test(provider): migrate more config-backed cases (#26984 )

test(provider): migrate provider tests to effect runner
test(project): stabilize VCS branch update test (#26979 )
2026-05-11 22:21:50 -04:00 · 2026-05-11 22:13:26 -04:00 · 2026-05-11 22:12:07 -04:00 · 2026-05-11 21:42:17 -04:00 · 2026-05-11 21:42:04 -04:00 · 2026-05-11 21:41:56 -04:00
422 changed files with 22104 additions and 13067 deletions
@@ -23,7 +23,7 @@ runs:
        fi

    - name: Setup Bun
-      uses: oven-sh/setup-bun@v2
+      uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
      with:
        bun-version-file: ${{ !steps.bun-url.outputs.url && 'package.json' || '' }}
        bun-download-url: ${{ steps.bun-url.outputs.url }}
@@ -34,7 +34,7 @@ runs:
      run: echo "dir=$(bun pm cache)" >> "$GITHUB_OUTPUT"

    - name: Cache Bun dependencies
-      uses: actions/cache@v4
+      uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
      with:
        path: ${{ steps.cache.outputs.dir }}
        key: ${{ runner.os }}-bun-${{ hashFiles('**/bun.lock') }}
@@ -19,7 +19,7 @@ runs:
  steps:
    - name: Create app token
      id: apptoken
-      uses: actions/create-github-app-token@v2
+      uses: actions/create-github-app-token@fee1f7d63c2ff003460e3d139729b119787bc349 # v2.2.2
      with:
        app-id: ${{ inputs.opencode-app-id }}
        private-key: ${{ inputs.opencode-app-secret }}
@@ -13,7 +13,7 @@ jobs:
      pull-requests: write
    steps:
      - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          fetch-depth: 0

@@ -12,9 +12,9 @@ jobs:
      contents: read
      issues: write
    steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1

-      - uses: oven-sh/setup-bun@v2
+      - uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
        with:
          bun-version: latest

@@ -21,7 +21,7 @@ jobs:
    timeout-minutes: 15
    steps:
      - name: Close inactive PRs
-        uses: actions/github-script@v8
+        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
        with:
          github-token: ${{ secrets.GITHUB_TOKEN }}
          script: |
@@ -16,7 +16,7 @@ jobs:
    runs-on: ubuntu-latest
    steps:
      - name: Close non-compliant issues and PRs after 2 hours
-        uses: actions/github-script@v7
+        uses: actions/github-script@f28e40c7f34bde8b3046d885e986cb6290c5673b # v7.1.0
        with:
          script: |
            const { data: items } = await github.rest.issues.listForRepo({
@@ -21,18 +21,18 @@ jobs:
      REGISTRY: ghcr.io/${{ github.repository_owner }}
      TAG: "24.04"
    steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1

      - uses: ./.github/actions/setup-bun

      - name: Set up QEMU
-        uses: docker/setup-qemu-action@v3
+        uses: docker/setup-qemu-action@c7c53464625b32c7a7e944ae62b3e17d2b600130 # v3.7.0

      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
+        uses: docker/setup-buildx-action@8d2750c68a42422c14e847fe6c8ac0403b4cbd6f # v3.12.0

      - name: Login to GHCR
-        uses: docker/login-action@v3
+        uses: docker/login-action@c94ce9fb468520275223c153574b00df6fe4bcc9 # v3.7.0
        with:
          registry: ghcr.io
          username: ${{ github.repository_owner }}
@@ -13,11 +13,11 @@ jobs:
  deploy:
    runs-on: ubuntu-latest
    steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0

      - uses: ./.github/actions/setup-bun

-      - uses: actions/setup-node@v4
+      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
        with:
          node-version: "24"

@@ -16,7 +16,7 @@ jobs:
      contents: write
    steps:
      - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          persist-credentials: false
          fetch-depth: 0
@@ -18,7 +18,7 @@ jobs:
      pull-requests: write
    steps:
      - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          fetch-depth: 0 # Fetch full history to access commits

@@ -43,7 +43,7 @@ jobs:

      - name: Run opencode
        if: steps.commits.outputs.has_commits == 'true'
-        uses: sst/opencode/github@latest
+        uses: sst/opencode/github@2c14fc5586fe0b88e5c04732d2e846769cc35671 # latest
        env:
          OPENCODE_API_KEY: ${{ secrets.OPENCODE_API_KEY }}
        with:
@@ -13,7 +13,7 @@ jobs:
      issues: write
    steps:
      - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          fetch-depth: 1

@@ -125,7 +125,7 @@ jobs:
      issues: write
    steps:
      - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          fetch-depth: 1

@@ -13,7 +13,7 @@ jobs:
      pull-requests: write
    steps:
      - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1

      - name: Setup Bun
        uses: ./.github/actions/setup-bun
@@ -20,10 +20,10 @@ jobs:
    timeout-minutes: 15
    steps:
      - name: Checkout repository
-        uses: actions/checkout@v6
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2

      - name: Setup Nix
-        uses: nixbuild/nix-quick-install-action@v34
+        uses: nixbuild/nix-quick-install-action@2c9db80fb984ceb1bcaa77cdda3fdf8cfba92035 # v34

      - name: Evaluate flake outputs (all systems)
        run: |
@@ -41,10 +41,10 @@ jobs:

    steps:
      - name: Checkout repository
-        uses: actions/checkout@v6
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2

      - name: Setup Nix
-        uses: nixbuild/nix-quick-install-action@v34
+        uses: nixbuild/nix-quick-install-action@2c9db80fb984ceb1bcaa77cdda3fdf8cfba92035 # v34

      - name: Compute node_modules hash
        id: hash
@@ -72,7 +72,7 @@ jobs:
          echo "Computed hash for ${SYSTEM}: $HASH"

      - name: Upload hash
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
        with:
          name: hash-${{ matrix.system }}
          path: hash.txt
@@ -85,7 +85,7 @@ jobs:

    steps:
      - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          persist-credentials: false
          fetch-depth: 0
@@ -102,7 +102,7 @@ jobs:
          git pull --rebase --autostash origin "$GITHUB_REF_NAME"

      - name: Download hash artifacts
-        uses: actions/download-artifact@v4
+        uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
        with:
          path: hashes
          pattern: hash-*
@@ -9,6 +9,6 @@ jobs:
    runs-on: blacksmith-4vcpu-ubuntu-2404
    steps:
      - name: Send nicely-formatted embed to Discord
-        uses: SethCohen/github-releases-to-discord@v1
+        uses: SethCohen/github-releases-to-discord@24d166886aee4646d448c8a389ff9e1ebcab3682 # v1.20.0
        with:
          webhook_url: ${{ secrets.DISCORD_WEBHOOK }}
@@ -21,12 +21,12 @@ jobs:
      issues: read
    steps:
      - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1

      - uses: ./.github/actions/setup-bun

      - name: Run opencode
-        uses: anomalyco/opencode/github@latest
+        uses: anomalyco/opencode/github@2c14fc5586fe0b88e5c04732d2e846769cc35671 # latest
        env:
          OPENCODE_API_KEY: ${{ secrets.OPENCODE_API_KEY }}
          OPENCODE_PERMISSION: '{"bash": "deny"}'
@@ -12,7 +12,7 @@ jobs:
      pull-requests: write
    steps:
      - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          fetch-depth: 1

@@ -78,7 +78,7 @@ jobs:
      issues: write
    steps:
      - name: Add Contributor Label
-        uses: actions/github-script@v8
+        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
        with:
          script: |
            const isPR = !!context.payload.pull_request;
@@ -12,7 +12,7 @@ jobs:
      pull-requests: write
    steps:
      - name: Check PR standards
-        uses: actions/github-script@v7
+        uses: actions/github-script@f28e40c7f34bde8b3046d885e986cb6290c5673b # v7.1.0
        with:
          script: |
            const pr = context.payload.pull_request;
@@ -159,7 +159,7 @@ jobs:
      pull-requests: write
    steps:
      - name: Check PR template compliance
-        uses: actions/github-script@v7
+        uses: actions/github-script@f28e40c7f34bde8b3046d885e986cb6290c5673b # v7.1.0
        with:
          script: |
            const pr = context.payload.pull_request;
@@ -16,7 +16,7 @@ jobs:
  publish:
    runs-on: blacksmith-4vcpu-ubuntu-2404
    steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
        with:
          fetch-depth: 0

@@ -15,7 +15,7 @@ jobs:
  publish:
    runs-on: blacksmith-4vcpu-ubuntu-2404
    steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
        with:
          fetch-depth: 0

@@ -35,7 +35,7 @@ jobs:
    runs-on: blacksmith-4vcpu-ubuntu-2404
    if: github.repository == 'anomalyco/opencode'
    steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
        with:
          fetch-depth: 0

@@ -72,7 +72,7 @@ jobs:
    runs-on: blacksmith-4vcpu-ubuntu-2404
    if: github.repository == 'anomalyco/opencode'
    steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
        with:
          fetch-tags: true

@@ -95,14 +95,14 @@ jobs:
          GH_REPO: ${{ needs.version.outputs.repo }}
          GH_TOKEN: ${{ steps.committer.outputs.token }}

-      - uses: actions/upload-artifact@v4
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
        with:
          name: opencode-cli
          path: |
            packages/opencode/dist/opencode-darwin*
            packages/opencode/dist/opencode-linux*

-      - uses: actions/upload-artifact@v4
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
        with:
          name: opencode-cli-windows
          path: packages/opencode/dist/opencode-windows*
@@ -123,9 +123,9 @@ jobs:
      AZURE_TRUSTED_SIGNING_CERTIFICATE_PROFILE: ${{ secrets.AZURE_TRUSTED_SIGNING_CERTIFICATE_PROFILE }}
      AZURE_TRUSTED_SIGNING_ENDPOINT: ${{ secrets.AZURE_TRUSTED_SIGNING_ENDPOINT }}
    steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0

-      - uses: actions/download-artifact@v4
+      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
        with:
          name: opencode-cli-windows
          path: packages/opencode/dist
@@ -138,13 +138,13 @@ jobs:
          opencode-app-secret: ${{ secrets.OPENCODE_APP_SECRET }}

      - name: Azure login
-        uses: azure/login@v2
+        uses: azure/login@a457da9ea143d694b1b9c7c869ebb04ebe844ef5 # v2.3.0
        with:
          client-id: ${{ env.AZURE_CLIENT_ID }}
          tenant-id: ${{ env.AZURE_TENANT_ID }}
          subscription-id: ${{ env.AZURE_SUBSCRIPTION_ID }}

-      - uses: azure/artifact-signing-action@v1
+      - uses: azure/artifact-signing-action@b443cf8ea4124818d2ea9f043cba29fc3ec47b16 # v1.2.0
        with:
          endpoint: ${{ env.AZURE_TRUSTED_SIGNING_ENDPOINT }}
          signing-account-name: ${{ env.AZURE_TRUSTED_SIGNING_ACCOUNT_NAME }}
@@ -201,7 +201,7 @@ jobs:
            --clobber `
            --repo "${{ needs.version.outputs.repo }}"

-      - uses: actions/upload-artifact@v4
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
        with:
          name: opencode-cli-signed-windows
          path: |
@@ -249,9 +249,9 @@ jobs:
            platform_flag: --linux
    runs-on: ${{ matrix.settings.host }}
    steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0

-      - uses: apple-actions/import-codesign-certs@v2
+      - uses: apple-actions/import-codesign-certs@8f3fb608891dd2244cdab3d69cd68c0d37a7fe93 # v2.0.0
        if: runner.os == 'macOS'
        with:
          keychain: build
@@ -268,19 +268,19 @@ jobs:

      - name: Azure login
        if: runner.os == 'Windows'
-        uses: azure/login@v2
+        uses: azure/login@a457da9ea143d694b1b9c7c869ebb04ebe844ef5 # v2.3.0
        with:
          client-id: ${{ env.AZURE_CLIENT_ID }}
          tenant-id: ${{ env.AZURE_TENANT_ID }}
          subscription-id: ${{ env.AZURE_SUBSCRIPTION_ID }}

-      - uses: actions/setup-node@v4
+      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
        with:
          node-version: "24"

      - name: Cache apt packages
        if: contains(matrix.settings.host, 'ubuntu')
-        uses: actions/cache@v4
+        uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
        with:
          path: ~/apt-cache
          key: ${{ runner.os }}-${{ matrix.settings.target }}-apt-electron-${{ hashFiles('.github/workflows/publish.yml') }}
@@ -388,12 +388,12 @@ jobs:
            }
          }

-      - uses: actions/upload-artifact@v4
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
        with:
          name: opencode-desktop-${{ matrix.settings.target }}
          path: packages/desktop/dist/*

-      - uses: actions/upload-artifact@v4
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
        if: needs.version.outputs.release
        with:
          name: latest-yml-${{ matrix.settings.target }}
@@ -408,44 +408,44 @@ jobs:
    if: always() && !failure() && !cancelled()
    runs-on: blacksmith-4vcpu-ubuntu-2404
    steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0

      - uses: ./.github/actions/setup-bun

      - name: Login to GitHub Container Registry
-        uses: docker/login-action@v3
+        uses: docker/login-action@c94ce9fb468520275223c153574b00df6fe4bcc9 # v3.7.0
        with:
          registry: ghcr.io
          username: ${{ github.repository_owner }}
          password: ${{ secrets.GITHUB_TOKEN }}

      - name: Set up QEMU
-        uses: docker/setup-qemu-action@v3
+        uses: docker/setup-qemu-action@c7c53464625b32c7a7e944ae62b3e17d2b600130 # v3.7.0

      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
+        uses: docker/setup-buildx-action@8d2750c68a42422c14e847fe6c8ac0403b4cbd6f # v3.12.0

-      - uses: actions/setup-node@v4
+      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
        with:
          node-version: "24"
          registry-url: "https://registry.npmjs.org"

-      - uses: actions/download-artifact@v4
+      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
        with:
          name: opencode-cli
          path: packages/opencode/dist

-      - uses: actions/download-artifact@v4
+      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
        with:
          name: opencode-cli-windows
          path: packages/opencode/dist

-      - uses: actions/download-artifact@v4
+      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
        with:
          name: opencode-cli-signed-windows
          path: packages/opencode/dist

-      - uses: actions/download-artifact@v4
+      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
        if: needs.version.outputs.release
        with:
          pattern: latest-yml-*
@@ -459,7 +459,7 @@ jobs:
          opencode-app-secret: ${{ secrets.OPENCODE_APP_SECRET }}

      - name: Cache apt packages (AUR)
-        uses: actions/cache@v4
+        uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
        with:
          path: /var/cache/apt/archives
          key: ${{ runner.os }}-apt-aur-${{ hashFiles('.github/workflows/publish.yml') }}
@@ -16,7 +16,7 @@ jobs:
  release:
    runs-on: blacksmith-4vcpu-ubuntu-2404
    steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          fetch-depth: 0

@@ -25,7 +25,7 @@ jobs:
          fi

      - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          fetch-depth: 1

@@ -16,7 +16,7 @@ jobs:

    steps:
      - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1

      - name: Setup Bun
        uses: ./.github/actions/setup-bun
@@ -29,7 +29,7 @@ jobs:
    runs-on: blacksmith-4vcpu-ubuntu-2404
    steps:
      - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1

      - name: Setup Bun
        uses: ./.github/actions/setup-bun
@@ -10,7 +10,7 @@ jobs:
    name: Release Zed Extension
    runs-on: blacksmith-4vcpu-ubuntu-2404
    steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          fetch-depth: 0

@@ -37,12 +37,12 @@ jobs:
        shell: bash
    steps:
      - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          token: ${{ secrets.GITHUB_TOKEN }}

      - name: Setup Node
-        uses: actions/setup-node@v4
+        uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
        with:
          node-version: "24"

@@ -55,7 +55,7 @@ jobs:
          git config --global user.name "opencode"

      - name: Cache Turbo
-        uses: actions/cache@v4
+        uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
        with:
          path: node_modules/.cache/turbo
          key: turbo-${{ runner.os }}-${{ hashFiles('turbo.json', '**/package.json') }}-${{ github.sha }}
@@ -75,7 +75,7 @@ jobs:

      - name: Publish unit reports
        if: always()
-        uses: mikepenz/action-junit-report@v6
+        uses: mikepenz/action-junit-report@bccf2e31636835cf0874589931c4116687171386 # v6.4.0
        with:
          report_paths: packages/*/.artifacts/unit/junit.xml
          check_name: "unit results (${{ matrix.settings.name }})"
@@ -85,7 +85,7 @@ jobs:

      - name: Upload unit artifacts
        if: always()
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
        with:
          name: unit-${{ matrix.settings.name }}-${{ github.run_attempt }}
          include-hidden-files: true
@@ -111,12 +111,12 @@ jobs:
        shell: bash
    steps:
      - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          token: ${{ secrets.GITHUB_TOKEN }}

      - name: Setup Node
-        uses: actions/setup-node@v4
+        uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
        with:
          node-version: "24"

@@ -131,7 +131,7 @@ jobs:

      - name: Cache Playwright browsers
        id: playwright-cache
-        uses: actions/cache@v4
+        uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
        with:
          path: ${{ github.workspace }}/.playwright-browsers
          key: ${{ runner.os }}-${{ runner.arch }}-playwright-${{ steps.playwright-version.outputs.version }}-chromium
@@ -155,7 +155,7 @@ jobs:

      - name: Upload Playwright artifacts
        if: always()
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
        with:
          name: playwright-${{ matrix.settings.name }}-${{ github.run_attempt }}
          if-no-files-found: ignore
@@ -12,7 +12,7 @@ jobs:
      issues: write
    steps:
      - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          fetch-depth: 1

@@ -12,7 +12,7 @@ jobs:
    runs-on: blacksmith-4vcpu-ubuntu-2404
    steps:
      - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1

      - name: Setup Bun
        uses: ./.github/actions/setup-bun
@@ -9,6 +9,7 @@
 ### General Principles

 - Keep things in one function unless composable or reusable
+- Do not extract single-use helpers preemptively. Inline the logic at the call site unless the helper is reused, hides a genuinely complex boundary, or has a clear independent name that improves the caller.
 - Avoid `try`/`catch` where possible
 - Avoid using the `any` type
 - Use Bun APIs when possible, like `Bun.file()`
@@ -29,7 +29,7 @@
    },
    "packages/app": {
      "name": "@opencode-ai/app",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "dependencies": {
        "@kobalte/core": "catalog:",
        "@opencode-ai/core": "workspace:*",
@@ -85,7 +85,7 @@
    },
    "packages/console/app": {
      "name": "@opencode-ai/console-app",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "dependencies": {
        "@cloudflare/vite-plugin": "1.15.2",
        "@ibm/plex": "6.4.1",
@@ -120,7 +120,7 @@
    },
    "packages/console/core": {
      "name": "@opencode-ai/console-core",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "dependencies": {
        "@aws-sdk/client-sts": "3.782.0",
        "@jsx-email/render": "1.1.1",
@@ -147,17 +147,15 @@
    },
    "packages/console/function": {
      "name": "@opencode-ai/console-function",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "dependencies": {
        "@ai-sdk/anthropic": "3.0.64",
        "@ai-sdk/openai": "3.0.48",
        "@ai-sdk/openai-compatible": "2.0.37",
-        "@hono/zod-validator": "catalog:",
        "@openauthjs/openauth": "0.0.0-20250322224806",
        "@opencode-ai/console-core": "workspace:*",
        "@opencode-ai/console-resource": "workspace:*",
        "ai": "catalog:",
-        "hono": "catalog:",
        "zod": "catalog:",
      },
      "devDependencies": {
@@ -171,7 +169,7 @@
    },
    "packages/console/mail": {
      "name": "@opencode-ai/console-mail",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "dependencies": {
        "@jsx-email/all": "2.2.3",
        "@jsx-email/cli": "1.4.3",
@@ -195,7 +193,7 @@
    },
    "packages/core": {
      "name": "@opencode-ai/core",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "bin": {
        "opencode": "./bin/opencode",
      },
@@ -229,7 +227,7 @@
    },
    "packages/desktop": {
      "name": "@opencode-ai/desktop",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "dependencies": {
        "drizzle-orm": "catalog:",
        "effect": "catalog:",
@@ -283,7 +281,7 @@
    },
    "packages/enterprise": {
      "name": "@opencode-ai/enterprise",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "dependencies": {
        "@opencode-ai/core": "workspace:*",
        "@opencode-ai/ui": "workspace:*",
@@ -313,7 +311,7 @@
    },
    "packages/function": {
      "name": "@opencode-ai/function",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "dependencies": {
        "@octokit/auth-app": "8.0.1",
        "@octokit/rest": "catalog:",
@@ -329,7 +327,7 @@
    },
    "packages/http-recorder": {
      "name": "@opencode-ai/http-recorder",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "dependencies": {
        "@effect/platform-node": "catalog:",
        "effect": "catalog:",
@@ -342,7 +340,7 @@
    },
    "packages/llm": {
      "name": "@opencode-ai/llm",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "dependencies": {
        "@smithy/eventstream-codec": "4.2.14",
        "@smithy/util-utf8": "4.2.2",
@@ -360,7 +358,7 @@
    },
    "packages/opencode": {
      "name": "opencode",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "bin": {
        "opencode": "./bin/opencode",
      },
@@ -412,6 +410,7 @@
        "@opentui/solid": "catalog:",
        "@parcel/watcher": "2.5.1",
        "@pierre/diffs": "catalog:",
+        "@silvia-odwyer/photon-node": "0.3.4",
        "@solid-primitives/event-bus": "1.1.2",
        "@solid-primitives/scheduled": "1.5.2",
        "@standard-schema/spec": "1.0.0",
@@ -495,7 +494,7 @@
    },
    "packages/plugin": {
      "name": "@opencode-ai/plugin",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "dependencies": {
        "@opencode-ai/sdk": "workspace:*",
        "effect": "catalog:",
@@ -533,7 +532,7 @@
    },
    "packages/sdk/js": {
      "name": "@opencode-ai/sdk",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "dependencies": {
        "cross-spawn": "catalog:",
      },
@@ -548,7 +547,7 @@
    },
    "packages/slack": {
      "name": "@opencode-ai/slack",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "dependencies": {
        "@opencode-ai/sdk": "workspace:*",
        "@slack/bolt": "^3.17.1",
@@ -583,7 +582,7 @@
    },
    "packages/ui": {
      "name": "@opencode-ai/ui",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "dependencies": {
        "@kobalte/core": "catalog:",
        "@opencode-ai/core": "workspace:*",
@@ -632,7 +631,7 @@
    },
    "packages/web": {
      "name": "@opencode-ai/web",
-      "version": "1.14.44",
+      "version": "1.14.48",
      "dependencies": {
        "@astrojs/cloudflare": "12.6.3",
        "@astrojs/markdown-remark": "6.3.1",
@@ -677,6 +676,7 @@
    "solid-js@1.9.10": "patches/solid-js@1.9.10.patch",
    "@standard-community/standard-openapi@0.2.9": "patches/@standard-community%2Fstandard-openapi@0.2.9.patch",
    "@npmcli/agent@4.0.0": "patches/@npmcli%2Fagent@4.0.0.patch",
+    "@silvia-odwyer/photon-node@0.3.4": "patches/@silvia-odwyer%2Fphoton-node@0.3.4.patch",
  },
  "overrides": {
    "@types/bun": "catalog:",
@@ -684,8 +684,8 @@
  },
  "catalog": {
    "@cloudflare/workers-types": "4.20251008.0",
-    "@effect/opentelemetry": "4.0.0-beta.57",
-    "@effect/platform-node": "4.0.0-beta.57",
+    "@effect/opentelemetry": "4.0.0-beta.65",
+    "@effect/platform-node": "4.0.0-beta.65",
    "@hono/zod-validator": "0.4.2",
    "@kobalte/core": "0.13.11",
    "@lydell/node-pty": "1.2.0-beta.10",
@@ -718,7 +718,7 @@
    "dompurify": "3.3.1",
    "drizzle-kit": "1.0.0-beta.19-d95b7a4",
    "drizzle-orm": "1.0.0-beta.19-d95b7a4",
-    "effect": "4.0.0-beta.59",
+    "effect": "4.0.0-beta.65",
    "fuzzysort": "3.1.0",
    "hono": "4.10.7",
    "hono-openapi": "1.1.2",
@@ -1079,11 +1079,11 @@

    "@drizzle-team/brocli": ["@drizzle-team/brocli@0.11.0", "", {}, "sha512-hD3pekGiPg0WPCCGAZmusBBJsDqGUR66Y452YgQsZOnkdQ7ViEPKuyP4huUGEZQefp8g34RRodXYmJ2TbCH+tg=="],

-    "@effect/opentelemetry": ["@effect/opentelemetry@4.0.0-beta.57", "", { "peerDependencies": { "@opentelemetry/api": "^1.9", "@opentelemetry/resources": "^2.0.0", "@opentelemetry/sdk-logs": ">=0.203.0 <0.300.0", "@opentelemetry/sdk-metrics": "^2.0.0", "@opentelemetry/sdk-trace-base": "^2.0.0", "@opentelemetry/sdk-trace-node": "^2.0.0", "@opentelemetry/sdk-trace-web": "^2.0.0", "@opentelemetry/semantic-conventions": "^1.33.0", "effect": "^4.0.0-beta.57" }, "optionalPeers": ["@opentelemetry/api", "@opentelemetry/resources", "@opentelemetry/sdk-logs", "@opentelemetry/sdk-metrics", "@opentelemetry/sdk-trace-base", "@opentelemetry/sdk-trace-node", "@opentelemetry/sdk-trace-web"] }, "sha512-gdjZPEP0QQg4qmI1vd+443kheeQZKytrjJIzCJncy6ZEpyk/SfrqeStLqLXdTRcms3IB0ls0vOV7KNq7YmBRVA=="],
+    "@effect/opentelemetry": ["@effect/opentelemetry@4.0.0-beta.65", "", { "peerDependencies": { "@opentelemetry/api": "^1.9", "@opentelemetry/api-logs": ">=0.203.0 <0.300.0", "@opentelemetry/resources": "^2.0.0", "@opentelemetry/sdk-logs": ">=0.203.0 <0.300.0", "@opentelemetry/sdk-metrics": "^2.0.0", "@opentelemetry/sdk-trace-base": "^2.0.0", "@opentelemetry/sdk-trace-node": "^2.0.0", "@opentelemetry/sdk-trace-web": "^2.0.0", "@opentelemetry/semantic-conventions": "^1.33.0", "effect": "^4.0.0-beta.65" }, "optionalPeers": ["@opentelemetry/api", "@opentelemetry/api-logs", "@opentelemetry/resources", "@opentelemetry/sdk-logs", "@opentelemetry/sdk-metrics", "@opentelemetry/sdk-trace-base", "@opentelemetry/sdk-trace-node", "@opentelemetry/sdk-trace-web"] }, "sha512-0CD2fSsXrDM7FP2WFkbGJO1DwMqWR3UKHh6oBDXPHAPA+RsJSKoh3pLQsbQfldLuKnhOy87Bv0v9r9IdrIHCQw=="],

-    "@effect/platform-node": ["@effect/platform-node@4.0.0-beta.57", "", { "dependencies": { "@effect/platform-node-shared": "^4.0.0-beta.57", "mime": "^4.1.0", "undici": "^8.0.2" }, "peerDependencies": { "effect": "^4.0.0-beta.57", "ioredis": "^5.7.0" } }, "sha512-la0xxPSAYOsY0d+uVxEBxok3jYB31iPQmIaZZRUj2SNWqcGGHJc6KorKtI8guqSLuv9FGZ255kBWXRbG6hMeeg=="],
+    "@effect/platform-node": ["@effect/platform-node@4.0.0-beta.65", "", { "dependencies": { "@effect/platform-node-shared": "^4.0.0-beta.65", "mime": "^4.1.0", "undici": "^8.0.2" }, "peerDependencies": { "effect": "^4.0.0-beta.65", "ioredis": "^5.7.0" } }, "sha512-QQy3KRcMwP0TngQdfQGl2u1zp03B7k7DuF5SNS8aZhD0dDBpKZpCwFad1ODY5qdY3ycPgMwBwKRRK7y/aw0C9w=="],

-    "@effect/platform-node-shared": ["@effect/platform-node-shared@4.0.0-beta.57", "", { "dependencies": { "@types/ws": "^8.18.1", "ws": "^8.20.0" }, "peerDependencies": { "effect": "^4.0.0-beta.57" } }, "sha512-C976X6f+qHUtLSqcqImuCrjhAHnJV17NC2RvvybsAuDfkyIWU4MyiO2XwgiBeijeNupyr1M/KPKnyjtkNxV9Hw=="],
+    "@effect/platform-node-shared": ["@effect/platform-node-shared@4.0.0-beta.65", "", { "dependencies": { "@types/ws": "^8.18.1", "ws": "^8.20.0" }, "peerDependencies": { "effect": "^4.0.0-beta.65" } }, "sha512-3rY8F3WLEax6Hj08GI/OvDIH+KqjfxH7RM2bAMfgR75NgRmwDtny1P49PtPkoRjH5dcdtThThtsvE4X9OTZkpQ=="],

    "@electron/asar": ["@electron/asar@3.4.1", "", { "dependencies": { "commander": "^5.0.0", "glob": "^7.1.6", "minimatch": "^3.0.4" }, "bin": { "asar": "bin/asar.js" } }, "sha512-i4/rNPRS84t0vSRa2HorerGRXWyF4vThfHesw0dmcWHp+cspK743UanA0suA5Q5y8kzY2y6YKrvbIUn69BCAiA=="],

@@ -1233,8 +1233,6 @@

    "@hono/standard-validator": ["@hono/standard-validator@0.1.5", "", { "peerDependencies": { "@standard-schema/spec": "1.0.0", "hono": ">=3.9.0" } }, "sha512-EIyZPPwkyLn6XKwFj5NBEWHXhXbgmnVh2ceIFo5GO7gKI9WmzTjPDKnppQB0KrqKeAkq3kpoW4SIbu5X1dgx3w=="],

-    "@hono/zod-validator": ["@hono/zod-validator@0.4.2", "", { "peerDependencies": { "hono": ">=3.9.0", "zod": "^3.19.1" } }, "sha512-1rrlBg+EpDPhzOV4hT9pxr5+xDVmKuz6YJl+la7VCwK6ass5ldyKm5fD+umJdV2zhHD6jROoCCv8NbTwyfhT0g=="],
-
    "@ibm/plex": ["@ibm/plex@6.4.1", "", { "dependencies": { "@ibm/telemetry-js": "^1.5.1" } }, "sha512-fnsipQywHt3zWvsnlyYKMikcVI7E2fEwpiPnIHFqlbByXVfQfANAAeJk1IV4mNnxhppUIDlhU0TzwYwL++Rn2g=="],

    "@ibm/telemetry-js": ["@ibm/telemetry-js@1.11.0", "", { "bin": { "ibmtelemetry": "dist/collect.js" } }, "sha512-RO/9j+URJnSfseWg9ZkEX9p+a3Ousd33DBU7rOafoZB08RqdzxFVYJ2/iM50dkBuD0o7WX7GYt1sLbNgCoE+pA=="],
@@ -2035,6 +2033,8 @@

    "@sigstore/verify": ["@sigstore/verify@3.1.0", "", { "dependencies": { "@sigstore/bundle": "^4.0.0", "@sigstore/core": "^3.1.0", "@sigstore/protobuf-specs": "^0.5.0" } }, "sha512-mNe0Iigql08YupSOGv197YdHpPPr+EzDZmfCgMc7RPNaZTw5aLN01nBl6CHJOh3BGtnMIj83EeN4butBchc8Ag=="],

+    "@silvia-odwyer/photon-node": ["@silvia-odwyer/photon-node@0.3.4", "", {}, "sha512-bnly4BKB3KDTFxrUIcgCLbaeVVS8lrAkri1pEzskpmxu9MdfGQTy8b8EgcD83ywD3RPMsIulY8xJH5Awa+t9fA=="],
+
    "@sindresorhus/is": ["@sindresorhus/is@4.6.0", "", {}, "sha512-t09vSN3MdfsyCHoFcTRCH/iUtG7OJ0CsjzB8cjAmKc/va/kIgeDI/TxsigdncE/4be734m0cvIYwNaV4i2XqAw=="],

    "@slack/bolt": ["@slack/bolt@3.22.0", "", { "dependencies": { "@slack/logger": "^4.0.0", "@slack/oauth": "^2.6.3", "@slack/socket-mode": "^1.3.6", "@slack/types": "^2.13.0", "@slack/web-api": "^6.13.0", "@types/express": "^4.16.1", "@types/promise.allsettled": "^1.0.3", "@types/tsscmp": "^1.0.0", "axios": "^1.7.4", "express": "^4.21.0", "path-to-regexp": "^8.1.0", "promise.allsettled": "^1.0.2", "raw-body": "^2.3.3", "tsscmp": "^1.0.6" } }, "sha512-iKDqGPEJDnrVwxSVlFW6OKTkijd7s4qLBeSufoBsTM0reTyfdp/5izIQVkxNfzjHi3o6qjdYbRXkYad5HBsBog=="],
@@ -3037,7 +3037,7 @@

    "ee-first": ["ee-first@1.1.1", "", {}, "sha512-WMwm9LhRUo+WUaRN+vRuETqG89IgZphVSNkdFgeb6sS/E4OrDIN7t48CAewSHXc6C8lefD8KKfr5vY61brQlow=="],

-    "effect": ["effect@4.0.0-beta.59", "", { "dependencies": { "@standard-schema/spec": "^1.1.0", "fast-check": "^4.6.0", "find-my-way-ts": "^0.1.6", "ini": "^6.0.0", "kubernetes-types": "^1.30.0", "msgpackr": "^1.11.9", "multipasta": "^0.2.7", "toml": "^4.1.1", "uuid": "^13.0.0", "yaml": "^2.8.3" } }, "sha512-xyUDLeHSe8d6lWGOvR6Fgn2HL6gYeTZ/S4Jzk9uc4ZUxMPPsNZlNXrvk0C7/utQFzeX7uAWcVnG2BjbA0SRoAA=="],
+    "effect": ["effect@4.0.0-beta.65", "", { "dependencies": { "@standard-schema/spec": "^1.1.0", "fast-check": "^4.6.0", "find-my-way-ts": "^0.1.6", "ini": "^6.0.0", "kubernetes-types": "^1.30.0", "msgpackr": "^1.11.9", "multipasta": "^0.2.7", "toml": "^4.1.1", "uuid": "^13.0.0", "yaml": "^2.8.3" } }, "sha512-QYKvQPAj3CmtsvWkHQww15wX4KG2gNsszDWEcOO5sZCMknp66u6Si/Opmt3wwWCwsyvRmDAdIg+JIz5qzbbFIw=="],

    "ejs": ["ejs@3.1.10", "", { "dependencies": { "jake": "^10.8.5" }, "bin": { "ejs": "bin/cli.js" } }, "sha512-UeJmFfOrAQS8OJWPZ4qtgHyWExa088/MtK5UEyoJGFH67cDEXkZSviOiKRCZ4Xij0zxI3JECgYs3oKx+AizQBA=="],

@@ -5465,8 +5465,6 @@

    "@hey-api/openapi-ts/semver": ["semver@7.7.3", "", { "bin": { "semver": "bin/semver.js" } }, "sha512-SdsKMrI9TdgjdweUSR9MweHA4EJ8YxHn8DFaDisvhVlUOe4BF1tLD7GAj0lIqWVl+dPb/rExr0Btby5loQm20Q=="],

-    "@hono/zod-validator/zod": ["zod@3.25.76", "", {}, "sha512-gzUt/qt81nXsFGKIFcC3YnfEAx5NkunCfnDlvuBSSFS02bcXu4Lmea0AFIUwbLWxWPx3d9p8S5QoaujKcNQxcQ=="],
-
    "@jimp/core/mime": ["mime@3.0.0", "", { "bin": { "mime": "cli.js" } }, "sha512-jSCU7/VB1loIWBZe14aEYHU/+1UMEHoaO7qxCOVJOw9GgH72VAWppxNcjU+x9a2k3GSIBXNKxXQFqRvvZ7vr3A=="],

    "@jimp/plugin-blit/zod": ["zod@3.25.76", "", {}, "sha512-gzUt/qt81nXsFGKIFcC3YnfEAx5NkunCfnDlvuBSSFS02bcXu4Lmea0AFIUwbLWxWPx3d9p8S5QoaujKcNQxcQ=="],
@@ -1,6 +1,8 @@
 import { SECRET } from "./secret"
 import { domain } from "./stage"

+const description = "Managed by SST (Don't edit in Honeycomb UI)"
+
 const webhookRecipient = new honeycomb.WebhookRecipient("DiscordAlerts", {
  name: $app.stage === "production" ? "Discord Alerts" : `Discord Alerts (${$app.stage})`,
  url: `https://${domain}/honeycomb/webhook`,
@@ -25,6 +27,16 @@ const webhookRecipient = new honeycomb.WebhookRecipient("DiscordAlerts", {
  ],
 })

+// Honeycomb can keep stale query-local calculated fields when the name is unchanged,
+// so tie the field name to the expression while avoiding deploy-to-deploy churn.
+// https://github.com/honeycombio/terraform-provider-honeycombio/issues/852
+const calculatedField = (field: { name: string; expression: string }) => ({
+  ...field,
+  name: `${field.name}_${(
+    Array.from(field.expression).reduce((result, char) => Math.imul(31, result) + char.charCodeAt(0), 0) >>> 0
+  ).toString(36)}`,
+})
+
 const modelHttpErrorsQuery = (product: "go" | "zen") => {
  const filters = [
    { column: "model", op: "exists" },
@@ -32,21 +44,26 @@ const modelHttpErrorsQuery = (product: "go" | "zen") => {
    { column: "user_agent", op: "contains", value: "opencode" },
    { column: "isGoTier", op: "=", value: product === "go" ? "true" : "false" },
  ]
+  const failedHttpStatus = calculatedField({
+    name: "is_failed_http_status",
+    expression:
+      product === "go"
+        ? `IF(AND(GTE($status, "400"), NOT(EQUALS($status, "401")), NOT(EQUALS($status, "429"))), 1, 0)`
+        : `IF(AND(EQUALS($status, "429"), $isFreeTier), 0, AND(GTE($status, "400"), NOT(EQUALS($status, "401"))), 1, 0)`,
+  })

  return honeycomb.getQuerySpecificationOutput({
    breakdowns: ["model"],
-    calculatedFields: [
-      {
-        name: "is_failed_http_status",
-        expression:
-          product === "go"
-            ? `IF(AND(GTE($status, "400"), NOT(EQUALS($status, "401")), NOT(EQUALS($status, "429"))), 1, 0)`
-            : `IF(AND(GTE($status, "400"), NOT(EQUALS($status, "401"))), 1, 0)`,
-      },
-    ],
+    calculatedFields: [failedHttpStatus],
    calculations: [
      { op: "COUNT", name: "TOTAL", filterCombination: "AND", filters },
-      { op: "SUM", name: "FAILED", column: "is_failed_http_status", filterCombination: "AND", filters },
+      {
+        op: "SUM",
+        name: "FAILED",
+        column: failedHttpStatus.name,
+        filterCombination: "AND",
+        filters,
+      },
    ],
    formulas: [{ name: "ERROR", expression: "IF(GTE($TOTAL, 100), DIV($FAILED, $TOTAL), 0)" }],
    timeRange: 900,
@@ -59,31 +76,30 @@ const providerHttpErrorsQuery = (product: "go" | "zen") => {
    { column: "user_agent", op: "contains", value: "opencode" },
    { column: "isGoTier", op: "=", value: product === "go" ? "true" : "false" },
  ]
+  const successHttpStatus = calculatedField({
+    name: "is_success_http_status",
+    expression: `IF(AND(GTE($status, "200"), LT($status, "400")), 1, 0)`,
+  })
+  const failedProviderHttpStatus = calculatedField({
+    name: "is_failed_provider_http_status",
+    expression: `IF(GT($llm.error.code, "400"), 1, 0)`,
+  })

  return honeycomb.getQuerySpecificationOutput({
    breakdowns: ["provider"],
-    calculatedFields: [
-      {
-        name: "is_success_http_status",
-        expression: `IF(AND(GTE($status, "200"), LT($status, "400")), 1, 0)`,
-      },
-      {
-        name: "is_failed_provider_http_status",
-        expression: `IF(GT($llm.error.code, "400"), 1, 0)`,
-      },
-    ],
+    calculatedFields: [successHttpStatus, failedProviderHttpStatus],
    calculations: [
      {
        op: "SUM",
        name: "SUCCESS",
-        column: "is_success_http_status",
+        column: successHttpStatus.name,
        filterCombination: "AND",
        filters: [...filters, { column: "event_type", op: "=", value: "completions" }],
      },
      {
        op: "SUM",
        name: "FAILED",
-        column: "is_failed_provider_http_status",
+        column: failedProviderHttpStatus.name,
        filterCombination: "AND",
        filters: [...filters, { column: "event_type", op: "=", value: "llm.error" }],
      },
@@ -95,8 +111,6 @@ const providerHttpErrorsQuery = (product: "go" | "zen") => {
  }).json
 }

-const description = "Managed by SST (Don't edit in Honeycomb UI)"
-
 new honeycomb.Trigger("IncreasedModelHttpErrorsGo", {
  name: "Increased Model HTTP Errors [Go]",
  description,
@@ -1,8 +1,8 @@
 {
  "nodeModules": {
-    "x86_64-linux": "sha256-LTo0ohJN5hBOubqFLVL45unVEIwBDkACNVv64k2nkq4=",
-    "aarch64-linux": "sha256-oYKY2UJRWG2fhufW4aGujX/Poou93023ZF2Fu7oyYOw=",
-    "aarch64-darwin": "sha256-618c9vqKN5I+no1nzylctAiWvqw7Bsa+bzSTNwXmSQA=",
-    "x86_64-darwin": "sha256-1ro3/gH0FC0TWXwWT+k675xR396GE98HpnBEeuD4t6k="
+    "x86_64-linux": "sha256-Q9r1S15YL9LQK7DRhuOpw3Fxi24BPovEM995GZJayKw=",
+    "aarch64-linux": "sha256-C0rRTLnxxuuEkCBc3JZbkR66TUVwpcPFif3BU9GRAuA=",
+    "aarch64-darwin": "sha256-1HvalOO/pOkRlYH8CZ93psapt90C+pYzui1JCadBE1Q=",
+    "x86_64-darwin": "sha256-RrndyLWfhWm4mZ88XytFF2NI+ly8la550Z5LBN/g5u4="
  }
 }
@@ -28,8 +28,8 @@
      "packages/slack"
    ],
    "catalog": {
-      "@effect/opentelemetry": "4.0.0-beta.57",
-      "@effect/platform-node": "4.0.0-beta.57",
+      "@effect/opentelemetry": "4.0.0-beta.65",
+      "@effect/platform-node": "4.0.0-beta.65",
      "@npmcli/arborist": "9.4.0",
      "@types/bun": "1.3.12",
      "@types/cross-spawn": "6.0.6",
@@ -55,7 +55,7 @@
      "dompurify": "3.3.1",
      "drizzle-kit": "1.0.0-beta.19-d95b7a4",
      "drizzle-orm": "1.0.0-beta.19-d95b7a4",
-      "effect": "4.0.0-beta.59",
+      "effect": "4.0.0-beta.65",
      "ai": "6.0.168",
      "cross-spawn": "7.0.6",
      "hono": "4.10.7",
@@ -133,6 +133,7 @@
  },
  "patchedDependencies": {
    "@npmcli/agent@4.0.0": "patches/@npmcli%2Fagent@4.0.0.patch",
+    "@silvia-odwyer/photon-node@0.3.4": "patches/@silvia-odwyer%2Fphoton-node@0.3.4.patch",
    "@standard-community/standard-openapi@0.2.9": "patches/@standard-community%2Fstandard-openapi@0.2.9.patch",
    "solid-js@1.9.10": "patches/solid-js@1.9.10.patch"
  }
@@ -1,6 +1,6 @@
 {
  "name": "@opencode-ai/app",
-  "version": "1.14.44",
+  "version": "1.14.48",
  "description": "",
  "type": "module",
  "exports": {
@@ -231,6 +231,7 @@ export function createChildStoreManager(input: {
            limit: 5,
            message: {},
            part: {},
+            part_text_accum_delta: {},
          })
          children[key] = child
          disposers.set(key, dispose)
@@ -81,6 +81,7 @@ const baseState = (input: Partial<State> = {}) =>
    limit: 10,
    message: {},
    part: {},
+    part_text_accum_delta: {},
    ...input,
  }) as State

@@ -211,6 +211,12 @@ export function applyDirectoryEvent(input: {
            const result = Binary.search(messages, props.messageID, (m) => m.id)
            if (result.found) messages.splice(result.index, 1)
          }
+          const parts = draft.part[props.messageID]
+          if (parts) {
+            for (const part of parts) {
+              delete draft.part_text_accum_delta[part.id]
+            }
+          }
          delete draft.part[props.messageID]
        }),
      )
@@ -219,6 +225,11 @@ export function applyDirectoryEvent(input: {
    case "message.part.updated": {
      const part = (event.properties as { part: Part }).part
      if (SKIP_PARTS.has(part.type)) break
+      input.setStore(
+        produce((draft) => {
+          delete draft.part_text_accum_delta[part.id]
+        }),
+      )
      const parts = input.store.part[part.messageID]
      if (!parts) {
        input.setStore("part", part.messageID, [part])
@@ -240,6 +251,11 @@ export function applyDirectoryEvent(input: {
    }
    case "message.part.removed": {
      const props = event.properties as { messageID: string; partID: string }
+      input.setStore(
+        produce((draft) => {
+          delete draft.part_text_accum_delta[props.partID]
+        }),
+      )
      const parts = input.store.part[props.messageID]
      if (!parts) break
      const result = Binary.search(parts, props.partID, (p) => p.id)
@@ -263,6 +279,7 @@ export function applyDirectoryEvent(input: {
      if (!parts) break
      const result = Binary.search(parts, props.partID, (p) => p.id)
      if (!result.found) break
+      input.setStore("part_text_accum_delta", props.partID, (existing) => (existing ?? "") + props.delta)
      input.setStore(
        "part",
        props.messageID,
@@ -39,6 +39,7 @@ describe("app session cache", () => {
      part: Record<string, Part[] | undefined>
      permission: Record<string, PermissionRequest[] | undefined>
      question: Record<string, QuestionRequest[] | undefined>
+      part_text_accum_delta: Record<string, string | undefined>
    } = {
      session_status: { ses_1: { type: "busy" } as SessionStatus },
      session_diff: { ses_1: [] },
@@ -47,12 +48,14 @@ describe("app session cache", () => {
      part: { msg_1: [part("prt_1", "ses_1", "msg_1")] },
      permission: { ses_1: [] as PermissionRequest[] },
      question: { ses_1: [] as QuestionRequest[] },
+      part_text_accum_delta: { prt_1: "streamed text" },
    }

    dropSessionCaches(store, ["ses_1"])

    expect(store.message.ses_1).toBeUndefined()
    expect(store.part.msg_1).toBeUndefined()
+    expect(store.part_text_accum_delta.prt_1).toBeUndefined()
    expect(store.todo.ses_1).toBeUndefined()
    expect(store.session_diff.ses_1).toBeUndefined()
    expect(store.session_status.ses_1).toBeUndefined()
@@ -70,6 +73,7 @@ describe("app session cache", () => {
      part: Record<string, Part[] | undefined>
      permission: Record<string, PermissionRequest[] | undefined>
      question: Record<string, QuestionRequest[] | undefined>
+      part_text_accum_delta: Record<string, string | undefined>
    } = {
      session_status: {},
      session_diff: {},
@@ -78,6 +82,7 @@ describe("app session cache", () => {
      part: { [m.id]: [part("prt_1", "ses_1", m.id)] },
      permission: {},
      question: {},
+      part_text_accum_delta: {},
    }

    dropSessionCaches(store, ["ses_1"])
@@ -18,6 +18,7 @@ type SessionCache = {
  part: Record<string, Part[] | undefined>
  permission: Record<string, PermissionRequest[] | undefined>
  question: Record<string, QuestionRequest[] | undefined>
+  part_text_accum_delta: Record<string, string | undefined>
 }

 export function dropSessionCaches(store: SessionCache, sessionIDs: Iterable<string>) {
@@ -27,6 +28,9 @@ export function dropSessionCaches(store: SessionCache, sessionIDs: Iterable<stri
  for (const key of Object.keys(store.part)) {
    const parts = store.part[key]
    if (!parts?.some((part) => stale.has(part?.sessionID ?? ""))) continue
+    for (const part of parts) {
+      delete store.part_text_accum_delta[part.id]
+    }
    delete store.part[key]
  }

@@ -72,6 +72,9 @@ export type State = {
  part: {
    [messageID: string]: Part[]
  }
+  part_text_accum_delta: {
+    [partID: string]: string
+  }
 }

 export type VcsCache = {
@@ -43,6 +43,7 @@ type SessionView = {
  reviewOpen?: string[]
  pendingMessage?: string
  pendingMessageAt?: number
+  todoCollapsed?: boolean
 }

 type TabHandoff = {
@@ -759,6 +760,18 @@ export const { use: useLayout, provider: LayoutProvider } = createSimpleContext(
          setScroll(tab: string, pos: SessionScroll) {
            scroll.setScroll(key(), tab, pos)
          },
+          todoCollapsed: {
+            get: () => s().todoCollapsed ?? false,
+            set(collapsed: boolean) {
+              const session = key()
+              const current = store.sessionView[session]
+              if (!current) {
+                setStore("sessionView", session, { scroll: {}, todoCollapsed: collapsed })
+              } else {
+                setStore("sessionView", session, "todoCollapsed", collapsed)
+              }
+            },
+          },
          terminal: {
            opened: terminalOpened,
            open() {
@@ -1934,7 +1934,7 @@ export default function Layout(props: ParentProps) {

    if (!created?.directory) return

-    setWorkspaceName(created.directory, created.branch, project.id, created.branch)
+    setWorkspaceName(created.directory, created.branch ?? getFilename(created.directory), project.id, created.branch)

    const local = project.worktree
    const key = pathKey(created.directory)
@@ -2,6 +2,7 @@ import { Show, createEffect, createMemo, onCleanup } from "solid-js"
 import { createStore } from "solid-js/store"
 import { useNavigate } from "@solidjs/router"
 import { useSpring } from "@opencode-ai/ui/motion-spring"
+import { useLayout } from "@/context/layout"
 import { PromptInput } from "@/components/prompt-input"
 import { useLanguage } from "@/context/language"
 import { usePrompt } from "@/context/prompt"
@@ -46,10 +47,12 @@ export function SessionComposerRegion(props: {
  setPromptDockRef: (el: HTMLDivElement) => void
 }) {
  const navigate = useNavigate()
+  const layout = useLayout()
  const prompt = usePrompt()
  const language = useLanguage()
  const route = useSessionKey()
  const sync = useSync()
+  const view = layout.view(route.sessionKey)

  const handoffPrompt = createMemo(() => getSessionHandoff(route.sessionKey())?.prompt)
  const info = createMemo(() => (route.params.id ? sync.session.get(route.params.id) : undefined))
@@ -207,6 +210,8 @@ export function SessionComposerRegion(props: {
                  <SessionTodoDock
                    sessionID={route.params.id}
                    todos={props.state.todos()}
+                    collapsed={view.todoCollapsed.get()}
+                    onToggle={() => view.todoCollapsed.set(!view.todoCollapsed.get())}
                    collapseLabel={language.t("session.todo.collapse")}
                    expandLabel={language.t("session.todo.expand")}
                    dockProgress={value()}
@@ -42,18 +42,17 @@ function dot(status: Todo["status"]) {
 export function SessionTodoDock(props: {
  sessionID?: string
  todos: Todo[]
+  collapsed: boolean
+  onToggle: () => void
  collapseLabel: string
  expandLabel: string
  dockProgress: number
 }) {
  const language = useLanguage()
  const [store, setStore] = createStore({
-    collapsed: false,
    height: 320,
  })

-  const toggle = () => setStore("collapsed", (value) => !value)
-
  const total = createMemo(() => props.todos.length)
  const done = createMemo(() => props.todos.filter((todo) => todo.status === "completed").length)
  const label = createMemo(() => language.t("session.todo.progress", { done: done(), total: total() }))
@@ -72,7 +71,7 @@ export function SessionTodoDock(props: {
  )

  const preview = createMemo(() => active()?.content ?? "")
-  const collapse = useSpring(() => (store.collapsed ? 1 : 0), { visualDuration: 0.3, bounce: 0 })
+  const collapse = useSpring(() => (props.collapsed ? 1 : 0), { visualDuration: 0.3, bounce: 0 })
  const dock = createMemo(() => Math.max(0, Math.min(1, props.dockProgress)))
  const shut = createMemo(() => 1 - dock())
  const value = createMemo(() => Math.max(0, Math.min(1, collapse())))
@@ -107,11 +106,11 @@ export function SessionTodoDock(props: {
          class="pl-3 pr-2 py-2 flex items-center gap-2 overflow-visible"
          role="button"
          tabIndex={0}
-          onClick={toggle}
+          onClick={props.onToggle}
          onKeyDown={(event) => {
            if (event.key !== "Enter" && event.key !== " ") return
            event.preventDefault()
-            toggle()
+            props.onToggle()
          }}
        >
          <span
@@ -148,7 +147,7 @@ export function SessionTodoDock(props: {
          >
            <TextReveal
              class="text-14-regular text-text-base cursor-default"
-              text={store.collapsed ? preview() : undefined}
+              text={props.collapsed ? preview() : undefined}
              duration={600}
              travel={25}
              edge={17}
@@ -161,7 +160,7 @@ export function SessionTodoDock(props: {
          <div class="ml-auto">
            <IconButton
              data-action="session-todo-toggle-button"
-              data-collapsed={store.collapsed ? "true" : "false"}
+              data-collapsed={props.collapsed ? "true" : "false"}
              icon="chevron-down"
              size="normal"
              variant="ghost"
@@ -172,16 +171,16 @@ export function SessionTodoDock(props: {
              }}
              onClick={(event) => {
                event.stopPropagation()
-                toggle()
+                props.onToggle()
              }}
-              aria-label={store.collapsed ? props.expandLabel : props.collapseLabel}
+              aria-label={props.collapsed ? props.expandLabel : props.collapseLabel}
            />
          </div>
        </div>

        <div
          data-slot="session-todo-list"
-          aria-hidden={store.collapsed || off()}
+          aria-hidden={props.collapsed || off()}
          classList={{
            "pointer-events-none": hide() > 0.1,
          }}
@@ -28,6 +28,12 @@ import { createOpenSessionFileTab, createSessionTabs, getTabReorderIndex, type S
 import { setSessionHandoff } from "@/pages/session/handoff"
 import { useSessionLayout } from "@/pages/session/session-layout"

+type RenderDiff = (SnapshotFileDiff & { file: string }) | VcsFileDiff
+
+function renderDiff(value: SnapshotFileDiff | VcsFileDiff): value is RenderDiff {
+  return typeof value.file === "string"
+}
+
 export function SessionSidePanel(props: {
  canReview: () => boolean
  diffs: () => (SnapshotFileDiff | VcsFileDiff)[]
@@ -70,7 +76,8 @@ export function SessionSidePanel(props: {
  })
  const treeWidth = createMemo(() => (fileOpen() ? `${layout.fileTree.width()}px` : "0px"))

-  const diffFiles = createMemo(() => props.diffs().map((d) => d.file))
+  const diffs = createMemo(() => props.diffs().filter(renderDiff))
+  const diffFiles = createMemo(() => diffs().map((d) => d.file))
  const kinds = createMemo(() => {
    const merge = (a: "add" | "del" | "mix" | undefined, b: "add" | "del" | "mix") => {
      if (!a) return b
@@ -81,7 +88,7 @@ export function SessionSidePanel(props: {
    const normalize = (p: string) => p.replaceAll("\\\\", "/").replace(/\/+$/, "")

    const out = new Map<string, "add" | "del" | "mix">()
-    for (const diff of props.diffs()) {
+    for (const diff of diffs()) {
      const file = normalize(diff.file)
      const kind = diff.status === "added" ? "add" : diff.status === "deleted" ? "del" : "mix"

@@ -1,6 +1,6 @@
 {
  "name": "@opencode-ai/console-app",
-  "version": "1.14.44",
+  "version": "1.14.48",
  "type": "module",
  "license": "MIT",
  "scripts": {
@@ -67,6 +67,7 @@
        display: inline-flex;
        align-items: center;
        justify-content: center;
+        flex-shrink: 0;
        padding: 0;
        background: transparent;
        border: none;
@@ -79,6 +80,7 @@
        }

        svg {
+          flex-shrink: 0;
          width: 16px;
          height: 16px;
        }
@@ -53,7 +53,7 @@ export function UsageSection() {
  }

  const calculateTotalOutputTokens = (u: Awaited<ReturnType<typeof getUsageInfo>>[0]) => {
-    return u.outputTokens + (u.reasoningTokens ?? 0)
+    return u.outputTokens
  }

  const goPrev = async () => {
@@ -47,6 +47,7 @@ import { Resource } from "@opencode-ai/console-resource"
 import { i18n, type Key } from "~/i18n"
 import { localeFromRequest } from "~/lib/language"
 import { createModelTpmLimiter } from "./modelTpmLimiter"
+import { createModelTpsLimiter } from "./modelTpsLimiter"

 type ZenData = Awaited<ReturnType<typeof ZenData.list>>
 type RetryOptions = {
@@ -129,6 +130,8 @@ export async function handler(
    logger.metric({ source: billingSource })
    const modelTpmLimiter = createModelTpmLimiter(modelInfo.providers)
    const modelTpmLimits = await modelTpmLimiter?.check()
+    const modelTpsLimiter = createModelTpsLimiter(modelInfo.providers)
+    const modelTpsLimits = await modelTpsLimiter?.check()

    const retriableRequest = async (retry: RetryOptions = { excludeProviders: [], retryCount: 0 }) => {
      const providerInfo = selectProvider(
@@ -142,6 +145,7 @@ export async function handler(
        retry,
        stickyProvider,
        modelTpmLimits,
+        modelTpsLimits,
      )
      validateModelSettings(billingSource, authInfo)
      updateProviderKey(authInfo, providerInfo)
@@ -294,14 +298,17 @@ export async function handler(

        let buffer = ""
        let responseLength = 0
+        let timestampFirstByte = 0
+        let timestampLastByte = 0

        function pump(): Promise<void> {
          return (
            reader?.read().then(async ({ done, value: rawValue }) => {
              if (done) {
+                const timestampLastByte = Date.now()
                logger.metric({
                  response_length: responseLength,
-                  "timestamp.last_byte": Date.now(),
+                  "timestamp.last_byte": timestampLastByte,
                })
                dataDumper?.flush()
                await rateLimiter?.track()
@@ -311,6 +318,13 @@ export async function handler(
                  const costInfo = calculateCost(modelInfo, usageInfo)
                  await trialLimiter?.track(usageInfo)
                  await modelTpmLimiter?.track(providerInfo.id, providerInfo.model, usageInfo)
+                  await modelTpsLimiter?.track(
+                    providerInfo.id,
+                    providerInfo.model,
+                    timestampFirstByte,
+                    timestampLastByte,
+                    usageInfo,
+                  )
                  await trackUsage(sessionId, billingSource, authInfo, modelInfo, providerInfo, usageInfo, costInfo)
                  await reload(billingSource, authInfo, costInfo)
                  const cost = calculateOccurredCost(billingSource, costInfo)
@@ -321,10 +335,10 @@ export async function handler(
              }

              if (responseLength === 0) {
-                const now = Date.now()
+                timestampFirstByte = Date.now()
                logger.metric({
-                  time_to_first_byte: now - startTimestamp,
-                  "timestamp.first_byte": now,
+                  time_to_first_byte: timestampFirstByte - startTimestamp,
+                  "timestamp.first_byte": timestampFirstByte,
                })
              }

@@ -478,6 +492,7 @@ export async function handler(
    retry: RetryOptions,
    stickyProvider: string | undefined,
    modelTpmLimits: Record<string, number> | undefined,
+    modelTpsLimits: Record<string, boolean> | undefined,
  ) {
    const modelProvider = (() => {
      // Byok is top priority b/c if user set their own API key, we should use it
@@ -509,6 +524,11 @@ export async function handler(
            const usage = modelTpmLimits?.[`${provider.id}/${provider.model}`] ?? 0
            return usage < provider.tpmLimit * 1_000_000
          })
+          .filter((provider) => {
+            if (!provider.tpsGoal) return true
+            const isLowTps = modelTpsLimits?.[`${provider.id}/${provider.model}`] ?? false
+            return !isLowTps
+          })
          .map((provider) => {
            topPriority = Math.min(topPriority, provider.priority)
            return provider
@@ -889,10 +909,6 @@ export async function handler(

    const inputCost = modelCost.input * inputTokens * 100
    const outputCost = modelCost.output * outputTokens * 100
-    const reasoningCost = (() => {
-      if (!reasoningTokens) return undefined
-      return modelCost.output * reasoningTokens * 100
-    })()
    const cacheReadCost = (() => {
      if (!cacheReadTokens) return undefined
      if (!modelCost.cacheRead) return undefined
@@ -909,17 +925,11 @@ export async function handler(
      return modelCost.cacheWrite1h * cacheWrite1hTokens * 100
    })()
    const totalCostInCent =
-      inputCost +
-      outputCost +
-      (reasoningCost ?? 0) +
-      (cacheReadCost ?? 0) +
-      (cacheWrite5mCost ?? 0) +
-      (cacheWrite1hCost ?? 0)
+      inputCost + outputCost + (cacheReadCost ?? 0) + (cacheWrite5mCost ?? 0) + (cacheWrite1hCost ?? 0)
    return {
      totalCostInCent,
      inputCost,
      outputCost,
-      reasoningCost,
      cacheReadCost,
      cacheWrite5mCost,
      cacheWrite1hCost,
@@ -941,8 +951,7 @@ export async function handler(
  ) {
    const { inputTokens, outputTokens, reasoningTokens, cacheReadTokens, cacheWrite5mTokens, cacheWrite1hTokens } =
      usageInfo
-    const { totalCostInCent, inputCost, outputCost, reasoningCost, cacheReadCost, cacheWrite5mCost, cacheWrite1hCost } =
-      costInfo
+    const { totalCostInCent, inputCost, outputCost, cacheReadCost, cacheWrite5mCost, cacheWrite1hCost } = costInfo

    logger.metric({
      "tokens.input": inputTokens,
@@ -953,14 +962,12 @@ export async function handler(
      "tokens.cache_write_1h": cacheWrite1hTokens,
      "cost.input.microcents": centsToMicroCents(inputCost),
      "cost.output.microcents": centsToMicroCents(outputCost),
-      "cost.reasoning.microcents": reasoningCost ? centsToMicroCents(reasoningCost) : undefined,
      "cost.cache_read.microcents": cacheReadCost ? centsToMicroCents(cacheReadCost) : undefined,
      "cost.cache_write.microcents": cacheWrite5mCost ? centsToMicroCents(cacheWrite5mCost) : undefined,
      "cost.total.microcents": centsToMicroCents(totalCostInCent),
      // deprecated - remove after May 20, 2026
      "cost.input": Math.round(inputCost),
      "cost.output": Math.round(outputCost),
-      "cost.reasoning": reasoningCost ? Math.round(reasoningCost) : undefined,
      "cost.cache_read": cacheReadCost ? Math.round(cacheReadCost) : undefined,
      "cost.cache_write_5m": cacheWrite5mCost ? Math.round(cacheWrite5mCost) : undefined,
      "cost.cache_write_1h": cacheWrite1hCost ? Math.round(cacheWrite1hCost) : undefined,
@@ -0,0 +1,89 @@
+import { and, Database, inArray, sql } from "@opencode-ai/console-core/drizzle/index.js"
+import { ModelTpsRateLimitTable } from "@opencode-ai/console-core/schema/ip.sql.js"
+import { UsageInfo } from "./provider/provider"
+
+export function createModelTpsLimiter(providers: { id: string; model: string; tpsGoal?: number }[]) {
+  const tpsGoals = Object.fromEntries(
+    providers.flatMap((p) => {
+      return p.tpsGoal ? [[`${p.id}/${p.model}`, p.tpsGoal]] : []
+    }),
+  )
+  const ids = Object.keys(tpsGoals)
+  if (ids.length === 0) return
+
+  const toInterval = (date: Date) =>
+    parseInt(
+      date
+        .toISOString()
+        .replace(/[^0-9]/g, "")
+        .substring(0, 12),
+    )
+  const now = Date.now()
+  const currInterval = toInterval(new Date(now))
+  const prevInterval = toInterval(new Date(now - 60 * 1000))
+
+  return {
+    check: async () => {
+      const data = await Database.use((tx) =>
+        tx
+          .select()
+          .from(ModelTpsRateLimitTable)
+          .where(
+            and(
+              inArray(ModelTpsRateLimitTable.id, ids),
+              inArray(ModelTpsRateLimitTable.interval, [currInterval, prevInterval]),
+            ),
+          ),
+      )
+
+      // convert to map of model to summed count across current and previous intervals
+      const result = data.reduce(
+        (acc, curr) => {
+          const existing = acc[curr.id] ?? { qualify: 0, unqualify: 0 }
+          acc[curr.id] = {
+            qualify: existing.qualify + curr.qualify,
+            unqualify: existing.unqualify + curr.unqualify,
+          }
+          return acc
+        },
+        {} as Record<string, { qualify: number; unqualify: number }>,
+      )
+
+      return Object.fromEntries(
+        Object.entries(result).map(([id, { qualify, unqualify }]) => {
+          const isLowTps = qualify + unqualify > 10 && qualify < unqualify
+          return [id, isLowTps]
+        }),
+      )
+    },
+    track: async (provider: string, model: string, tsFirstByte: number, tsLastByte: number, usageInfo: UsageInfo) => {
+      const id = `${provider}/${model}`
+      if (!ids.includes(id)) return
+      const tpsGoal = tpsGoals[id]
+      if (!tpsGoal) return
+      if (tsFirstByte <= 0 || tsLastByte <= 0) return
+      const tokens = usageInfo.outputTokens
+      if (tokens <= 10) return
+
+      const tps = (tokens / (tsLastByte - tsFirstByte)) * 1000
+      const qualify = tps >= tpsGoal ? 1 : 0
+      const unqualify = tps < tpsGoal ? 1 : 0
+      await Database.use((tx) =>
+        tx
+          .insert(ModelTpsRateLimitTable)
+          .values({
+            id,
+            interval: currInterval,
+            qualify,
+            unqualify,
+          })
+          .onDuplicateKeyUpdate({
+            set: {
+              qualify: sql`${ModelTpsRateLimitTable.qualify} + ${qualify}`,
+              unqualify: sql`${ModelTpsRateLimitTable.unqualify} + ${unqualify}`,
+            },
+          }),
+      )
+    },
+  }
+}
@@ -50,7 +50,7 @@ export const openaiHelper: ProviderHelper = ({ workspaceID }) => ({
    const cacheReadTokens = usage.input_tokens_details?.cached_tokens ?? undefined
    return {
      inputTokens: inputTokens - (cacheReadTokens ?? 0),
-      outputTokens: outputTokens - (reasoningTokens ?? 0),
+      outputTokens,
      reasoningTokens,
      cacheReadTokens,
      cacheWrite5mTokens: undefined,
@@ -0,0 +1,7 @@
+CREATE TABLE `model_tps_rate_limit` (
+	`id` varchar(255) NOT NULL,
+	`interval` bigint NOT NULL,
+	`qualify` int NOT NULL,
+	`unqualify` int NOT NULL,
+	CONSTRAINT PRIMARY KEY(`id`,`interval`)
+);
@@ -1,7 +1,7 @@
 {
  "$schema": "https://json.schemastore.org/package.json",
  "name": "@opencode-ai/console-core",
-  "version": "1.14.44",
+  "version": "1.14.48",
  "private": true,
  "type": "module",
  "license": "MIT",
@@ -36,6 +36,7 @@ export namespace ZenData {
        model: z.string(),
        priority: z.number().optional(),
        tpmLimit: z.number().optional(),
+        tpsGoal: z.number().optional(),
        weight: z.number().optional(),
        disabled: z.boolean().optional(),
        storeModel: z.string().optional(),
@@ -40,3 +40,14 @@ export const ModelTpmRateLimitTable = mysqlTable(
  },
  (table) => [primaryKey({ columns: [table.id, table.interval] })],
 )
+
+export const ModelTpsRateLimitTable = mysqlTable(
+  "model_tps_rate_limit",
+  {
+    id: varchar("id", { length: 255 }).notNull(),
+    interval: bigint("interval", { mode: "number" }).notNull(),
+    qualify: int("qualify").notNull(),
+    unqualify: int("unqualify").notNull(),
+  },
+  (table) => [primaryKey({ columns: [table.id, table.interval] })],
+)
@@ -1,6 +1,6 @@
 {
  "name": "@opencode-ai/console-function",
-  "version": "1.14.44",
+  "version": "1.14.48",
  "$schema": "https://json.schemastore.org/package.json",
  "private": true,
  "type": "module",
@@ -20,12 +20,10 @@
    "@ai-sdk/anthropic": "3.0.64",
    "@ai-sdk/openai": "3.0.48",
    "@ai-sdk/openai-compatible": "2.0.37",
-    "@hono/zod-validator": "catalog:",
    "@opencode-ai/console-core": "workspace:*",
    "@opencode-ai/console-resource": "workspace:*",
    "@openauthjs/openauth": "0.0.0-20250322224806",
    "ai": "catalog:",
-    "hono": "catalog:",
    "zod": "catalog:"
  }
 }
@@ -1,6 +1,6 @@
 {
  "name": "@opencode-ai/console-mail",
-  "version": "1.14.44",
+  "version": "1.14.48",
  "dependencies": {
    "@jsx-email/all": "2.2.3",
    "@jsx-email/cli": "1.4.3",
@@ -1,6 +1,6 @@
 {
  "$schema": "https://json.schemastore.org/package.json",
-  "version": "1.14.44",
+  "version": "1.14.48",
  "name": "@opencode-ai/core",
  "type": "module",
  "license": "MIT",
@@ -85,6 +85,7 @@ export const Flag = {
  OPENCODE_WORKSPACE_ID: process.env["OPENCODE_WORKSPACE_ID"],
  OPENCODE_EXPERIMENTAL_WORKSPACES: OPENCODE_EXPERIMENTAL || truthy("OPENCODE_EXPERIMENTAL_WORKSPACES"),
  OPENCODE_EXPERIMENTAL_EVENT_SYSTEM: OPENCODE_EXPERIMENTAL || truthy("OPENCODE_EXPERIMENTAL_EVENT_SYSTEM"),
+  OPENCODE_EXPERIMENTAL_SESSION_SWITCHING: OPENCODE_EXPERIMENTAL || truthy("OPENCODE_EXPERIMENTAL_SESSION_SWITCHING"),

  // Evaluated at access time (not module load) because tests, the CLI, and
  // external tooling set these env vars at runtime.
@@ -1,5 +1,4 @@
 import { Option, Schema, SchemaGetter } from "effect"
-import { zod, ZodOverride } from "./effect-zod"

 /**
 * Integer greater than zero.
@@ -21,7 +20,6 @@ export const optionalOmitUndefined = <S extends Schema.Top>(schema: S) =>
      decode: SchemaGetter.passthrough({ strict: false }),
      encode: SchemaGetter.transformOptional(Option.filter((value) => value !== undefined)),
    }),
-    Schema.annotate({ [ZodOverride]: zod(schema).optional() }),
  )

 /**
@@ -1,7 +1,7 @@
 {
  "name": "@opencode-ai/desktop",
  "private": true,
-  "version": "1.14.44",
+  "version": "1.14.48",
  "type": "module",
  "license": "MIT",
  "homepage": "https://opencode.ai",
@@ -1,6 +1,6 @@
 {
  "name": "@opencode-ai/enterprise",
-  "version": "1.14.44",
+  "version": "1.14.48",
  "private": true,
  "type": "module",
  "license": "MIT",
@@ -1,7 +1,7 @@
 id = "opencode"
 name = "OpenCode"
 description = "The open source coding agent."
-version = "1.14.44"
+version = "1.14.48"
 schema_version = 1
 authors = ["Anomaly"]
 repository = "https://github.com/anomalyco/opencode"
@@ -11,26 +11,26 @@ name = "OpenCode"
 icon = "./icons/opencode.svg"

 [agent_servers.opencode.targets.darwin-aarch64]
-archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.44/opencode-darwin-arm64.zip"
+archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.48/opencode-darwin-arm64.zip"
 cmd = "./opencode"
 args = ["acp"]

 [agent_servers.opencode.targets.darwin-x86_64]
-archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.44/opencode-darwin-x64.zip"
+archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.48/opencode-darwin-x64.zip"
 cmd = "./opencode"
 args = ["acp"]

 [agent_servers.opencode.targets.linux-aarch64]
-archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.44/opencode-linux-arm64.tar.gz"
+archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.48/opencode-linux-arm64.tar.gz"
 cmd = "./opencode"
 args = ["acp"]

 [agent_servers.opencode.targets.linux-x86_64]
-archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.44/opencode-linux-x64.tar.gz"
+archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.48/opencode-linux-x64.tar.gz"
 cmd = "./opencode"
 args = ["acp"]

 [agent_servers.opencode.targets.windows-x86_64]
-archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.44/opencode-windows-x64.zip"
+archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.48/opencode-windows-x64.zip"
 cmd = "./opencode.exe"
 args = ["acp"]
@@ -1,6 +1,6 @@
 {
  "name": "@opencode-ai/function",
-  "version": "1.14.44",
+  "version": "1.14.48",
  "$schema": "https://json.schemastore.org/package.json",
  "private": true,
  "type": "module",
@@ -0,0 +1,209 @@
+# @opencode-ai/http-recorder
+
+Record and replay HTTP and WebSocket traffic for Effect's `HttpClient`. Tests
+exercise real request shapes against deterministic, version-controlled
+cassettes — no manual mocks, no flakes from upstream drift.
+
+## Install
+
+Internal package; depended on as `@opencode-ai/http-recorder` from another
+workspace package.
+
+```ts
+import { HttpRecorder } from "@opencode-ai/http-recorder"
+```
+
+## Quickstart
+
+Provide `cassetteLayer(name)` in place of (or layered over) your `HttpClient`.
+By default the layer records on first run and replays on subsequent runs —
+no env-var ternary at the call site, and `CI=true` forces strict replay so
+missing cassettes fail loudly in CI rather than silently re-recording.
+
+```ts
+import { Effect } from "effect"
+import { HttpClient, HttpClientRequest } from "effect/unstable/http"
+import { HttpRecorder } from "@opencode-ai/http-recorder"
+
+const program = Effect.gen(function* () {
+  const http = yield* HttpClient.HttpClient
+  const response = yield* http.execute(HttpClientRequest.get("https://api.example.com/users/1"))
+  return yield* response.json
+})
+
+// Records if the cassette is missing, replays if it exists.
+// In CI (CI=true) always replays — fails loudly on missing fixtures.
+Effect.runPromise(program.pipe(Effect.provide(HttpRecorder.cassetteLayer("users/get-one"))))
+
+// Force a refresh — always hits upstream and overwrites.
+Effect.runPromise(program.pipe(Effect.provide(HttpRecorder.cassetteLayer("users/get-one", { mode: "record" }))))
+```
+
+## Modes
+
+| Mode          | Behavior                                                                            |
+| ------------- | ----------------------------------------------------------------------------------- |
+| `auto`        | Default. Replay if the cassette exists; record if missing. `CI=true` forces replay. |
+| `replay`      | Strict — match the request to a recorded interaction; error if none.                |
+| `record`      | Execute upstream, append the interaction, write the cassette.                       |
+| `passthrough` | Bypass the recorder entirely — just call upstream.                                  |
+
+## Cassette format
+
+A cassette is JSON at `test/fixtures/recordings/<name>.json`:
+
+```json
+{
+  "version": 1,
+  "metadata": { "name": "users/get-one", "recordedAt": "2026-05-09T..." },
+  "interactions": [
+    {
+      "transport": "http",
+      "request":  { "method": "GET", "url": "...", "headers": {...}, "body": "" },
+      "response": { "status": 200, "headers": {...}, "body": "..." }
+    }
+  ]
+}
+```
+
+Cassettes are normal source files — review them, diff them, commit them.
+
+## Request matching
+
+Replay walks the cassette in record order via an internal cursor: the Nth
+request executed at runtime is served by the Nth recorded interaction, and
+each one is validated as the cursor advances. Request equality is computed
+on canonicalized method, URL, headers, and JSON body (object keys sorted).
+
+This is deliberately strict — content-based dispatch was removed because
+it silently returns the first recorded response for repeated identical
+requests, masking state changes that retry/polling/cache-hit tests need to
+observe. If you reorder requests in a test, re-record the cassette.
+
+Supply your own matcher via `match: (incoming, recorded) => boolean` for
+custom equivalence (e.g. ignoring a timestamp field in the body).
+
+## Redaction & secret safety
+
+Cassettes get checked in, so the recorder is aggressive about not letting
+secrets escape. Redaction is configured by composing a `Redactor`:
+
+```ts
+import { HttpRecorder, Redactor } from "@opencode-ai/http-recorder"
+
+HttpRecorder.cassetteLayer("anthropic/messages", {
+  redactor: Redactor.defaults({
+    requestHeaders: { allow: ["content-type", "anthropic-version"] },
+    url: { transform: (url) => url.replace(/\/accounts\/[^/]+/, "/accounts/{account}") },
+    body: (parsed) => ({ ...(parsed as object), user_id: "{user}" }),
+  }),
+})
+```
+
+`Redactor.defaults({ … })` composes the four built-in redactors with your
+overrides. For full control, build the stack yourself:
+
+```ts
+const redactor = Redactor.compose(
+  Redactor.requestHeaders({ allow: ["content-type", "x-custom"] }),
+  Redactor.responseHeaders(),
+  Redactor.url({ query: ["session-id"] }),
+  Redactor.body((parsed) => /* … */),
+)
+```
+
+What each layer does:
+
+- **`requestHeaders` / `responseHeaders`** — strip headers to a small
+  allow-list (request default: `content-type`, `accept`, `openai-beta`;
+  response default: `content-type`). Sensitive headers within the
+  allow-list (`authorization`, `cookie`, API-key headers, AWS/GCP tokens,
+  …) are replaced with `[REDACTED]`.
+- **`url`** — query parameters matching common secret names (`api_key`,
+  `token`, `signature`, AWS signing params, …) are replaced with
+  `[REDACTED]`. URL user/password are replaced. `transform` runs after
+  built-in redaction for path-level scrubbing.
+- **`body`** — receives the parsed JSON request body and returns a redacted
+  version. No-op for non-JSON bodies.
+
+After assembling the cassette, the recorder scans every string for known
+secret patterns (Bearer tokens, `sk-…`, `sk-ant-…`, Google `AIza…` keys,
+AWS access keys, GitHub tokens, PEM blocks) and for values matching any
+environment variable named like a credential. If anything is found, the
+cassette is **not written** and the request fails with `UnsafeCassetteError`
+listing what was detected.
+
+## WebSocket recording
+
+WebSocket support records the open frame plus client/server message
+streams. It uses the shared `Cassette.Service`, so HTTP and WS interactions
+can live in the same cassette.
+
+```ts
+import { HttpRecorder } from "@opencode-ai/http-recorder"
+import { Effect } from "effect"
+
+const program = Effect.gen(function* () {
+  const cassette = yield* HttpRecorder.Cassette.Service
+  const executor = yield* HttpRecorder.makeWebSocketExecutor({
+    name: "ws/subscribe",
+    cassette,
+    live: liveExecutor,
+  })
+  // use executor.open(...)
+})
+```
+
+## Inspecting cassettes programmatically
+
+`Cassette.Service` exposes `read`, `append`, `exists`, and `list`. `read`
+returns the recorded interactions for a name; the file format is hidden
+behind the seam. Useful for CI checks:
+
+```ts
+import { HttpRecorder } from "@opencode-ai/http-recorder"
+import { Effect } from "effect"
+
+const audit = Effect.gen(function* () {
+  const cassettes = yield* HttpRecorder.Cassette.Service
+  const entries = yield* cassettes.list()
+  const issues = yield* Effect.forEach(entries, (entry) =>
+    cassettes
+      .read(entry.name)
+      .pipe(Effect.map((interactions) => ({ name: entry.name, findings: HttpRecorder.secretFindings(interactions) }))),
+  )
+  return issues.filter((i) => i.findings.length > 0)
+})
+```
+
+`cassetteLayer` is the batteries-included entry point — it provides
+`Cassette.fileSystem({ directory })` automatically. If you want to provide
+your own `Cassette.Service` (e.g. an in-memory adapter for the recorder's
+own unit tests), use `recordingLayer` and supply `Cassette.fileSystem` /
+`Cassette.memory` yourself.
+
+## Options reference
+
+```ts
+type RecordReplayOptions = {
+  mode?: "auto" | "replay" | "record" | "passthrough" // default: "auto" (CI=true forces "replay")
+  directory?: string // default: <cwd>/test/fixtures/recordings
+  metadata?: Record<string, unknown> // merged into cassette.metadata
+  redactor?: Redactor // default: Redactor.defaults()
+  match?: (incoming, recorded) => boolean // custom matcher
+}
+```
+
+## Layout
+
+| File           | Purpose                                                                          |
+| -------------- | -------------------------------------------------------------------------------- |
+| `effect.ts`    | `cassetteLayer` / `recordingLayer` — the `HttpClient` adapter.                   |
+| `websocket.ts` | `makeWebSocketExecutor` — WebSocket record/replay.                               |
+| `cassette.ts`  | `Cassette.Service` — reads/writes cassette files, accumulates state.             |
+| `recorder.ts`  | Shared transport plumbing: `UnsafeCassetteError`, `appendOrFail`, `ReplayState`. |
+| `redactor.ts`  | Composable `Redactor` — headers, url, body redaction.                            |
+| `redaction.ts` | Lower-level header/URL primitives + secret pattern detection.                    |
+| `schema.ts`    | Effect Schema definitions for the cassette JSON format.                          |
+| `storage.ts`   | Path resolution, JSON encode/decode, sync existence check.                       |
+| `matching.ts`  | Request matcher, canonicalization, sequential cursor, mismatch diagnostics.      |
@@ -1,6 +1,6 @@
 {
  "$schema": "https://json.schemastore.org/package.json",
-  "version": "1.14.44",
+  "version": "1.14.48",
  "name": "@opencode-ai/http-recorder",
  "type": "module",
  "license": "MIT",
@@ -1,54 +1,76 @@
-import { Context, Effect, FileSystem, Layer, PlatformError, Ref } from "effect"
+import { Context, Effect, FileSystem, Layer, Schema } from "effect"
+import * as fs from "node:fs"
 import * as path from "node:path"
-import { cassetteSecretFindings, type SecretFinding } from "./redaction"
-import type { Cassette, CassetteMetadata, Interaction } from "./schema"
-import { cassetteFor, cassettePath, DEFAULT_RECORDINGS_DIR, formatCassette, parseCassette } from "./storage"
+import { secretFindings, type SecretFinding } from "./redaction"
+import { decodeCassette, encodeCassette, type Cassette, type CassetteMetadata, type Interaction } from "./schema"

-export interface Entry {
-  readonly name: string
-  readonly path: string
+const DEFAULT_RECORDINGS_DIR = path.resolve(process.cwd(), "test", "fixtures", "recordings")
+
+export class CassetteNotFoundError extends Schema.TaggedErrorClass<CassetteNotFoundError>()("CassetteNotFoundError", {
+  cassetteName: Schema.String,
+}) {
+  override get message() {
+    return `Cassette "${this.cassetteName}" not found`
+  }
+}
+
+export interface AppendResult {
+  readonly findings: ReadonlyArray<SecretFinding>
 }

 export interface Interface {
-  readonly path: (name: string) => string
-  readonly read: (name: string) => Effect.Effect<Cassette, PlatformError.PlatformError>
-  readonly write: (name: string, cassette: Cassette) => Effect.Effect<void, PlatformError.PlatformError>
-  readonly append: (
-    name: string,
-    interaction: Interaction,
-    metadata: CassetteMetadata | undefined,
-  ) => Effect.Effect<
-    {
-      readonly cassette: Cassette
-      readonly findings: ReadonlyArray<SecretFinding>
-    },
-    PlatformError.PlatformError
-  >
+  readonly read: (name: string) => Effect.Effect<ReadonlyArray<Interaction>, CassetteNotFoundError>
+  readonly append: (name: string, interaction: Interaction, metadata?: CassetteMetadata) => Effect.Effect<AppendResult>
  readonly exists: (name: string) => Effect.Effect<boolean>
-  readonly list: () => Effect.Effect<ReadonlyArray<Entry>, PlatformError.PlatformError>
-  readonly scan: (cassette: Cassette) => ReadonlyArray<SecretFinding>
+  readonly list: () => Effect.Effect<ReadonlyArray<string>>
 }

 export class Service extends Context.Service<Service, Interface>()("@opencode-ai/http-recorder/Cassette") {}

-export const layer = (options: { readonly directory?: string } = {}) =>
+export const hasCassetteSync = (name: string, options: { readonly directory?: string } = {}) =>
+  fs.existsSync(path.join(options.directory ?? DEFAULT_RECORDINGS_DIR, `${name}.json`))
+
+const buildCassette = (
+  name: string,
+  interactions: ReadonlyArray<Interaction>,
+  metadata: CassetteMetadata | undefined,
+): Cassette => ({
+  version: 1,
+  metadata: { name, recordedAt: new Date().toISOString(), ...(metadata ?? {}) },
+  interactions,
+})
+
+const formatCassette = (cassette: Cassette) => `${JSON.stringify(encodeCassette(cassette), null, 2)}\n`
+
+const parseCassette = (raw: string) => decodeCassette(JSON.parse(raw))
+
+export const fileSystem = (
+  options: { readonly directory?: string } = {},
+): Layer.Layer<Service, never, FileSystem.FileSystem> =>
  Layer.effect(
    Service,
    Effect.gen(function* () {
-      const fileSystem = yield* FileSystem.FileSystem
+      const fs = yield* FileSystem.FileSystem
      const directory = options.directory ?? DEFAULT_RECORDINGS_DIR
-      const recorded = yield* Ref.make(new Map<string, ReadonlyArray<Interaction>>())
+      const recorded = new Map<string, { interactions: Interaction[]; findings: SecretFinding[] }>()
+      const directoriesEnsured = new Set<string>()

-      const pathFor = (name: string) => cassettePath(name, directory)
+      const cassettePath = (name: string) => path.join(directory, `${name}.json`)

-      const walk = (directory: string): Effect.Effect<ReadonlyArray<string>, PlatformError.PlatformError> =>
+      const ensureDirectory = (name: string) =>
        Effect.gen(function* () {
-          const entries = yield* fileSystem
-            .readDirectory(directory)
-            .pipe(Effect.catch(() => Effect.succeed([] as string[])))
+          const dir = path.dirname(cassettePath(name))
+          if (directoriesEnsured.has(dir)) return
+          yield* fs.makeDirectory(dir, { recursive: true }).pipe(Effect.orDie)
+          directoriesEnsured.add(dir)
+        })
+
+      const walk = (current: string): Effect.Effect<ReadonlyArray<string>> =>
+        Effect.gen(function* () {
+          const entries = yield* fs.readDirectory(current).pipe(Effect.catch(() => Effect.succeed([] as string[])))
          const nested = yield* Effect.forEach(entries, (entry) => {
-            const full = path.join(directory, entry)
-            return fileSystem.stat(full).pipe(
+            const full = path.join(current, entry)
+            return fs.stat(full).pipe(
              Effect.flatMap((stat) => (stat.type === "Directory" ? walk(full) : Effect.succeed([full]))),
              Effect.catch(() => Effect.succeed([] as string[])),
            )
@@ -56,53 +78,73 @@ export const layer = (options: { readonly directory?: string } = {}) =>
          return nested.flat()
        })

-      const read = Effect.fn("Cassette.read")(function* (name: string) {
-        return parseCassette(yield* fileSystem.readFileString(pathFor(name)))
+      return Service.of({
+        read: (name) =>
+          fs.readFileString(cassettePath(name)).pipe(
+            Effect.map((raw) => parseCassette(raw).interactions),
+            Effect.catch(() => Effect.fail(new CassetteNotFoundError({ cassetteName: name }))),
+          ),
+        append: (name, interaction, metadata) =>
+          Effect.gen(function* () {
+            const entry = recorded.get(name) ?? { interactions: [], findings: [] }
+            if (!recorded.has(name)) recorded.set(name, entry)
+            entry.interactions.push(interaction)
+            entry.findings.push(...secretFindings(interaction))
+            const cassette = buildCassette(name, entry.interactions, metadata)
+            const findings = [...entry.findings, ...secretFindings(cassette.metadata ?? {})]
+            if (findings.length === 0) {
+              yield* ensureDirectory(name)
+              yield* fs.writeFileString(cassettePath(name), formatCassette(cassette)).pipe(Effect.orDie)
+            }
+            return { findings }
+          }),
+        exists: (name) =>
+          fs.access(cassettePath(name)).pipe(
+            Effect.as(true),
+            Effect.catch(() => Effect.succeed(false)),
+          ),
+        list: () =>
+          walk(directory).pipe(
+            Effect.map((files) =>
+              files
+                .filter((file) => file.endsWith(".json"))
+                .map((file) =>
+                  path
+                    .relative(directory, file)
+                    .replace(/\\/g, "/")
+                    .replace(/\.json$/, ""),
+                )
+                .toSorted((a, b) => a.localeCompare(b)),
+            ),
+          ),
      })
-
-      const write = Effect.fn("Cassette.write")(function* (name: string, cassette: Cassette) {
-        yield* fileSystem.makeDirectory(path.dirname(pathFor(name)), { recursive: true })
-        yield* fileSystem.writeFileString(pathFor(name), formatCassette(cassette))
-      })
-
-      const append = Effect.fn("Cassette.append")(function* (
-        name: string,
-        interaction: Interaction,
-        metadata: CassetteMetadata | undefined,
-      ) {
-        const interactions = yield* Ref.updateAndGet(recorded, (previous) =>
-          new Map(previous).set(name, [...(previous.get(name) ?? []), interaction]),
-        )
-        const cassette = cassetteFor(name, interactions.get(name) ?? [], metadata)
-        const findings = cassetteSecretFindings(cassette)
-        if (findings.length === 0) yield* write(name, cassette)
-        return { cassette, findings }
-      })
-
-      const exists = Effect.fn("Cassette.exists")(function* (name: string) {
-        return yield* fileSystem.access(pathFor(name)).pipe(
-          Effect.as(true),
-          Effect.catch(() => Effect.succeed(false)),
-        )
-      })
-
-      const list = Effect.fn("Cassette.list")(function* () {
-        return (yield* walk(directory))
-          .filter((file) => file.endsWith(".json"))
-          .map((file) => ({
-            name: path
-              .relative(directory, file)
-              .replace(/\\/g, "/")
-              .replace(/\.json$/, ""),
-            path: file,
-          }))
-          .toSorted((a, b) => a.name.localeCompare(b.name))
-      })
-
-      return Service.of({ path: pathFor, read, write, append, exists, list, scan: cassetteSecretFindings })
    }),
  )

-export const defaultLayer = layer()
+export const memory = (initial: Record<string, ReadonlyArray<Interaction>> = {}): Layer.Layer<Service> =>
+  Layer.sync(Service, () => {
+    const stored = new Map<string, Interaction[]>(
+      Object.entries(initial).map(([name, interactions]) => [name, [...interactions]]),
+    )
+    const accumulatedFindings = new Map<string, SecretFinding[]>()

-export * as Cassette from "./cassette"
+    return Service.of({
+      read: (name) =>
+        stored.has(name)
+          ? Effect.succeed(stored.get(name) ?? [])
+          : Effect.fail(new CassetteNotFoundError({ cassetteName: name })),
+      append: (name, interaction, metadata) =>
+        Effect.sync(() => {
+          const existing = stored.get(name)
+          if (existing) existing.push(interaction)
+          else stored.set(name, [interaction])
+          const findings = accumulatedFindings.get(name)
+          if (findings) findings.push(...secretFindings(interaction))
+          else accumulatedFindings.set(name, [...secretFindings(interaction)])
+          if (metadata) accumulatedFindings.get(name)!.push(...secretFindings({ name, ...metadata }))
+          return { findings: accumulatedFindings.get(name) ?? [] }
+        }),
+      exists: (name) => Effect.sync(() => stored.has(name)),
+      list: () => Effect.sync(() => Array.from(stored.keys()).toSorted()),
+    })
+  })
@@ -1,95 +0,0 @@
-import { Option } from "effect"
-import { Headers, HttpBody, HttpClientRequest, UrlParams } from "effect/unstable/http"
-import { decodeJson } from "./matching"
-import { REDACTED, redactUrl, secretFindings } from "./redaction"
-import { httpInteractions, type Cassette, type RequestSnapshot } from "./schema"
-
-const safeText = (value: unknown) => {
-  if (value === undefined) return "undefined"
-  if (secretFindings(value).length > 0) return JSON.stringify(REDACTED)
-  const text = typeof value === "string" ? JSON.stringify(value) : JSON.stringify(value)
-  if (!text) return String(value)
-  return text.length > 300 ? `${text.slice(0, 300)}...` : text
-}
-
-const jsonBody = (body: string) => Option.getOrUndefined(decodeJson(body))
-
-const valueDiffs = (expected: unknown, received: unknown, base = "$", limit = 8): ReadonlyArray<string> => {
-  if (Object.is(expected, received)) return []
-  if (
-    expected &&
-    received &&
-    typeof expected === "object" &&
-    typeof received === "object" &&
-    !Array.isArray(expected) &&
-    !Array.isArray(received)
-  ) {
-    return [...new Set([...Object.keys(expected), ...Object.keys(received)])]
-      .toSorted()
-      .flatMap((key) =>
-        valueDiffs(
-          (expected as Record<string, unknown>)[key],
-          (received as Record<string, unknown>)[key],
-          `${base}.${key}`,
-          limit,
-        ),
-      )
-      .slice(0, limit)
-  }
-  if (Array.isArray(expected) && Array.isArray(received)) {
-    return Array.from({ length: Math.max(expected.length, received.length) }, (_, index) => index)
-      .flatMap((index) => valueDiffs(expected[index], received[index], `${base}[${index}]`, limit))
-      .slice(0, limit)
-  }
-  return [`${base} expected ${safeText(expected)}, received ${safeText(received)}`]
-}
-
-const headerDiffs = (expected: Record<string, string>, received: Record<string, string>) =>
-  [...new Set([...Object.keys(expected), ...Object.keys(received)])].toSorted().flatMap((key) => {
-    if (expected[key] === received[key]) return []
-    if (expected[key] === undefined) return [`  ${key} unexpected ${safeText(received[key])}`]
-    if (received[key] === undefined) return [`  ${key} missing expected ${safeText(expected[key])}`]
-    return [`  ${key} expected ${safeText(expected[key])}, received ${safeText(received[key])}`]
-  })
-
-export const requestDiff = (expected: RequestSnapshot, received: RequestSnapshot) => {
-  const lines = []
-  if (expected.method !== received.method) {
-    lines.push("method:", `  expected ${expected.method}, received ${received.method}`)
-  }
-  if (expected.url !== received.url) {
-    lines.push("url:", `  expected ${expected.url}`, `  received ${received.url}`)
-  }
-  const headers = headerDiffs(expected.headers, received.headers)
-  if (headers.length > 0) lines.push("headers:", ...headers.slice(0, 8))
-  const expectedBody = jsonBody(expected.body)
-  const receivedBody = jsonBody(received.body)
-  const body =
-    expectedBody !== undefined && receivedBody !== undefined
-      ? valueDiffs(expectedBody, receivedBody).map((line) => `  ${line}`)
-      : expected.body === received.body
-        ? []
-        : [`  expected ${safeText(expected.body)}, received ${safeText(received.body)}`]
-  if (body.length > 0) lines.push("body:", ...body)
-  return lines
-}
-
-export const mismatchDetail = (cassette: Cassette, incoming: RequestSnapshot) => {
-  const interactions = httpInteractions(cassette)
-  if (interactions.length === 0) return "cassette has no recorded HTTP interactions"
-  const ranked = interactions
-    .map((interaction, index) => ({ index, lines: requestDiff(interaction.request, incoming) }))
-    .toSorted((a, b) => a.lines.length - b.lines.length || a.index - b.index)
-  const best = ranked[0]
-  return ["no recorded interaction matched", `closest interaction: #${best.index + 1}`, ...best.lines].join("\n")
-}
-
-export const redactedErrorRequest = (request: HttpClientRequest.HttpClientRequest) =>
-  HttpClientRequest.makeWith(
-    request.method,
-    redactUrl(request.url),
-    UrlParams.empty,
-    Option.none(),
-    Headers.empty,
-    HttpBody.empty,
-  )
@@ -1,62 +1,36 @@
 import { NodeFileSystem } from "@effect/platform-node"
-import { Effect, Layer, Option, Ref } from "effect"
+import { Effect, Layer, Option } from "effect"
 import {
  FetchHttpClient,
+  Headers,
+  HttpBody,
  HttpClient,
  HttpClientError,
  HttpClientRequest,
  HttpClientResponse,
+  UrlParams,
 } from "effect/unstable/http"
-import { redactedErrorRequest, mismatchDetail, requestDiff } from "./diff"
-import { defaultMatcher, decodeJson, type RequestMatcher } from "./matching"
-import { redactHeaders, redactUrl, type SecretFinding } from "./redaction"
-import {
-  httpInteractions,
-  type Cassette,
-  type CassetteMetadata,
-  type HttpInteraction,
-  type ResponseSnapshot,
-} from "./schema"
 import * as CassetteService from "./cassette"
+import { defaultMatcher, selectSequential, type RequestMatcher } from "./matching"
+import { appendOrFail, makeReplayState, resolveAutoMode } from "./recorder"
+import { defaults, type Redactor } from "./redactor"
+import { redactUrl } from "./redaction"
+import { httpInteractions, type CassetteMetadata, type HttpInteraction, type ResponseSnapshot } from "./schema"

-export const DEFAULT_REQUEST_HEADERS: ReadonlyArray<string> = ["content-type", "accept", "openai-beta"]
-const DEFAULT_RESPONSE_HEADERS: ReadonlyArray<string> = ["content-type"]
-
-export type RecordReplayMode = "record" | "replay" | "passthrough"
+export type RecordReplayMode = "auto" | "record" | "replay" | "passthrough"

 export interface RecordReplayOptions {
  readonly mode?: RecordReplayMode
  readonly directory?: string
  readonly metadata?: CassetteMetadata
-  readonly redact?: {
-    readonly headers?: ReadonlyArray<string>
-    readonly query?: ReadonlyArray<string>
-    readonly url?: (url: string) => string
-  }
-  readonly requestHeaders?: ReadonlyArray<string>
-  readonly responseHeaders?: ReadonlyArray<string>
-  readonly redactBody?: (body: unknown) => unknown
-  readonly dispatch?: "match" | "sequential"
+  readonly redactor?: Redactor
  readonly match?: RequestMatcher
 }

-const responseHeaders = (
-  response: HttpClientResponse.HttpClientResponse,
-  allow: ReadonlyArray<string>,
-  redact: ReadonlyArray<string> | undefined,
-) => {
-  const merged = redactHeaders(response.headers as Record<string, string>, allow, redact)
-  if (!merged["content-type"]) merged["content-type"] = "text/event-stream"
-  return merged
-}
-
 const BINARY_CONTENT_TYPES: ReadonlyArray<string> = ["vnd.amazon.eventstream", "octet-stream"]

-const isBinaryContentType = (contentType: string | undefined) => {
-  if (!contentType) return false
-  const lower = contentType.toLowerCase()
-  return BINARY_CONTENT_TYPES.some((token) => lower.includes(token))
-}
+const isBinaryContentType = (contentType: string | undefined) =>
+  contentType !== undefined && BINARY_CONTENT_TYPES.some((token) => contentType.toLowerCase().includes(token))

 const captureResponseBody = (response: HttpClientResponse.HttpClientResponse, contentType: string | undefined) =>
  isBinaryContentType(contentType)
@@ -68,34 +42,19 @@ const captureResponseBody = (response: HttpClientResponse.HttpClientResponse, co
 const decodeResponseBody = (snapshot: ResponseSnapshot) =>
  snapshot.bodyEncoding === "base64" ? Buffer.from(snapshot.body, "base64") : snapshot.body

-const fixtureMissing = (request: HttpClientRequest.HttpClientRequest, name: string) =>
-  new HttpClientError.HttpClientError({
-    reason: new HttpClientError.TransportError({
-      request: redactedErrorRequest(request),
-      description: `Fixture "${name}" not found. Run with RECORD=true to create it.`,
-    }),
-  })
+export const redactedErrorRequest = (request: HttpClientRequest.HttpClientRequest) =>
+  HttpClientRequest.makeWith(
+    request.method,
+    redactUrl(request.url),
+    UrlParams.empty,
+    Option.none(),
+    Headers.empty,
+    HttpBody.empty,
+  )

-const fixtureMismatch = (request: HttpClientRequest.HttpClientRequest, name: string, detail: string) =>
+const transportError = (request: HttpClientRequest.HttpClientRequest, description: string) =>
  new HttpClientError.HttpClientError({
-    reason: new HttpClientError.TransportError({
-      request: redactedErrorRequest(request),
-      description: `Fixture "${name}" does not match the current request: ${detail}. Run with RECORD=true to update it.`,
-    }),
-  })
-
-const unsafeCassette = (
-  request: HttpClientRequest.HttpClientRequest,
-  name: string,
-  findings: ReadonlyArray<SecretFinding>,
-) =>
-  new HttpClientError.HttpClientError({
-    reason: new HttpClientError.TransportError({
-      request: redactedErrorRequest(request),
-      description: `Refusing to write cassette "${name}" because it contains possible secrets: ${findings
-        .map((item) => `${item.path} (${item.reason})`)
-        .join(", ")}`,
-    }),
+    reason: new HttpClientError.TransportError({ request: redactedErrorRequest(request), description }),
  })

 export const recordingLayer = (
@@ -107,61 +66,21 @@ export const recordingLayer = (
    Effect.gen(function* () {
      const upstream = yield* HttpClient.HttpClient
      const cassetteService = yield* CassetteService.Service
-      const requestHeadersAllow = options.requestHeaders ?? DEFAULT_REQUEST_HEADERS
-      const responseHeadersAllow = options.responseHeaders ?? DEFAULT_RESPONSE_HEADERS
+      const redactor = options.redactor ?? defaults()
      const match = options.match ?? defaultMatcher
-      const mode = options.mode ?? "replay"
-      const sequential = options.dispatch === "sequential"
-      const replay = yield* Ref.make<Cassette | undefined>(undefined)
-      const cursor = yield* Ref.make(0)
+      const requested = options.mode ?? "auto"
+      const mode = requested === "auto" ? yield* resolveAutoMode(cassetteService, name) : requested
+      const replay = yield* makeReplayState(cassetteService, name, httpInteractions)

      const snapshotRequest = (request: HttpClientRequest.HttpClientRequest) =>
        Effect.gen(function* () {
          const web = yield* HttpClientRequest.toWeb(request).pipe(Effect.orDie)
-          const raw = yield* Effect.promise(() => web.text())
-          const body = options.redactBody
-            ? Option.match(decodeJson(raw), {
-                onNone: () => raw,
-                onSome: (parsed) => JSON.stringify(options.redactBody?.(parsed)),
-              })
-            : raw
-          return {
+          return redactor.request({
            method: web.method,
-            url: redactUrl(web.url, options.redact?.query, options.redact?.url),
-            headers: redactHeaders(
-              Object.fromEntries(web.headers.entries()),
-              requestHeadersAllow,
-              options.redact?.headers,
-            ),
-            body,
-          }
-        })
-
-      const selectInteraction = (cassette: Cassette, incoming: HttpInteraction["request"]) =>
-        Effect.gen(function* () {
-          const interactions = httpInteractions(cassette)
-          if (sequential) {
-            const index = yield* Ref.get(cursor)
-            const interaction = interactions[index]
-            if (!interaction)
-              return { interaction, detail: `interaction ${index + 1} of ${interactions.length} not recorded` }
-            if (!match(incoming, interaction.request)) {
-              return { interaction: undefined, detail: requestDiff(interaction.request, incoming).join("\n") }
-            }
-            yield* Ref.update(cursor, (n) => n + 1)
-            return { interaction, detail: "" }
-          }
-          const interaction = interactions.find((candidate) => match(incoming, candidate.request))
-          return { interaction, detail: interaction ? "" : mismatchDetail(cassette, incoming) }
-        })
-
-      const loadReplay = (request: HttpClientRequest.HttpClientRequest) =>
-        Effect.gen(function* () {
-          const cached = yield* Ref.get(replay)
-          if (cached) return cached
-          const cassette = yield* cassetteService.read(name).pipe(Effect.mapError(() => fixtureMissing(request, name)))
-          yield* Ref.set(replay, cassette)
-          return cassette
+            url: web.url,
+            headers: Object.fromEntries(web.headers.entries()),
+            body: yield* Effect.promise(() => web.text()),
+          })
        })

      return HttpClient.make((request) => {
@@ -169,18 +88,21 @@ export const recordingLayer = (

        if (mode === "record") {
          return Effect.gen(function* () {
-            const currentRequest = yield* snapshotRequest(request)
+            const incoming = yield* snapshotRequest(request)
            const response = yield* upstream.execute(request)
-            const headers = responseHeaders(response, responseHeadersAllow, options.redact?.headers)
-            const captured = yield* captureResponseBody(response, headers["content-type"])
+            const captured = yield* captureResponseBody(response, response.headers["content-type"])
            const interaction: HttpInteraction = {
              transport: "http",
-              request: currentRequest,
-              response: { status: response.status, headers, ...captured },
+              request: incoming,
+              response: redactor.response({
+                status: response.status,
+                headers: response.headers as Record<string, string>,
+                ...captured,
+              }),
            }
-            const result = yield* cassetteService.append(name, interaction, options.metadata).pipe(Effect.orDie)
-            const findings = result.findings
-            if (findings.length > 0) return yield* unsafeCassette(request, name, findings)
+            yield* appendOrFail(cassetteService, name, interaction, options.metadata).pipe(
+              Effect.catchTag("UnsafeCassetteError", (error) => Effect.fail(transportError(request, error.message))),
+            )
            return HttpClientResponse.fromWeb(
              request,
              new Response(decodeResponseBody(interaction.response), interaction.response),
@@ -189,14 +111,21 @@ export const recordingLayer = (
        }

        return Effect.gen(function* () {
-          const cassette = yield* loadReplay(request)
          const incoming = yield* snapshotRequest(request)
-          const { interaction, detail } = yield* selectInteraction(cassette, incoming)
-          if (!interaction) return yield* fixtureMismatch(request, name, detail)
-
+          const interactions = yield* replay.load.pipe(
+            Effect.mapError(() =>
+              transportError(request, `Fixture "${name}" not found. Run locally to record it (CI=true forces replay).`),
+            ),
+          )
+          const result = selectSequential(interactions, incoming, match, yield* replay.cursor)
+          if (!result.interaction)
+            return yield* Effect.fail(
+              transportError(request, `Fixture "${name}" does not match the current request: ${result.detail}.`),
+            )
+          yield* replay.advance
          return HttpClientResponse.fromWeb(
            request,
-            new Response(decodeResponseBody(interaction.response), interaction.response),
+            new Response(decodeResponseBody(result.interaction.response), result.interaction.response),
          )
        })
      })
@@ -205,7 +134,7 @@ export const recordingLayer = (

 export const cassetteLayer = (name: string, options: RecordReplayOptions = {}): Layer.Layer<HttpClient.HttpClient> =>
  recordingLayer(name, options).pipe(
-    Layer.provide(CassetteService.layer({ directory: options.directory })),
+    Layer.provide(CassetteService.fileSystem({ directory: options.directory })),
    Layer.provide(FetchHttpClient.layer),
    Layer.provide(NodeFileSystem.layer),
  )
@@ -1,10 +1,26 @@
-export * from "./schema"
-export * from "./redaction"
-export * from "./matching"
-export * from "./diff"
-export * from "./storage"
-export * from "./websocket"
-export * from "./effect"
+export type {
+  CassetteMetadata,
+  HttpInteraction,
+  Interaction,
+  RequestSnapshot,
+  ResponseSnapshot,
+  WebSocketFrame,
+  WebSocketInteraction,
+} from "./schema"
+export { CassetteNotFoundError, hasCassetteSync } from "./cassette"
+export { defaultMatcher, type RequestMatcher } from "./matching"
+export { redactHeaders, redactUrl, secretFindings, type SecretFinding } from "./redaction"
+export { UnsafeCassetteError } from "./recorder"
+export { cassetteLayer, recordingLayer, type RecordReplayMode, type RecordReplayOptions } from "./effect"
+export {
+  makeWebSocketExecutor,
+  type WebSocketConnection,
+  type WebSocketExecutor,
+  type WebSocketRecordReplayOptions,
+  type WebSocketRequest,
+} from "./websocket"
+
 export * as Cassette from "./cassette"
+export * as Redactor from "./redactor"

 export * as HttpRecorder from "."
@@ -1,5 +1,6 @@
 import { Option, Schema } from "effect"
-import type { RequestSnapshot } from "./schema"
+import { REDACTED, secretFindings } from "./redaction"
+import type { HttpInteraction, RequestSnapshot } from "./schema"

 const JsonValue = Schema.fromJsonString(Schema.Unknown)
 export const decodeJson = Schema.decodeUnknownOption(JsonValue)
@@ -34,3 +35,72 @@ export const canonicalSnapshot = (snapshot: RequestSnapshot): string =>

 export const defaultMatcher: RequestMatcher = (incoming, recorded) =>
  canonicalSnapshot(incoming) === canonicalSnapshot(recorded)
+
+const safeText = (value: unknown) => {
+  if (value === undefined) return "undefined"
+  if (secretFindings(value).length > 0) return JSON.stringify(REDACTED)
+  const text = JSON.stringify(value)
+  if (!text) return String(value)
+  return text.length > 300 ? `${text.slice(0, 300)}...` : text
+}
+
+const jsonBody = (body: string) => Option.getOrUndefined(decodeJson(body))
+
+const valueDiffs = (expected: unknown, received: unknown, base = "$", limit = 8): ReadonlyArray<string> => {
+  if (Object.is(expected, received)) return []
+  if (isRecord(expected) && isRecord(received)) {
+    return [...new Set([...Object.keys(expected), ...Object.keys(received)])]
+      .toSorted()
+      .flatMap((key) => valueDiffs(expected[key], received[key], `${base}.${key}`, limit))
+      .slice(0, limit)
+  }
+  if (Array.isArray(expected) && Array.isArray(received)) {
+    return Array.from({ length: Math.max(expected.length, received.length) }, (_, index) => index)
+      .flatMap((index) => valueDiffs(expected[index], received[index], `${base}[${index}]`, limit))
+      .slice(0, limit)
+  }
+  return [`${base} expected ${safeText(expected)}, received ${safeText(received)}`]
+}
+
+const headerDiffs = (expected: Record<string, string>, received: Record<string, string>) =>
+  [...new Set([...Object.keys(expected), ...Object.keys(received)])].toSorted().flatMap((key) => {
+    if (expected[key] === received[key]) return []
+    if (expected[key] === undefined) return [`  ${key} unexpected ${safeText(received[key])}`]
+    if (received[key] === undefined) return [`  ${key} missing expected ${safeText(expected[key])}`]
+    return [`  ${key} expected ${safeText(expected[key])}, received ${safeText(received[key])}`]
+  })
+
+export const requestDiff = (expected: RequestSnapshot, received: RequestSnapshot): ReadonlyArray<string> => {
+  const lines: string[] = []
+  if (expected.method !== received.method) {
+    lines.push("method:", `  expected ${expected.method}, received ${received.method}`)
+  }
+  if (expected.url !== received.url) {
+    lines.push("url:", `  expected ${expected.url}`, `  received ${received.url}`)
+  }
+  const headers = headerDiffs(expected.headers, received.headers)
+  if (headers.length > 0) lines.push("headers:", ...headers.slice(0, 8))
+  const expectedBody = jsonBody(expected.body)
+  const receivedBody = jsonBody(received.body)
+  const body =
+    expectedBody !== undefined && receivedBody !== undefined
+      ? valueDiffs(expectedBody, receivedBody).map((line) => `  ${line}`)
+      : expected.body === received.body
+        ? []
+        : [`  expected ${safeText(expected.body)}, received ${safeText(received.body)}`]
+  if (body.length > 0) lines.push("body:", ...body)
+  return lines
+}
+
+export const selectSequential = (
+  interactions: ReadonlyArray<HttpInteraction>,
+  incoming: RequestSnapshot,
+  match: RequestMatcher,
+  index: number,
+): { readonly interaction: HttpInteraction | undefined; readonly detail: string } => {
+  const interaction = interactions[index]
+  if (!interaction) return { interaction, detail: `interaction ${index + 1} of ${interactions.length} not recorded` }
+  if (!match(incoming, interaction.request))
+    return { interaction: undefined, detail: requestDiff(interaction.request, incoming).join("\n") }
+  return { interaction, detail: "" }
+}
@@ -0,0 +1,73 @@
+import { Effect, Ref, Schema, Scope } from "effect"
+import type * as CassetteService from "./cassette"
+import type { CassetteNotFoundError } from "./cassette"
+import { SecretFindingSchema } from "./redaction"
+import type { CassetteMetadata, Interaction } from "./schema"
+
+export class UnsafeCassetteError extends Schema.TaggedErrorClass<UnsafeCassetteError>()("UnsafeCassetteError", {
+  cassetteName: Schema.String,
+  findings: Schema.Array(SecretFindingSchema),
+}) {
+  override get message() {
+    return `Refusing to write cassette "${this.cassetteName}" because it contains possible secrets: ${this.findings
+      .map((finding) => `${finding.path} (${finding.reason})`)
+      .join(", ")}`
+  }
+}
+
+export type ResolvedMode = "record" | "replay" | "passthrough"
+
+const isCI = () => {
+  const value = process.env.CI
+  return value !== undefined && value !== "" && value !== "false" && value !== "0"
+}
+
+export const resolveAutoMode = (cassette: CassetteService.Interface, name: string): Effect.Effect<ResolvedMode> =>
+  Effect.gen(function* () {
+    if (isCI()) return "replay"
+    return (yield* cassette.exists(name)) ? "replay" : "record"
+  })
+
+export const appendOrFail = (
+  cassette: CassetteService.Interface,
+  name: string,
+  interaction: Interaction,
+  metadata: CassetteMetadata | undefined,
+): Effect.Effect<void, UnsafeCassetteError> =>
+  cassette
+    .append(name, interaction, metadata)
+    .pipe(
+      Effect.flatMap(({ findings }) =>
+        findings.length === 0 ? Effect.void : Effect.fail(new UnsafeCassetteError({ cassetteName: name, findings })),
+      ),
+    )
+
+export interface ReplayState<T> {
+  readonly load: Effect.Effect<ReadonlyArray<T>, CassetteNotFoundError>
+  readonly cursor: Effect.Effect<number>
+  readonly advance: Effect.Effect<void>
+}
+
+export const makeReplayState = <T>(
+  cassette: CassetteService.Interface,
+  name: string,
+  project: (interactions: ReadonlyArray<Interaction>) => ReadonlyArray<T>,
+): Effect.Effect<ReplayState<T>, never, Scope.Scope> =>
+  Effect.gen(function* () {
+    const load = yield* Effect.cached(cassette.read(name).pipe(Effect.map(project)))
+    const position = yield* Ref.make(0)
+
+    yield* Effect.addFinalizer(() =>
+      Effect.gen(function* () {
+        const used = yield* Ref.get(position)
+        if (used === 0) return
+        const interactions = yield* load.pipe(Effect.orDie)
+        if (used < interactions.length)
+          yield* Effect.die(
+            new Error(`Unused recorded interactions in ${name}: used ${used} of ${interactions.length}`),
+          )
+      }),
+    )
+
+    return { load, cursor: Ref.get(position), advance: Ref.update(position, (n) => n + 1) }
+  })
@@ -1,5 +1,3 @@
-import type { Cassette } from "./schema"
-
 export const REDACTED = "[REDACTED]"

 const DEFAULT_REDACT_HEADERS = [
@@ -97,10 +95,13 @@ export const redactHeaders = (
  )
 }

-export type SecretFinding = {
-  readonly path: string
-  readonly reason: string
-}
+import { Schema } from "effect"
+
+export const SecretFindingSchema = Schema.Struct({
+  path: Schema.String,
+  reason: Schema.String,
+})
+export type SecretFinding = Schema.Schema.Type<typeof SecretFindingSchema>

 export const secretFindings = (value: unknown): ReadonlyArray<SecretFinding> =>
  stringEntries(value).flatMap((entry) => [
@@ -112,5 +113,3 @@ export const secretFindings = (value: unknown): ReadonlyArray<SecretFinding> =>
      .filter((item) => entry.value.includes(item.value))
      .map((item) => ({ path: entry.path, reason: `environment secret ${item.name}` })),
  ])
-
-export const cassetteSecretFindings = (cassette: Cassette) => secretFindings(cassette)
@@ -0,0 +1,76 @@
+import { Option } from "effect"
+import { decodeJson } from "./matching"
+import { redactHeaders, redactUrl } from "./redaction"
+import type { RequestSnapshot, ResponseSnapshot } from "./schema"
+
+export const DEFAULT_REQUEST_HEADERS: ReadonlyArray<string> = ["content-type", "accept", "openai-beta"]
+export const DEFAULT_RESPONSE_HEADERS: ReadonlyArray<string> = ["content-type"]
+
+const identity = <T>(value: T) => value
+
+export interface Redactor {
+  readonly request: (snapshot: RequestSnapshot) => RequestSnapshot
+  readonly response: (snapshot: ResponseSnapshot) => ResponseSnapshot
+}
+
+export const compose = (...redactors: ReadonlyArray<Partial<Redactor>>): Redactor => {
+  const requests = redactors.map((r) => r.request).filter((fn): fn is Redactor["request"] => fn !== undefined)
+  const responses = redactors.map((r) => r.response).filter((fn): fn is Redactor["response"] => fn !== undefined)
+  return {
+    request: requests.length === 0 ? identity : (snapshot) => requests.reduce((acc, fn) => fn(acc), snapshot),
+    response: responses.length === 0 ? identity : (snapshot) => responses.reduce((acc, fn) => fn(acc), snapshot),
+  }
+}
+
+export interface HeaderOptions {
+  readonly allow?: ReadonlyArray<string>
+  readonly redact?: ReadonlyArray<string>
+}
+
+export const requestHeaders = (options: HeaderOptions = {}): Partial<Redactor> => ({
+  request: (snapshot) => ({
+    ...snapshot,
+    headers: redactHeaders(snapshot.headers, options.allow ?? DEFAULT_REQUEST_HEADERS, options.redact),
+  }),
+})
+
+export const responseHeaders = (options: HeaderOptions = {}): Partial<Redactor> => ({
+  response: (snapshot) => ({
+    ...snapshot,
+    headers: redactHeaders(snapshot.headers, options.allow ?? DEFAULT_RESPONSE_HEADERS, options.redact),
+  }),
+})
+
+export interface UrlOptions {
+  readonly query?: ReadonlyArray<string>
+  readonly transform?: (url: string) => string
+}
+
+export const url = (options: UrlOptions = {}): Partial<Redactor> => ({
+  request: (snapshot) => ({ ...snapshot, url: redactUrl(snapshot.url, options.query, options.transform) }),
+})
+
+export const body = (transform: (parsed: unknown) => unknown): Partial<Redactor> => ({
+  request: (snapshot) => ({
+    ...snapshot,
+    body: Option.match(decodeJson(snapshot.body), {
+      onNone: () => snapshot.body,
+      onSome: (parsed) => JSON.stringify(transform(parsed)),
+    }),
+  }),
+})
+
+export interface DefaultRedactorOverrides {
+  readonly requestHeaders?: HeaderOptions
+  readonly responseHeaders?: HeaderOptions
+  readonly url?: UrlOptions
+  readonly body?: (parsed: unknown) => unknown
+}
+
+export const defaults = (overrides: DefaultRedactorOverrides = {}): Redactor =>
+  compose(
+    requestHeaders(overrides.requestHeaders),
+    responseHeaders(overrides.responseHeaders),
+    url(overrides.url),
+    ...(overrides.body ? [body(overrides.body)] : []),
+  )
@@ -52,9 +52,10 @@ export const isHttpInteraction = InteractionSchema.guards.http

 export const isWebSocketInteraction = InteractionSchema.guards.websocket

-export const httpInteractions = (cassette: Cassette) => cassette.interactions.filter(isHttpInteraction)
+export const httpInteractions = (interactions: ReadonlyArray<Interaction>) => interactions.filter(isHttpInteraction)

-export const webSocketInteractions = (cassette: Cassette) => cassette.interactions.filter(isWebSocketInteraction)
+export const webSocketInteractions = (interactions: ReadonlyArray<Interaction>) =>
+  interactions.filter(isWebSocketInteraction)

 export const CassetteSchema = Schema.Struct({
  version: Schema.Literal(1),
@@ -1,34 +0,0 @@
-import { Option } from "effect"
-import * as fs from "node:fs"
-import * as path from "node:path"
-import { encodeCassette, decodeCassette, type Cassette, type CassetteMetadata, type Interaction } from "./schema"
-
-export const DEFAULT_RECORDINGS_DIR = path.resolve(process.cwd(), "test", "fixtures", "recordings")
-
-export const cassettePath = (name: string, directory = DEFAULT_RECORDINGS_DIR) => path.join(directory, `${name}.json`)
-
-export const metadataFor = (name: string, metadata: CassetteMetadata | undefined): CassetteMetadata => ({
-  name,
-  recordedAt: new Date().toISOString(),
-  ...(metadata ?? {}),
-})
-
-export const cassetteFor = (
-  name: string,
-  interactions: ReadonlyArray<Interaction>,
-  metadata: CassetteMetadata | undefined,
-): Cassette => ({
-  version: 1,
-  metadata: metadataFor(name, metadata),
-  interactions,
-})
-
-export const formatCassette = (cassette: Cassette) => `${JSON.stringify(encodeCassette(cassette), null, 2)}\n`
-
-export const parseCassette = (raw: string) => decodeCassette(JSON.parse(raw))
-
-export const hasCassetteSync = (name: string, options: { readonly directory?: string } = {}) => {
-  const file = cassettePath(name, options.directory)
-  if (!fs.existsSync(file)) return false
-  return Option.isSome(Option.liftThrowable(parseCassette)(fs.readFileSync(file, "utf8")))
-}
@@ -2,10 +2,10 @@ import { Effect, Option, Ref, Scope, Stream } from "effect"
 import type { Headers } from "effect/unstable/http"
 import * as CassetteService from "./cassette"
 import { canonicalizeJson, decodeJson } from "./matching"
-import { redactHeaders, redactUrl, type SecretFinding } from "./redaction"
-import { webSocketInteractions, type CassetteMetadata, type WebSocketFrame, type WebSocketInteraction } from "./schema"
-
-export const DEFAULT_WEBSOCKET_REQUEST_HEADERS: ReadonlyArray<string> = ["content-type", "accept", "openai-beta"]
+import { appendOrFail, makeReplayState, resolveAutoMode } from "./recorder"
+import type { RecordReplayMode } from "./effect"
+import { defaults, type Redactor } from "./redactor"
+import { webSocketInteractions, type CassetteMetadata, type WebSocketFrame } from "./schema"

 export interface WebSocketRequest {
  readonly url: string
@@ -24,67 +24,36 @@ export interface WebSocketExecutor<E> {

 export interface WebSocketRecordReplayOptions<E> {
  readonly name: string
-  readonly mode?: "record" | "replay" | "passthrough"
+  readonly mode?: RecordReplayMode
  readonly metadata?: CassetteMetadata
  readonly cassette: CassetteService.Interface
  readonly live: WebSocketExecutor<E>
-  readonly redact?: {
-    readonly headers?: ReadonlyArray<string>
-    readonly query?: ReadonlyArray<string>
-    readonly url?: (url: string) => string
-  }
-  readonly requestHeaders?: ReadonlyArray<string>
+  readonly redactor?: Redactor
  readonly compareClientMessagesAsJson?: boolean
 }

-const headersRecord = (headers: Headers.Headers) =>
+const headersRecord = (headers: Headers.Headers): Record<string, string> =>
  Object.fromEntries(
-    Object.entries(headers as Record<string, unknown>)
-      .filter((entry): entry is [string, string] => typeof entry[1] === "string")
-      .toSorted(([a], [b]) => a.localeCompare(b)),
+    Object.entries(headers as Record<string, unknown>).filter(
+      (entry): entry is [string, string] => typeof entry[1] === "string",
+    ),
  )

-const openSnapshot = (
-  request: WebSocketRequest,
-  options: Pick<WebSocketRecordReplayOptions<never>, "redact" | "requestHeaders"> = {},
-) => ({
-  url: redactUrl(request.url, options.redact?.query, options.redact?.url),
-  headers: redactHeaders(
-    headersRecord(request.headers),
-    options.requestHeaders ?? DEFAULT_WEBSOCKET_REQUEST_HEADERS,
-    options.redact?.headers,
-  ),
-})
-
-const textFrame = (body: string): WebSocketFrame => ({ kind: "text", body })
-
-const frameText = (frame: WebSocketFrame) => {
-  if (frame.kind === "text") return frame.body
-  return new TextDecoder().decode(Buffer.from(frame.body, "base64"))
-}
-
-const frameMessage = (frame: WebSocketFrame) =>
-  frame.kind === "text" ? frame.body : new Uint8Array(Buffer.from(frame.body, "base64"))
-
-const receivedFrame = (message: string | Uint8Array): WebSocketFrame =>
+const encodeFrame = (message: string | Uint8Array): WebSocketFrame =>
  typeof message === "string"
-    ? textFrame(message)
+    ? { kind: "text", body: message }
    : { kind: "binary", body: Buffer.from(message).toString("base64"), bodyEncoding: "base64" }

-const unsafeCassette = (name: string, findings: ReadonlyArray<SecretFinding>) =>
-  new Error(
-    `Refusing to write WebSocket cassette "${name}" because it contains possible secrets: ${findings
-      .map((item) => `${item.path} (${item.reason})`)
-      .join(", ")}`,
-  )
+const decodeFrameMessage = (frame: WebSocketFrame): string | Uint8Array =>
+  frame.kind === "text" ? frame.body : new Uint8Array(Buffer.from(frame.body, "base64"))

-const mismatch = (message: string, actual: unknown, expected: unknown) =>
-  new Error(`${message}: expected ${JSON.stringify(expected)}, received ${JSON.stringify(actual)}`)
+const decodeFrameText = (frame: WebSocketFrame) =>
+  frame.kind === "text" ? frame.body : new TextDecoder().decode(Buffer.from(frame.body, "base64"))

 const assertEqual = (message: string, actual: unknown, expected: unknown) =>
  Effect.sync(() => {
    if (JSON.stringify(actual) === JSON.stringify(expected)) return
-    throw mismatch(message, actual, expected)
+    throw new Error(`${message}: expected ${JSON.stringify(expected)}, received ${JSON.stringify(actual)}`)
  })

 const jsonOrText = (value: string) => Option.match(decodeJson(value), { onNone: () => value, onSome: canonicalizeJson })
@@ -94,7 +63,7 @@ const compareClientMessage = (actual: string, expected: WebSocketFrame | undefin
    return Effect.sync(() => {
      throw new Error(`Unexpected WebSocket client frame ${index + 1}: ${actual}`)
    })
-  const expectedText = frameText(expected)
+  const expectedText = decodeFrameText(expected)
  if (!asJson) return assertEqual(`WebSocket client frame ${index + 1}`, actual, expectedText)
  return assertEqual(`WebSocket client JSON frame ${index + 1}`, jsonOrText(actual), jsonOrText(expectedText))
 }
@@ -103,7 +72,18 @@ export const makeWebSocketExecutor = <E>(
  options: WebSocketRecordReplayOptions<E>,
 ): Effect.Effect<WebSocketExecutor<E>, never, Scope.Scope> =>
  Effect.gen(function* () {
-    const mode = options.mode ?? "replay"
+    const requested = options.mode ?? "auto"
+    const mode = requested === "auto" ? yield* resolveAutoMode(options.cassette, options.name) : requested
+    const redactor = options.redactor ?? defaults()
+    const openSnapshot = (request: WebSocketRequest) => {
+      const redacted = redactor.request({
+        method: "GET",
+        url: request.url,
+        headers: headersRecord(request.headers),
+        body: "",
+      })
+      return { url: redacted.url, headers: redacted.headers }
+    }

    if (mode === "passthrough") return options.live

@@ -118,21 +98,21 @@ export const makeWebSocketExecutor = <E>(
            const closeOnce = Effect.gen(function* () {
              if (yield* Ref.getAndSet(closed, true)) return
              yield* connection.close
-              const result = yield* options.cassette
-                .append(
-                  options.name,
-                  { transport: "websocket", open: openSnapshot(request, options), client, server },
-                  options.metadata,
-                )
-                .pipe(Effect.orDie)
-              if (result.findings.length > 0) yield* Effect.die(unsafeCassette(options.name, result.findings))
+              yield* appendOrFail(
+                options.cassette,
+                options.name,
+                { transport: "websocket", open: openSnapshot(request), client, server },
+                options.metadata,
+              ).pipe(Effect.orDie)
            })
            return {
-              sendText: (message: string) =>
-                connection.sendText(message).pipe(Effect.tap(() => Effect.sync(() => client.push(textFrame(message))))),
+              sendText: (message) =>
+                connection
+                  .sendText(message)
+                  .pipe(Effect.tap(() => Effect.sync(() => client.push(encodeFrame(message))))),
              messages: connection.messages.pipe(
                Stream.map((message) => {
-                  server.push(receivedFrame(message))
+                  server.push(encodeFrame(message))
                  return message
                }),
              ),
@@ -142,44 +122,20 @@ export const makeWebSocketExecutor = <E>(
      }
    }

-    const replay = yield* Ref.make<{ readonly interactions: ReadonlyArray<WebSocketInteraction> } | undefined>(
-      undefined,
-    )
-    const cursor = yield* Ref.make(0)
-
-    yield* Effect.addFinalizer(() =>
-      Effect.gen(function* () {
-        const input = yield* Ref.get(replay)
-        if (!input) return
-        yield* assertEqual(
-          `Unused recorded WebSocket interactions in ${options.name}`,
-          yield* Ref.get(cursor),
-          input.interactions.length,
-        )
-      }),
-    )
-
-    const loadReplay = Effect.fn("WebSocketRecorder.loadReplay")(function* () {
-      const cached = yield* Ref.get(replay)
-      if (cached) return cached
-      const input = {
-        interactions: webSocketInteractions(yield* options.cassette.read(options.name).pipe(Effect.orDie)),
-      }
-      yield* Ref.set(replay, input)
-      return input
-    })
+    const replay = yield* makeReplayState(options.cassette, options.name, webSocketInteractions)

    return {
-      open: (request) => {
-        return Effect.gen(function* () {
-          const input = yield* loadReplay()
-          const index = yield* Ref.getAndUpdate(cursor, (value) => value + 1)
-          const interaction = input.interactions[index]
+      open: (request) =>
+        Effect.gen(function* () {
+          const interactions = yield* replay.load.pipe(Effect.orDie)
+          const index = yield* replay.cursor
+          const interaction = interactions[index]
          if (!interaction) return yield* Effect.die(new Error(`No recorded WebSocket interaction for ${request.url}`))
-          yield* assertEqual(`WebSocket open frame ${index + 1}`, openSnapshot(request, options), interaction.open)
+          yield* replay.advance
+          yield* assertEqual(`WebSocket open frame ${index + 1}`, openSnapshot(request), interaction.open)
          const messageIndex = yield* Ref.make(0)
          return {
-            sendText: (message: string) =>
+            sendText: (message) =>
              Effect.gen(function* () {
                const current = yield* Ref.getAndUpdate(messageIndex, (value) => value + 1)
                yield* compareClientMessage(
@@ -189,7 +145,7 @@ export const makeWebSocketExecutor = <E>(
                  options.compareClientMessagesAsJson === true,
                )
              }),
-            messages: Stream.fromIterable(interaction.server).pipe(Stream.map(frameMessage)),
+            messages: Stream.fromIterable(interaction.server).pipe(Stream.map(decodeFrameMessage)),
            close: Effect.gen(function* () {
              yield* assertEqual(
                `WebSocket client frame count for interaction ${index + 1}`,
@@ -198,7 +154,6 @@ export const makeWebSocketExecutor = <E>(
              )
            }),
          }
-        })
-      },
+        }),
    }
  })
@@ -0,0 +1,10 @@
+/* This file is auto-generated by SST. Do not edit. */
+/* tslint:disable */
+/* eslint-disable */
+/* deno-fmt-ignore-file */
+/* biome-ignore-all lint: auto-generated */
+
+/// <reference path="../../sst-env.d.ts" />
+
+import "sst"
+export {}
@@ -6,7 +6,16 @@ import * as fs from "node:fs"
 import * as os from "node:os"
 import * as path from "node:path"
 import { HttpRecorder } from "../src"
-import { redactedErrorRequest } from "../src/diff"
+import { redactedErrorRequest } from "../src/effect"
+import type { Interaction } from "../src/schema"
+
+const seedCassetteDirectory = (directory: string, name: string, interactions: ReadonlyArray<Interaction>) =>
+  Effect.runPromise(
+    Effect.gen(function* () {
+      const cassette = yield* HttpRecorder.Cassette.Service
+      yield* Effect.forEach(interactions, (interaction) => cassette.append(name, interaction))
+    }).pipe(Effect.provide(HttpRecorder.Cassette.fileSystem({ directory })), Effect.provide(NodeFileSystem.layer)),
+  )

 const post = (url: string, body: object) =>
  Effect.gen(function* () {
@@ -33,7 +42,7 @@ const runRecorder = <A, E>(effect: Effect.Effect<A, E, HttpRecorder.Cassette.Ser
    Effect.scoped(
      effect.pipe(
        Effect.provide(
-          HttpRecorder.Cassette.layer({ directory: fs.mkdtempSync(path.join(os.tmpdir(), "http-recorder-")) }),
+          HttpRecorder.Cassette.fileSystem({ directory: fs.mkdtempSync(path.join(os.tmpdir(), "http-recorder-")) }),
        ),
        Effect.provide(NodeFileSystem.layer),
      ),
@@ -108,7 +117,7 @@ describe("http-recorder", () => {

  test("detects secret-looking values without returning the secret", () => {
    expect(
-      HttpRecorder.cassetteSecretFindings({
+      HttpRecorder.secretFindings({
        version: 1,
        interactions: [
          {
@@ -136,7 +145,7 @@ describe("http-recorder", () => {

  test("detects secret-looking values inside metadata", () => {
    expect(
-      HttpRecorder.cassetteSecretFindings({
+      HttpRecorder.secretFindings({
        version: 1,
        metadata: { token: "sk-123456789012345678901234" },
        interactions: [],
@@ -144,60 +153,42 @@ describe("http-recorder", () => {
    ).toEqual([{ path: "metadata.token", reason: "API key" }])
  })

-  test("formats websocket cassettes with shared metadata", () => {
-    const cassette = HttpRecorder.cassetteFor(
-      "websocket/basic",
-      [
-        {
-          transport: "websocket",
-          open: { url: "wss://example.test/realtime", headers: { "content-type": "application/json" } },
-          client: [{ kind: "text", body: JSON.stringify({ type: "response.create" }) }],
-          server: [{ kind: "text", body: JSON.stringify({ type: "response.completed" }) }],
-        },
-      ],
-      { provider: "openai" },
-    )
+  test("replays websocket interactions seeded into the in-memory cassette adapter", async () => {
+    await Effect.runPromise(
+      Effect.scoped(
+        Effect.gen(function* () {
+          const cassette = yield* HttpRecorder.Cassette.Service
+          const executor = yield* HttpRecorder.makeWebSocketExecutor({
+            name: "websocket/replay",
+            cassette,
+            compareClientMessagesAsJson: true,
+            live: { open: () => Effect.die(new Error("unexpected live WebSocket open")) },
+          })
+          const connection = yield* executor.open({
+            url: "wss://example.test/realtime",
+            headers: Headers.fromInput({ "content-type": "application/json" }),
+          })
+          yield* connection.sendText(JSON.stringify({ type: "response.create" }))
+          const messages: Array<string | Uint8Array> = []
+          yield* connection.messages.pipe(Stream.runForEach((message) => Effect.sync(() => messages.push(message))))
+          yield* connection.close

-    expect(cassette.metadata).toMatchObject({ name: "websocket/basic", provider: "openai" })
-    expect(HttpRecorder.parseCassette(HttpRecorder.formatCassette(cassette))).toEqual(cassette)
-  })
-
-  test("replays websocket interactions from the shared cassette service", async () => {
-    await runRecorder(
-      Effect.gen(function* () {
-        const cassette = yield* HttpRecorder.Cassette.Service
-        yield* cassette.write(
-          "websocket/replay",
-          HttpRecorder.cassetteFor(
-            "websocket/replay",
-            [
-              {
-                transport: "websocket",
-                open: { url: "wss://example.test/realtime", headers: { "content-type": "application/json" } },
-                client: [{ kind: "text", body: JSON.stringify({ type: "response.create" }) }],
-                server: [{ kind: "text", body: JSON.stringify({ type: "response.completed" }) }],
-              },
-            ],
-            undefined,
+          expect(messages).toEqual([JSON.stringify({ type: "response.completed" })])
+        }).pipe(
+          Effect.provide(
+            HttpRecorder.Cassette.memory({
+              "websocket/replay": [
+                {
+                  transport: "websocket",
+                  open: { url: "wss://example.test/realtime", headers: { "content-type": "application/json" } },
+                  client: [{ kind: "text", body: JSON.stringify({ type: "response.create" }) }],
+                  server: [{ kind: "text", body: JSON.stringify({ type: "response.completed" }) }],
+                },
+              ],
+            }),
          ),
-        )
-        const executor = yield* HttpRecorder.makeWebSocketExecutor({
-          name: "websocket/replay",
-          cassette,
-          compareClientMessagesAsJson: true,
-          live: { open: () => Effect.die(new Error("unexpected live WebSocket open")) },
-        })
-        const connection = yield* executor.open({
-          url: "wss://example.test/realtime",
-          headers: Headers.fromInput({ "content-type": "application/json" }),
-        })
-        yield* connection.sendText(JSON.stringify({ type: "response.create" }))
-        const messages: Array<string | Uint8Array> = []
-        yield* connection.messages.pipe(Stream.runForEach((message) => Effect.sync(() => messages.push(message))))
-        yield* connection.close
-
-        expect(messages).toEqual([JSON.stringify({ type: "response.completed" })])
-      }),
+        ),
+      ),
    )
  })

@@ -227,34 +218,22 @@ describe("http-recorder", () => {
        yield* connection.messages.pipe(Stream.runDrain)
        yield* connection.close

-        expect(yield* cassette.read("websocket/record")).toMatchObject({
-          metadata: { name: "websocket/record", provider: "test" },
-          interactions: [
-            {
-              transport: "websocket",
-              open: { url: "wss://example.test/realtime", headers: { "content-type": "application/json" } },
-              client: [{ kind: "text", body: JSON.stringify({ type: "response.create" }) }],
-              server: [{ kind: "text", body: JSON.stringify({ type: "response.completed" }) }],
-            },
-          ],
-        })
+        expect(yield* cassette.read("websocket/record")).toMatchObject([
+          {
+            transport: "websocket",
+            open: { url: "wss://example.test/realtime", headers: { "content-type": "application/json" } },
+            client: [{ kind: "text", body: JSON.stringify({ type: "response.create" }) }],
+            server: [{ kind: "text", body: JSON.stringify({ type: "response.completed" }) }],
+          },
+        ])
      }),
    )
  })

-  test("default matcher dispatches multi-interaction cassettes by request shape", async () => {
-    await run(
-      Effect.gen(function* () {
-        expect(yield* post("https://example.test/echo", { step: 2 })).toBe('{"reply":"second"}')
-        expect(yield* post("https://example.test/echo", { step: 1 })).toBe('{"reply":"first"}')
-      }),
-    )
-  })
-
-  test("sequential dispatch returns recorded responses in order for identical requests", async () => {
+  test("replay returns recorded responses in order for identical requests", async () => {
    await runWith(
      "record-replay/retry",
-      { dispatch: "sequential" },
+      {},
      Effect.gen(function* () {
        expect(yield* post("https://example.test/poll", { id: "job_1" })).toBe('{"status":"pending"}')
        expect(yield* post("https://example.test/poll", { id: "job_1" })).toBe('{"status":"complete"}')
@@ -262,21 +241,8 @@ describe("http-recorder", () => {
    )
  })

-  test("default matcher returns the first match for identical requests", async () => {
-    await runWith(
-      "record-replay/retry",
-      {},
-      Effect.gen(function* () {
-        expect(yield* post("https://example.test/poll", { id: "job_1" })).toBe('{"status":"pending"}')
-        expect(yield* post("https://example.test/poll", { id: "job_1" })).toBe('{"status":"pending"}')
-      }),
-    )
-  })
-
-  test("sequential dispatch reports cursor exhaustion when more requests are made than recorded", async () => {
-    await runWith(
-      "record-replay/multi-step",
-      { dispatch: "sequential" },
+  test("replay reports cursor exhaustion when more requests are made than recorded", async () => {
+    await run(
      Effect.gen(function* () {
        yield* post("https://example.test/echo", { step: 1 })
        yield* post("https://example.test/echo", { step: 2 })
@@ -286,10 +252,8 @@ describe("http-recorder", () => {
    )
  })

-  test("sequential dispatch still validates each recorded request", async () => {
-    await runWith(
-      "record-replay/multi-step",
-      { dispatch: "sequential" },
+  test("replay validates each recorded request in order", async () => {
+    await run(
      Effect.gen(function* () {
        yield* post("https://example.test/echo", { step: 1 })
        const exit = yield* Effect.exit(post("https://example.test/echo", { step: 3 }))
@@ -300,14 +264,56 @@ describe("http-recorder", () => {
    )
  })

-  test("mismatch diagnostics show closest redacted request differences", async () => {
+  test("auto mode replays when the cassette exists", async () => {
+    const directory = fs.mkdtempSync(path.join(os.tmpdir(), "http-recorder-auto-"))
+    await seedCassetteDirectory(directory, "auto-replay", [
+      {
+        transport: "http",
+        request: {
+          method: "POST",
+          url: "https://example.test/echo",
+          headers: { "content-type": "application/json" },
+          body: JSON.stringify({ step: 1 }),
+        },
+        response: { status: 200, headers: { "content-type": "application/json" }, body: '{"reply":"hi"}' },
+      },
+    ])
+
+    const result = await runWith(
+      "auto-replay",
+      { directory, mode: "auto" },
+      post("https://example.test/echo", { step: 1 }),
+    )
+    expect(result).toBe('{"reply":"hi"}')
+  })
+
+  test("auto mode forces replay when CI=true even if cassette is missing", async () => {
+    const directory = fs.mkdtempSync(path.join(os.tmpdir(), "http-recorder-auto-ci-"))
+    const previous = process.env.CI
+    process.env.CI = "true"
+    try {
+      const exit = await Effect.runPromise(
+        Effect.exit(
+          post("https://example.test/echo", { step: 1 }).pipe(
+            Effect.provide(HttpRecorder.cassetteLayer("missing-cassette", { directory, mode: "auto" })),
+          ),
+        ),
+      )
+      expect(Exit.isFailure(exit)).toBe(true)
+      expect(failureText(exit)).toContain('Fixture "missing-cassette" not found')
+    } finally {
+      if (previous === undefined) delete process.env.CI
+      else process.env.CI = previous
+    }
+  })
+
+  test("mismatch diagnostics show redacted request differences against the expected interaction", async () => {
    await run(
      Effect.gen(function* () {
        const exit = yield* Effect.exit(
          post("https://example.test/echo?api_key=secret-value", { step: 3, token: "sk-123456789012345678901234" }),
        )
        const message = failureText(exit)
-        expect(message).toContain("closest interaction: #1")
        expect(message).toContain("url:")
        expect(message).toContain("https://example.test/echo?api_key=%5BREDACTED%5D")
        expect(message).toContain("body:")
@@ -8,6 +8,10 @@
 - In `Effect.gen`, yield yieldable errors directly (`return yield* new MyError(...)`) instead of `Effect.fail(new MyError(...))`.
 - Use `Effect.void` instead of `Effect.succeed(undefined)` when the successful value is intentionally void.

+## Conventions
+
+Per-type constructors live on the type's namespace, not as top-level re-exports. Use `Message.user(...)`, `Message.assistant(...)`, `Message.tool(...)`, `ToolDefinition.make(...)`, `ToolCallPart.make(...)`, `ToolResultPart.make(...)`, `ToolChoice.make(...)`, `ToolChoice.named(...)`, `SystemPart.make(...)`, and `GenerationOptions.make(...)` directly. The top-level `LLM` namespace is reserved for the request-shaped call API: `LLM.request`, `LLM.generate`, `LLM.stream`, `LLM.model`, `LLM.updateRequest`, `LLM.generateObject`. Two ways to construct the same thing is one too many.
+
 ## Tests

 - Use `testEffect(...)` from `test/lib/effect.ts` for tests requiring Effect layers.
@@ -166,12 +170,12 @@ If you find yourself copying a 3-to-5-line snippet between two protocols, lift i
 Tool loops are represented in common messages and events:

 ```ts
-const call = LLM.toolCall({ id: "call_1", name: "lookup", input: { query: "weather" } })
-const result = LLM.toolMessage({ id: "call_1", name: "lookup", result: { forecast: "sunny" } })
+const call = ToolCallPart.make({ id: "call_1", name: "lookup", input: { query: "weather" } })
+const result = Message.tool({ id: "call_1", name: "lookup", result: { forecast: "sunny" } })

 const followUp = LLM.request({
  model,
-  messages: [LLM.user("Weather?"), LLM.assistant([call]), result],
+  messages: [Message.user("Weather?"), Message.assistant([call]), result],
 })
 ```

@@ -289,6 +293,6 @@ Filters apply in replay and record mode. Combine them with `RECORD=true` when re

 **Binary response bodies.** Most providers stream text (SSE, JSON). AWS Bedrock streams binary AWS event-stream frames whose CRC32 fields would be mangled by a UTF-8 round-trip — those bodies are stored as base64 with `bodyEncoding: "base64"` on the response snapshot. Detection is by `Content-Type` in `@opencode-ai/http-recorder` (currently `application/vnd.amazon.eventstream` and `application/octet-stream`); cassettes for SSE/JSON routes omit the field and decode as text.

-**Matching strategies.** Replay defaults to structural matching, which finds an interaction by comparing method, URL, allow-listed headers, and the canonical JSON body. This is the right choice for tool loops because each round's request differs (the message history grows). For scenarios where successive requests are byte-identical and expect different responses (retries, polling), pass `dispatch: "sequential"` in `RecordReplayOptions` — replay then walks the cassette in record order via an internal cursor. `scriptedResponses` (in `test/lib/http.ts`) is the deterministic counterpart for tests that don't need a live provider; it scripts response bodies in order without reading from disk.
+**Matching strategy.** Replay walks the cassette in record order via an internal cursor: the Nth runtime request is served by the Nth recorded interaction, and each one is validated by comparing method, URL, allow-listed headers, and the canonical JSON body. This handles tool loops (each round's request differs as history grows) and retry/polling scenarios (successive byte-identical requests with different responses) uniformly. If a test reorders its requests, re-record the cassette. `scriptedResponses` (in `test/lib/http.ts`) is the deterministic counterpart for tests that don't need a live provider; it scripts response bodies in order without reading from disk.

 Do not blanket re-record an entire test file when adding one cassette. `RECORD=true` rewrites every recorded case that runs, and provider streams contain volatile IDs, timestamps, fingerprints, and obfuscation fields. Prefer deleting the one cassette you intend to refresh, or run a focused test pattern that only registers the scenario you want to record. Keep stable existing cassettes unchanged unless their request shape or expected behavior changed.
@@ -0,0 +1,129 @@
+# @opencode-ai/llm
+
+Schema-first LLM core for opencode. One typed request, response, event, and tool language; provider quirks live in adapters, not in calling code.
+
+```ts
+import { Effect } from "effect"
+import { LLM, LLMClient } from "@opencode-ai/llm"
+import { OpenAI } from "@opencode-ai/llm/providers"
+
+const model = OpenAI.model("gpt-4o-mini", { apiKey: process.env.OPENAI_API_KEY })
+
+const request = LLM.request({
+  model,
+  system: "You are concise.",
+  prompt: "Say hello in one short sentence.",
+  generation: { maxTokens: 40 },
+})
+
+const program = Effect.gen(function* () {
+  const response = yield* LLMClient.generate(request)
+  console.log(response.text)
+})
+```
+
+Run `LLMClient.stream(request)` instead of `generate` when you want incremental `LLMEvent`s. The event stream is provider-neutral — same shape across OpenAI Chat, OpenAI Responses, Anthropic Messages, Gemini, Bedrock Converse, and any OpenAI-compatible deployment.
+
+## Public API
+
+- **`LLM.request({...})`** — build a provider-neutral `LLMRequest`. Accepts ergonomic inputs (`system: string`, `prompt: string`) that normalize into the canonical Schema classes.
+- **`LLM.generate` / `LLM.stream`** — re-exported from `LLMClient` for one-import use.
+- **`LLM.user(...)` / `LLM.assistant(...)` / `LLM.toolMessage(...)`** — message constructors.
+- **`LLM.toolCall(...)` / `LLM.toolResult(...)` / `LLM.toolDefinition(...)`** — tool-related parts.
+- **`LLMClient.prepare(request)`** — compile a request through protocol body construction, validation, and HTTP preparation without sending. Useful for inspection and testing.
+- **`LLMEvent.is.*`** — typed guards (`is.text`, `is.toolCall`, `is.requestFinish`, …) for filtering streams.
+
+## Caching
+
+Prompt caching is **on by default**. Every `LLMRequest` resolves to `cache: "auto"` unless the caller opts out with `cache: "none"`. Each protocol translates `CacheHint`s to its wire format (`cache_control` on Anthropic, `cachePoint` on Bedrock; OpenAI and Gemini do implicit caching server-side and don't need inline markers — auto is a no-op there).
+
+### Auto placement
+
+`"auto"` places three breakpoints — last tool definition, last system part, latest user message. The last-user-message boundary is the load-bearing detail: in a tool-use loop, a single user turn expands into many assistant/tool round-trips, all sharing that prefix. Caching at that boundary lets every intra-turn API call hit.
+
+The math justifies the default: Anthropic's 5-minute cache write is 1.25× base, read is 0.1×, so a single reuse within 5 minutes already wins. One-shot completions below the per-model minimum-cacheable-token threshold silently no-op on the wire, so the worst case is harmless.
+
+### Opting out
+
+```ts
+LLM.request({
+  model,
+  system,
+  prompt: "one-off question",
+  cache: "none",
+})
+```
+
+### Granular policy
+
+```ts
+cache: {
+  tools?: boolean,
+  system?: boolean,
+  messages?: "latest-user-message" | "latest-assistant" | { tail: number },
+  ttlSeconds?: number,         // ≥ 3600 → 1h on Anthropic/Bedrock; else 5m
+}
+```
+
+### Manual hints
+
+Inline `CacheHint` on any text / system / tool / tool-result part overrides automatic placement. The auto policy preserves manual hints; it only fills gaps.
+
+```ts
+LLM.request({
+  model,
+  system: [
+    { type: "text", text: "stable system prompt", cache: { type: "ephemeral" } },
+  ],
+  ...
+})
+```
+
+### Provider behavior table
+
+| Protocol                | `cache: "auto"`                                                           |
+| ----------------------- | ------------------------------------------------------------------------- |
+| Anthropic Messages      | emits up to 3 `cache_control` markers (4-breakpoint cap enforced)         |
+| Bedrock Converse        | emits up to 3 `cachePoint` blocks (4-breakpoint cap enforced)             |
+| OpenAI Chat / Responses | no-op (implicit caching above 1024 tokens)                                |
+| Gemini                  | no-op (implicit caching on 2.5+; explicit `CachedContent` is out-of-band) |
+
+Normalized cache usage is read back into `response.usage.cacheReadInputTokens` and `cacheWriteInputTokens` across every provider.
+
+## Providers
+
+Each provider exports a `model(...)` helper that records identity, protocol, capabilities, auth, and defaults.
+
+```ts
+import { Anthropic } from "@opencode-ai/llm/providers"
+
+const model = Anthropic.model("claude-sonnet-4-6", {
+  apiKey: process.env.ANTHROPIC_API_KEY,
+})
+```
+
+Included providers: OpenAI, Anthropic, Google (Gemini), Amazon Bedrock, Azure OpenAI, Cloudflare, GitHub Copilot, OpenRouter, xAI, plus generic OpenAI-compatible helpers for DeepSeek, Cerebras, Groq, Fireworks, Together, etc.
+
+## Provider options & HTTP overlays
+
+Three escape hatches in order of stability:
+
+1. **`generation`** — portable knobs (`maxTokens`, `temperature`, `topP`, `topK`, penalties, seed, stop).
+2. **`providerOptions: { <provider>: {...} }`** — typed-at-the-facade provider-specific knobs (OpenAI `promptCacheKey`, Anthropic `thinking`, Gemini `thinkingConfig`, OpenRouter routing).
+3. **`http: { body, headers, query }`** — last-resort serializable overlays merged into the final HTTP request. Reach for this only when a stable typed path doesn't yet exist.
+
+Model-level defaults are overridden by request-level values for each axis.
+
+## Routes
+
+Adding a new model or deployment is usually 5–15 lines using `Route.make({ protocol, transport, ... })`. The four orthogonal pieces are protocol (body construction + stream parsing), transport (endpoint + auth + framing + encoding), defaults, and capabilities. See `AGENTS.md` for the architectural detail.
+
+## Effect
+
+This package is built on Effect. Public methods return `Effect` or `Stream`; provide `LLMClient.layer` (the default registers every shipped route) for runtime dispatch. The example at `example/tutorial.ts` is a runnable walkthrough.
+
+## See also
+
+- `AGENTS.md` — architecture, route construction, contributor guide
+- `example/tutorial.ts` — runnable end-to-end walkthrough
+- `test/provider/*.test.ts` — fixture-first protocol tests; `*.recorded.test.ts` files cover live cassettes
@@ -184,7 +184,7 @@ const FakeProtocol = Protocol.make<FakeBody, string, string, void>({
  stream: {
    event: Schema.String,
    initial: () => undefined,
-    step: (_, frame) => Effect.succeed([undefined, [{ type: "text-delta", text: frame }]] as const),
+    step: (_, frame) => Effect.succeed([undefined, [{ type: "text-delta", id: "text-0", text: frame }]] as const),
    onHalt: () => [{ type: "request-finish", reason: "stop" }],
  },
 })
@@ -1,6 +1,6 @@
 {
  "$schema": "https://json.schemastore.org/package.json",
-  "version": "1.14.44",
+  "version": "1.14.48",
  "name": "@opencode-ai/llm",
  "type": "module",
  "license": "MIT",
@@ -0,0 +1,111 @@
+// Apply an `LLMRequest.cache` policy by injecting `CacheHint`s onto the parts
+// the policy designates. Runs once at compile time, before the per-protocol
+// body builder, so the existing inline-hint lowering path handles the rest.
+//
+// The default `"auto"` shape places one breakpoint at the last tool definition,
+// one at the last system part, and one at the latest user message. This
+// matches what production agent harnesses (LangChain's caching middleware,
+// kern-ai's 10x cost-reduction playbook) converge on for tool-use loops: the
+// latest user message stays put while a single turn explodes into many
+// assistant/tool round-trips, so caching at that boundary lets every
+// intra-turn API call hit the prefix.
+//
+// Manual `cache: CacheHint` placements on individual parts are preserved —
+// this function only fills gaps the caller left empty.
+import { CacheHint, type CachePolicy, type CachePolicyObject } from "./schema/options"
+import { LLMRequest, Message, ToolDefinition, type ContentPart } from "./schema/messages"
+
+const AUTO: CachePolicyObject = {
+  tools: true,
+  system: true,
+  messages: "latest-user-message",
+}
+
+const NONE: CachePolicyObject = {}
+
+// Resolution rules:
+//   - undefined   → "auto" — caching is on by default. The math favors it:
+//                   Anthropic 5m-cache write is 1.25x base, read is 0.1x,
+//                   so a single reuse within 5 minutes already wins.
+//   - "auto"      → tools + system + latest user msg.
+//   - "none"      → no auto placement; manual `CacheHint`s still flow.
+//   - object form → exactly what the caller asked for.
+const resolve = (policy: CachePolicy | undefined): CachePolicyObject => {
+  if (policy === undefined || policy === "auto") return AUTO
+  if (policy === "none") return NONE
+  return policy
+}
+
+// Protocols whose wire format ignores inline cache markers (OpenAI's implicit
+// prefix caching, Gemini's implicit + out-of-band CachedContent). Skip the
+// whole policy pass for these — emitting hints would be harmless but pointless.
+const RESPECTS_INLINE_HINTS = new Set(["anthropic-messages", "bedrock-converse"])
+
+const makeHint = (ttlSeconds: number | undefined): CacheHint =>
+  ttlSeconds !== undefined ? new CacheHint({ type: "ephemeral", ttlSeconds }) : new CacheHint({ type: "ephemeral" })
+
+const markLastTool = (tools: ReadonlyArray<ToolDefinition>, hint: CacheHint): ReadonlyArray<ToolDefinition> => {
+  if (tools.length === 0) return tools
+  const last = tools.length - 1
+  if (tools[last]!.cache) return tools
+  return tools.map((tool, i) => (i === last ? new ToolDefinition({ ...tool, cache: hint }) : tool))
+}
+
+const markLastSystem = (system: LLMRequest["system"], hint: CacheHint): LLMRequest["system"] => {
+  if (system.length === 0) return system
+  const last = system.length - 1
+  if (system[last]!.cache) return system
+  return system.map((part, i) => (i === last ? { ...part, cache: hint } : part))
+}
+
+const lastIndexOfRole = (messages: ReadonlyArray<Message>, role: Message["role"]): number =>
+  messages.findLastIndex((m) => m.role === role)
+
+// Mark the last text part of `messages[index]`. If no text part exists, mark
+// the last content part regardless of type — that's the breakpoint position
+// in tool-result-only messages too.
+const markMessageAt = (messages: ReadonlyArray<Message>, index: number, hint: CacheHint): ReadonlyArray<Message> => {
+  if (index < 0 || index >= messages.length) return messages
+  const target = messages[index]!
+  if (target.content.length === 0) return messages
+  const lastTextIndex = target.content.findLastIndex((part) => part.type === "text")
+  const markAt = lastTextIndex >= 0 ? lastTextIndex : target.content.length - 1
+  const existing = target.content[markAt]!
+  if ("cache" in existing && existing.cache) return messages
+  const nextContent = target.content.map((part, i) => (i === markAt ? ({ ...part, cache: hint } as ContentPart) : part))
+  const next = new Message({ ...target, content: nextContent })
+  // Single pass over `messages`, substituting the one updated entry. Long
+  // conversations call this on every request, so avoid `.map()` here — its
+  // closure dispatch and identity copies show up in profiling.
+  const result = messages.slice()
+  result[index] = next
+  return result
+}
+
+const markMessages = (
+  messages: ReadonlyArray<Message>,
+  strategy: NonNullable<CachePolicyObject["messages"]>,
+  hint: CacheHint,
+): ReadonlyArray<Message> => {
+  if (messages.length === 0) return messages
+  if (strategy === "latest-user-message") return markMessageAt(messages, lastIndexOfRole(messages, "user"), hint)
+  if (strategy === "latest-assistant") return markMessageAt(messages, lastIndexOfRole(messages, "assistant"), hint)
+  const start = Math.max(0, messages.length - strategy.tail)
+  let next = messages
+  for (let i = start; i < messages.length; i++) next = markMessageAt(next, i, hint)
+  return next
+}
+
+export const applyCachePolicy = (request: LLMRequest): LLMRequest => {
+  if (!RESPECTS_INLINE_HINTS.has(request.model.route)) return request
+  const policy = resolve(request.cache)
+  if (!policy.tools && !policy.system && !policy.messages) return request
+
+  const hint = makeHint(policy.ttlSeconds)
+  const tools = policy.tools ? markLastTool(request.tools, hint) : request.tools
+  const system = policy.system ? markLastSystem(request.system, hint) : request.system
+  const messages = policy.messages ? markMessages(request.messages, policy.messages, hint) : request.messages
+
+  if (tools === request.tools && system === request.system && messages === request.messages) return request
+  return LLMRequest.update(request, { tools, system, messages })
+}
@@ -44,32 +44,8 @@ export type RequestInput = Omit<

 export const limits = modelLimits

-export const text = Message.text
-
-export const system = SystemPart.make
-
-export const message = Message.make
-
-export const user = Message.user
-
-export const assistant = Message.assistant
-
 export const model = modelRef

-export const toolDefinition = ToolDefinition.make
-
-export const toolCall = ToolCallPart.make
-
-export const toolResult = ToolResultPart.make
-
-export const toolMessage = Message.tool
-
-export const toolChoiceName = ToolChoice.named
-
-export const toolChoice = ToolChoice.make
-
-export const generation = GenerationOptions.make
-
 export const generate = LLMClient.generate

 export const stream = LLMClient.stream
@@ -95,10 +71,10 @@ export const request = (input: RequestInput) => {
  return new LLMRequest({
    ...rest,
    system: SystemPart.content(requestSystem),
-    messages: [...(messages?.map(message) ?? []), ...(prompt === undefined ? [] : [user(prompt)])],
-    tools: tools?.map(toolDefinition) ?? [],
-    toolChoice: requestToolChoice ? toolChoice(requestToolChoice) : undefined,
-    generation: requestGeneration === undefined ? undefined : generation(requestGeneration),
+    messages: [...(messages?.map(Message.make) ?? []), ...(prompt === undefined ? [] : [Message.user(prompt)])],
+    tools: tools?.map(ToolDefinition.make) ?? [],
+    toolChoice: requestToolChoice ? ToolChoice.make(requestToolChoice) : undefined,
+    generation: requestGeneration === undefined ? undefined : GenerationOptions.make(requestGeneration),
    providerOptions: requestProviderOptions,
    http: requestHttp === undefined ? undefined : HttpOptions.make(requestHttp),
  })
@@ -5,10 +5,10 @@ import { Endpoint } from "../route/endpoint"
 import { Framing } from "../route/framing"
 import { Protocol } from "../route/protocol"
 import {
+  LLMEvent,
  Usage,
  type CacheHint,
  type FinishReason,
-  type LLMEvent,
  type LLMRequest,
  type ProviderMetadata,
  type ToolCallPart,
@@ -16,6 +16,8 @@ import {
  type ToolResultPart,
 } from "../schema"
 import { JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared"
+import * as Cache from "./utils/cache"
+import { Lifecycle } from "./utils/lifecycle"
 import { ToolStream } from "./utils/tool-stream"

 const ADAPTER = "anthropic-messages"
@@ -25,7 +27,10 @@ export const PATH = "/messages"
 // =============================================================================
 // Request Body Schema
 // =============================================================================
-const AnthropicCacheControl = Schema.Struct({ type: Schema.tag("ephemeral") })
+const AnthropicCacheControl = Schema.Struct({
+  type: Schema.tag("ephemeral"),
+  ttl: Schema.optional(Schema.Literals(["5m", "1h"])),
+})

 const AnthropicTextBlock = Schema.Struct({
  type: Schema.tag("text"),
@@ -186,6 +191,7 @@ type AnthropicEvent = Schema.Schema.Type<typeof AnthropicEvent>
 interface ParserState {
  readonly tools: ToolStream.State<number>
  readonly usage?: Usage
+  readonly lifecycle: Lifecycle.State
 }

 const invalid = ProviderShared.invalidRequest
@@ -193,8 +199,24 @@ const invalid = ProviderShared.invalidRequest
 // =============================================================================
 // Request Lowering
 // =============================================================================
-const cacheControl = (cache: CacheHint | undefined) =>
-  cache?.type === "ephemeral" ? { type: "ephemeral" as const } : undefined
+// Anthropic accepts at most 4 explicit cache_control breakpoints per request,
+// across `tools`, `system`, and `messages`. Beyond the cap the API returns a
+// 400 — so the lowering layer counts emitted markers and silently drops any
+// that exceed it.
+const ANTHROPIC_BREAKPOINT_CAP = 4
+
+const EPHEMERAL_5M = { type: "ephemeral" as const }
+const EPHEMERAL_1H = { type: "ephemeral" as const, ttl: "1h" as const }
+
+const cacheControl = (breakpoints: Cache.Breakpoints, cache: CacheHint | undefined) => {
+  if (cache?.type !== "ephemeral" && cache?.type !== "persistent") return undefined
+  if (breakpoints.remaining <= 0) {
+    breakpoints.dropped += 1
+    return undefined
+  }
+  breakpoints.remaining -= 1
+  return Cache.ttlBucket(cache.ttlSeconds) === "1h" ? EPHEMERAL_1H : EPHEMERAL_5M
+}

 const anthropicMetadata = (metadata: Record<string, unknown>): ProviderMetadata => ({ anthropic: metadata })

@@ -204,10 +226,11 @@ const signatureFromMetadata = (metadata: ProviderMetadata | undefined): string |
  return typeof anthropic.signature === "string" ? anthropic.signature : undefined
 }

-const lowerTool = (tool: ToolDefinition): AnthropicTool => ({
+const lowerTool = (breakpoints: Cache.Breakpoints, tool: ToolDefinition): AnthropicTool => ({
  name: tool.name,
  description: tool.description,
  input_schema: tool.inputSchema,
+  cache_control: cacheControl(breakpoints, tool.cache),
 })

 const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>) =>
@@ -249,7 +272,10 @@ const lowerServerToolResult = Effect.fn("AnthropicMessages.lowerServerToolResult
  return { type: wireType, tool_use_id: part.id, content: part.result.value } satisfies AnthropicServerToolResultBlock
 })

-const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (request: LLMRequest) {
+const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
+  request: LLMRequest,
+  breakpoints: Cache.Breakpoints,
+) {
  const messages: AnthropicMessage[] = []

  for (const message of request.messages) {
@@ -258,7 +284,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (re
      for (const part of message.content) {
        if (!ProviderShared.supportsContent(part, ["text"]))
          return yield* ProviderShared.unsupportedContent("Anthropic Messages", "user", ["text"])
-        content.push({ type: "text", text: part.text, cache_control: cacheControl(part.cache) })
+        content.push({ type: "text", text: part.text, cache_control: cacheControl(breakpoints, part.cache) })
      }
      messages.push({ role: "user", content })
      continue
@@ -268,7 +294,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (re
      const content: AnthropicAssistantBlock[] = []
      for (const part of message.content) {
        if (part.type === "text") {
-          content.push({ type: "text", text: part.text, cache_control: cacheControl(part.cache) })
+          content.push({ type: "text", text: part.text, cache_control: cacheControl(breakpoints, part.cache) })
          continue
        }
        if (part.type === "reasoning") {
@@ -304,6 +330,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (re
        tool_use_id: part.id,
        content: ProviderShared.toolResultText(part),
        is_error: part.result.type === "error" ? true : undefined,
+        cache_control: cacheControl(breakpoints, part.cache),
      })
    }
    messages.push({ role: "user", content })
@@ -330,18 +357,33 @@ const lowerThinking = Effect.fn("AnthropicMessages.lowerThinking")(function* (re
 const fromRequest = Effect.fn("AnthropicMessages.fromRequest")(function* (request: LLMRequest) {
  const toolChoice = request.toolChoice ? yield* lowerToolChoice(request.toolChoice) : undefined
  const generation = request.generation
+  // Allocate the 4-breakpoint budget in invalidation order: tools → system →
+  // messages. Tools live highest in the cache hierarchy, so when callers
+  // over-mark we keep their tool hints and shed the message-tail ones first.
+  const breakpoints = Cache.newBreakpoints(ANTHROPIC_BREAKPOINT_CAP)
+  const tools =
+    request.tools.length === 0 || request.toolChoice?.type === "none"
+      ? undefined
+      : request.tools.map((tool) => lowerTool(breakpoints, tool))
+  const system =
+    request.system.length === 0
+      ? undefined
+      : request.system.map((part) => ({
+          type: "text" as const,
+          text: part.text,
+          cache_control: cacheControl(breakpoints, part.cache),
+        }))
+  const messages = yield* lowerMessages(request, breakpoints)
+  if (breakpoints.dropped > 0) {
+    yield* Effect.logWarning(
+      `Anthropic Messages: dropped ${breakpoints.dropped} cache breakpoint(s); the API allows at most ${ANTHROPIC_BREAKPOINT_CAP} per request.`,
+    )
+  }
  return {
    model: request.model.id,
-    system:
-      request.system.length === 0
-        ? undefined
-        : request.system.map((part) => ({
-            type: "text" as const,
-            text: part.text,
-            cache_control: cacheControl(part.cache),
-          })),
-    messages: yield* lowerMessages(request),
-    tools: request.tools.length === 0 || request.toolChoice?.type === "none" ? undefined : request.tools.map(lowerTool),
+    system,
+    messages,
+    tools,
    tool_choice: toolChoice,
    stream: true as const,
    max_tokens: generation?.maxTokens ?? request.model.limits.output ?? 4096,
@@ -364,34 +406,56 @@ const mapFinishReason = (reason: string | null | undefined): FinishReason => {
  return "unknown"
 }

+// Anthropic reports the non-overlapping breakdown natively — its
+// `input_tokens` is the *non-cached* count per the Messages API docs, with
+// cache reads and writes as separate fields. We sum them to derive the
+// inclusive `inputTokens` the rest of the contract expects. Extended
+// thinking tokens are *not* broken out by Anthropic — they're billed as
+// part of `output_tokens`, so `reasoningTokens` stays `undefined` and
+// `outputTokens` carries the combined total.
 const mapUsage = (usage: AnthropicUsage | undefined): Usage | undefined => {
  if (!usage) return undefined
+  const nonCached = usage.input_tokens
+  const cacheRead = usage.cache_read_input_tokens ?? undefined
+  const cacheWrite = usage.cache_creation_input_tokens ?? undefined
+  const inputTokens = ProviderShared.sumTokens(nonCached, cacheRead, cacheWrite)
  return new Usage({
-    inputTokens: usage.input_tokens,
+    inputTokens,
    outputTokens: usage.output_tokens,
-    cacheReadInputTokens: usage.cache_read_input_tokens ?? undefined,
-    cacheWriteInputTokens: usage.cache_creation_input_tokens ?? undefined,
-    totalTokens: ProviderShared.totalTokens(usage.input_tokens, usage.output_tokens, undefined),
-    native: usage,
+    nonCachedInputTokens: nonCached,
+    cacheReadInputTokens: cacheRead,
+    cacheWriteInputTokens: cacheWrite,
+    totalTokens: ProviderShared.totalTokens(inputTokens, usage.output_tokens, undefined),
+    providerMetadata: { anthropic: usage },
  })
 }

 // Anthropic emits usage on `message_start` and again on `message_delta` — the
 // final delta carries the authoritative totals. Right-biased merge: each
-// field prefers `right` when defined, falls back to `left`. `totalTokens` is
-// recomputed from the merged input/output to stay consistent.
+// field prefers `right` when defined, falls back to `left`. `inputTokens` is
+// recomputed from the merged breakdown so the inclusive total stays
+// consistent with `nonCached + cacheRead + cacheWrite`.
 const mergeUsage = (left: Usage | undefined, right: Usage | undefined) => {
  if (!left) return right
  if (!right) return left
-  const inputTokens = right.inputTokens ?? left.inputTokens
+  const nonCachedInputTokens = right.nonCachedInputTokens ?? left.nonCachedInputTokens
+  const cacheReadInputTokens = right.cacheReadInputTokens ?? left.cacheReadInputTokens
+  const cacheWriteInputTokens = right.cacheWriteInputTokens ?? left.cacheWriteInputTokens
+  const inputTokens = ProviderShared.sumTokens(nonCachedInputTokens, cacheReadInputTokens, cacheWriteInputTokens)
  const outputTokens = right.outputTokens ?? left.outputTokens
  return new Usage({
    inputTokens,
    outputTokens,
-    cacheReadInputTokens: right.cacheReadInputTokens ?? left.cacheReadInputTokens,
-    cacheWriteInputTokens: right.cacheWriteInputTokens ?? left.cacheWriteInputTokens,
+    nonCachedInputTokens,
+    cacheReadInputTokens,
+    cacheWriteInputTokens,
    totalTokens: ProviderShared.totalTokens(inputTokens, outputTokens, undefined),
-    native: { ...left.native, ...right.native },
+    providerMetadata: {
+      anthropic: {
+        ...(left.providerMetadata?.["anthropic"] ?? {}),
+        ...(right.providerMetadata?.["anthropic"] ?? {}),
+      },
+    },
  })
 }

@@ -415,14 +479,13 @@ const serverToolResultEvent = (block: NonNullable<AnthropicEvent["content_block"
      ? String((block.content as Record<string, unknown>).type)
      : ""
  const isError = errorPayload.endsWith("_tool_result_error")
-  return {
-    type: "tool-result",
+  return LLMEvent.toolResult({
    id: block.tool_use_id ?? "",
    name: SERVER_TOOL_RESULT_NAMES[block.type],
    result: isError ? { type: "error", value: block.content } : { type: "json", value: block.content },
    providerExecuted: true,
    providerMetadata: anthropicMetadata({ blockType: block.type }),
-  }
+  })
 }

 type StepResult = readonly [ParserState, ReadonlyArray<LLMEvent>]
@@ -439,38 +502,45 @@ const onContentBlockStart = (state: ParserState, event: AnthropicEvent): StepRes
  if (!block) return [state, NO_EVENTS]

  if ((block.type === "tool_use" || block.type === "server_tool_use") && event.index !== undefined) {
+    const events: LLMEvent[] = []
+    const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
    return [
      {
        ...state,
+        lifecycle,
        tools: ToolStream.start(state.tools, event.index, {
          id: block.id ?? String(event.index),
          name: block.name ?? "",
          providerExecuted: block.type === "server_tool_use",
        }),
      },
-      NO_EVENTS,
+      [...events, LLMEvent.toolInputStart({ id: block.id ?? String(event.index), name: block.name ?? "" })],
    ]
  }

  if (block.type === "text" && block.text) {
-    return [state, [{ type: "text-delta", text: block.text }]]
+    const events: LLMEvent[] = []
+    return [
+      { ...state, lifecycle: Lifecycle.textDelta(state.lifecycle, events, `text-${event.index ?? 0}`, block.text) },
+      events,
+    ]
  }

  if (block.type === "thinking" && block.thinking) {
+    const events: LLMEvent[] = []
    return [
-      state,
-      [
-        {
-          type: "reasoning-delta",
-          text: block.thinking,
-          ...(block.signature ? { providerMetadata: anthropicMetadata({ signature: block.signature }) } : {}),
-        },
-      ],
+      {
+        ...state,
+        lifecycle: Lifecycle.reasoningDelta(state.lifecycle, events, `reasoning-${event.index ?? 0}`, block.thinking),
+      },
+      events,
    ]
  }

  const result = serverToolResultEvent(block)
-  return [state, result ? [result] : NO_EVENTS]
+  if (!result) return [state, NO_EVENTS]
+  const events: LLMEvent[] = []
+  return [{ ...state, lifecycle: Lifecycle.stepStart(state.lifecycle, events) }, [...events, result]]
 }

 const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(function* (
@@ -480,17 +550,37 @@ const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(f
  const delta = event.delta

  if (delta?.type === "text_delta" && delta.text) {
-    return [state, [{ type: "text-delta", text: delta.text }]] satisfies StepResult
+    const events: LLMEvent[] = []
+    return [
+      { ...state, lifecycle: Lifecycle.textDelta(state.lifecycle, events, `text-${event.index ?? 0}`, delta.text) },
+      events,
+    ] satisfies StepResult
  }

  if (delta?.type === "thinking_delta" && delta.thinking) {
-    return [state, [{ type: "reasoning-delta", text: delta.thinking }]] satisfies StepResult
+    const events: LLMEvent[] = []
+    return [
+      {
+        ...state,
+        lifecycle: Lifecycle.reasoningDelta(state.lifecycle, events, `reasoning-${event.index ?? 0}`, delta.thinking),
+      },
+      events,
+    ] satisfies StepResult
  }

  if (delta?.type === "signature_delta" && delta.signature) {
+    const events: LLMEvent[] = []
    return [
-      state,
-      [{ type: "reasoning-delta", text: "", providerMetadata: anthropicMetadata({ signature: delta.signature }) }],
+      {
+        ...state,
+        lifecycle: Lifecycle.reasoningEnd(
+          state.lifecycle,
+          events,
+          `reasoning-${event.index ?? 0}`,
+          anthropicMetadata({ signature: delta.signature }),
+        ),
+      },
+      events,
    ] satisfies StepResult
  }

@@ -504,7 +594,10 @@ const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(f
      "Anthropic Messages tool argument delta is missing its tool call",
    )
    if (ToolStream.isError(result)) return yield* result
-    return [{ ...state, tools: result.tools }, result.event ? [result.event] : NO_EVENTS] satisfies StepResult
+    const events: LLMEvent[] = []
+    const lifecycle = result.events.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
+    events.push(...result.events)
+    return [{ ...state, lifecycle, tools: result.tools }, events] satisfies StepResult
  }

  return [state, NO_EVENTS] satisfies StepResult
@@ -516,29 +609,35 @@ const onContentBlockStop = Effect.fn("AnthropicMessages.onContentBlockStop")(fun
 ) {
  if (event.index === undefined) return [state, NO_EVENTS] satisfies StepResult
  const result = yield* ToolStream.finish(ADAPTER, state.tools, event.index)
-  return [{ ...state, tools: result.tools }, result.event ? [result.event] : NO_EVENTS] satisfies StepResult
+  const events: LLMEvent[] = []
+  const resultEvents = result.events ?? []
+  const lifecycle = resultEvents.length
+    ? Lifecycle.stepStart(state.lifecycle, events)
+    : Lifecycle.reasoningEnd(
+        Lifecycle.textEnd(state.lifecycle, events, `text-${event.index}`),
+        events,
+        `reasoning-${event.index}`,
+      )
+  events.push(...resultEvents)
+  return [{ ...state, lifecycle, tools: result.tools }, events] satisfies StepResult
 })

 const onMessageDelta = (state: ParserState, event: AnthropicEvent): StepResult => {
  const usage = mergeUsage(state.usage, mapUsage(event.usage))
-  return [
-    { ...state, usage },
-    [
-      {
-        type: "request-finish",
-        reason: mapFinishReason(event.delta?.stop_reason),
-        usage,
-        ...(event.delta?.stop_sequence
-          ? { providerMetadata: anthropicMetadata({ stopSequence: event.delta.stop_sequence }) }
-          : {}),
-      },
-    ],
-  ]
+  const events: LLMEvent[] = []
+  const lifecycle = Lifecycle.finish(state.lifecycle, events, {
+    reason: mapFinishReason(event.delta?.stop_reason),
+    usage,
+    providerMetadata: event.delta?.stop_sequence
+      ? anthropicMetadata({ stopSequence: event.delta.stop_sequence })
+      : undefined,
+  })
+  return [{ ...state, lifecycle, usage }, events]
 }

 const onError = (state: ParserState, event: AnthropicEvent): StepResult => [
  state,
-  [{ type: "provider-error", message: event.error?.message ?? "Anthropic Messages stream error" }],
+  [LLMEvent.providerError({ message: event.error?.message ?? "Anthropic Messages stream error" })],
 ]

 const step = (state: ParserState, event: AnthropicEvent) => {
@@ -567,7 +666,7 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: Protocol.jsonEvent(AnthropicEvent),
-    initial: () => ({ tools: ToolStream.empty<number>() }),
+    initial: () => ({ tools: ToolStream.empty<number>(), lifecycle: Lifecycle.initial() }),
    step,
  },
 })
@@ -3,10 +3,10 @@ import { Route, type RouteModelInput } from "../route/client"
 import { Endpoint } from "../route/endpoint"
 import { Protocol } from "../route/protocol"
 import {
+  LLMEvent,
  Usage,
  type CacheHint,
  type FinishReason,
-  type LLMEvent,
  type LLMRequest,
  type ToolCallPart,
  type ToolDefinition,
@@ -17,6 +17,7 @@ import { JsonObject, optionalArray, ProviderShared } from "./shared"
 import { BedrockAuth, type Credentials as BedrockCredentials } from "./utils/bedrock-auth"
 import { BedrockCache } from "./utils/bedrock-cache"
 import { BedrockMedia } from "./utils/bedrock-media"
+import { Lifecycle } from "./utils/lifecycle"
 import { ToolStream } from "./utils/tool-stream"

 const ADAPTER = "bedrock-converse"
@@ -108,7 +109,7 @@ type BedrockMessage = Schema.Schema.Type<typeof BedrockMessage>
 const BedrockSystemBlock = Schema.Union([BedrockTextBlock, BedrockCache.CachePointBlock])
 type BedrockSystemBlock = Schema.Schema.Type<typeof BedrockSystemBlock>

-const BedrockTool = Schema.Struct({
+const BedrockToolSpec = Schema.Struct({
  toolSpec: Schema.Struct({
    name: Schema.String,
    description: Schema.String,
@@ -117,6 +118,9 @@ const BedrockTool = Schema.Struct({
    }),
  }),
 })
+type BedrockToolSpec = Schema.Schema.Type<typeof BedrockToolSpec>
+
+const BedrockTool = Schema.Union([BedrockToolSpec, BedrockCache.CachePointBlock])
 type BedrockTool = Schema.Schema.Type<typeof BedrockTool>

 const BedrockToolChoice = Schema.Union([
@@ -214,7 +218,7 @@ type BedrockEvent = Schema.Schema.Type<typeof BedrockEvent>
 // =============================================================================
 // Request Lowering
 // =============================================================================
-const lowerTool = (tool: ToolDefinition): BedrockTool => ({
+const lowerToolSpec = (tool: ToolDefinition): BedrockToolSpec => ({
  toolSpec: {
    name: tool.name,
    description: tool.description,
@@ -222,11 +226,22 @@ const lowerTool = (tool: ToolDefinition): BedrockTool => ({
  },
 })

+const lowerTools = (breakpoints: BedrockCache.Breakpoints, tools: ReadonlyArray<ToolDefinition>): BedrockTool[] => {
+  const result: BedrockTool[] = []
+  for (const tool of tools) {
+    result.push(lowerToolSpec(tool))
+    const cachePoint = BedrockCache.block(breakpoints, tool.cache)
+    if (cachePoint) result.push(cachePoint)
+  }
+  return result
+}
+
 const textWithCache = (
+  breakpoints: BedrockCache.Breakpoints,
  text: string,
  cache: CacheHint | undefined,
 ): Array<BedrockTextBlock | BedrockCache.CachePointBlock> => {
-  const cachePoint = BedrockCache.block(cache)
+  const cachePoint = BedrockCache.block(breakpoints, cache)
  return cachePoint ? [{ text }, cachePoint] : [{ text }]
 }

@@ -257,7 +272,10 @@ const lowerToolResult = (part: ToolResultPart): BedrockToolResultBlock => ({
  },
 })

-const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (request: LLMRequest) {
+const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
+  request: LLMRequest,
+  breakpoints: BedrockCache.Breakpoints,
+) {
  const messages: BedrockMessage[] = []

  for (const message of request.messages) {
@@ -267,7 +285,7 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (requ
        if (!ProviderShared.supportsContent(part, ["text", "media"]))
          return yield* ProviderShared.unsupportedContent("Bedrock Converse", "user", ["text", "media"])
        if (part.type === "text") {
-          content.push(...textWithCache(part.text, part.cache))
+          content.push(...textWithCache(breakpoints, part.text, part.cache))
          continue
        }
        if (part.type === "media") {
@@ -289,7 +307,7 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (requ
            "tool-call",
          ])
        if (part.type === "text") {
-          content.push(...textWithCache(part.text, part.cache))
+          content.push(...textWithCache(breakpoints, part.text, part.cache))
          continue
        }
        if (part.type === "reasoning") {
@@ -309,11 +327,13 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (requ
      continue
    }

-    const content: BedrockToolResultBlock[] = []
+    const content: BedrockUserBlock[] = []
    for (const part of message.content) {
      if (!ProviderShared.supportsContent(part, ["tool-result"]))
        return yield* ProviderShared.unsupportedContent("Bedrock Converse", "tool", ["tool-result"])
      content.push(lowerToolResult(part))
+      const cachePoint = BedrockCache.block(breakpoints, part.cache)
+      if (cachePoint) content.push(cachePoint)
    }
    messages.push({ role: "user", content })
  }
@@ -323,16 +343,32 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (requ

 // System prompts share the cache-point convention: emit the text block, then
 // optionally a positional `cachePoint` marker.
-const lowerSystem = (system: ReadonlyArray<LLMRequest["system"][number]>): BedrockSystemBlock[] =>
-  system.flatMap((part) => textWithCache(part.text, part.cache))
+const lowerSystem = (
+  breakpoints: BedrockCache.Breakpoints,
+  system: ReadonlyArray<LLMRequest["system"][number]>,
+): BedrockSystemBlock[] => system.flatMap((part) => textWithCache(breakpoints, part.text, part.cache))

 const fromRequest = Effect.fn("BedrockConverse.fromRequest")(function* (request: LLMRequest) {
  const toolChoice = request.toolChoice ? yield* lowerToolChoice(request.toolChoice) : undefined
  const generation = request.generation
+  // Bedrock-Claude shares Anthropic's 4-breakpoint cap. Spend the budget in
+  // tools → system → messages order to favour the highest-impact prefixes.
+  const breakpoints = BedrockCache.breakpoints()
+  const toolConfig =
+    request.tools.length > 0 && request.toolChoice?.type !== "none"
+      ? { tools: lowerTools(breakpoints, request.tools), toolChoice }
+      : undefined
+  const system = request.system.length === 0 ? undefined : lowerSystem(breakpoints, request.system)
+  const messages = yield* lowerMessages(request, breakpoints)
+  if (breakpoints.dropped > 0) {
+    yield* Effect.logWarning(
+      `Bedrock Converse: dropped ${breakpoints.dropped} cache breakpoint(s); the API allows at most ${BedrockCache.BEDROCK_BREAKPOINT_CAP} per request.`,
+    )
+  }
  return {
    modelId: request.model.id,
-    messages: yield* lowerMessages(request),
-    system: request.system.length === 0 ? undefined : lowerSystem(request.system),
+    messages,
+    system,
    inferenceConfig:
      generation?.maxTokens === undefined &&
      generation?.temperature === undefined &&
@@ -345,10 +381,7 @@ const fromRequest = Effect.fn("BedrockConverse.fromRequest")(function* (request:
            topP: generation?.topP,
            stopSequences: generation?.stop,
          },
-    toolConfig:
-      request.tools.length > 0 && request.toolChoice?.type !== "none"
-        ? { tools: request.tools.map(lowerTool), toolChoice }
-        : undefined,
+    toolConfig,
  }
 })

@@ -363,15 +396,22 @@ const mapFinishReason = (reason: string): FinishReason => {
  return "unknown"
 }

+// AWS Bedrock Converse reports `inputTokens` (inclusive total) with
+// `cacheReadInputTokens` and `cacheWriteInputTokens` as subsets. Pass
+// the total through and derive the non-cached breakdown. Bedrock does
+// not break reasoning out of `outputTokens` for any current model.
 const mapUsage = (usage: BedrockUsageSchema | undefined): Usage | undefined => {
  if (!usage) return undefined
+  const cacheTotal = (usage.cacheReadInputTokens ?? 0) + (usage.cacheWriteInputTokens ?? 0)
+  const nonCached = ProviderShared.subtractTokens(usage.inputTokens, cacheTotal)
  return new Usage({
    inputTokens: usage.inputTokens,
    outputTokens: usage.outputTokens,
-    totalTokens: ProviderShared.totalTokens(usage.inputTokens, usage.outputTokens, usage.totalTokens),
+    nonCachedInputTokens: nonCached,
    cacheReadInputTokens: usage.cacheReadInputTokens,
    cacheWriteInputTokens: usage.cacheWriteInputTokens,
-    native: usage,
+    totalTokens: ProviderShared.totalTokens(usage.inputTokens, usage.outputTokens, usage.totalTokens),
+    providerMetadata: { bedrock: usage },
  })
 }

@@ -381,32 +421,64 @@ interface ParserState {
  // `metadata` (carries usage). Hold the terminal event in state so `onHalt`
  // can emit exactly one finish after both chunks have had a chance to arrive.
  readonly pendingFinish: { readonly reason: FinishReason; readonly usage?: Usage } | undefined
+  readonly hasToolCalls: boolean
+  readonly lifecycle: Lifecycle.State
 }

 const step = (state: ParserState, event: BedrockEvent) =>
  Effect.gen(function* () {
    if (event.contentBlockStart?.start?.toolUse) {
      const index = event.contentBlockStart.contentBlockIndex
+      const events: LLMEvent[] = []
+      const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
      return [
        {
          ...state,
+          lifecycle,
          tools: ToolStream.start(state.tools, index, {
            id: event.contentBlockStart.start.toolUse.toolUseId,
            name: event.contentBlockStart.start.toolUse.name,
          }),
        },
-        [],
+        [
+          ...events,
+          LLMEvent.toolInputStart({
+            id: event.contentBlockStart.start.toolUse.toolUseId,
+            name: event.contentBlockStart.start.toolUse.name,
+          }),
+        ],
      ] as const
    }

    if (event.contentBlockDelta?.delta?.text) {
-      return [state, [{ type: "text-delta" as const, text: event.contentBlockDelta.delta.text }]] as const
+      const events: LLMEvent[] = []
+      return [
+        {
+          ...state,
+          lifecycle: Lifecycle.textDelta(
+            state.lifecycle,
+            events,
+            `text-${event.contentBlockDelta.contentBlockIndex}`,
+            event.contentBlockDelta.delta.text,
+          ),
+        },
+        events,
+      ] as const
    }

    if (event.contentBlockDelta?.delta?.reasoningContent?.text) {
+      const events: LLMEvent[] = []
      return [
-        state,
-        [{ type: "reasoning-delta" as const, text: event.contentBlockDelta.delta.reasoningContent.text }],
+        {
+          ...state,
+          lifecycle: Lifecycle.reasoningDelta(
+            state.lifecycle,
+            events,
+            `reasoning-${event.contentBlockDelta.contentBlockIndex}`,
+            event.contentBlockDelta.delta.reasoningContent.text,
+          ),
+        },
+        events,
      ] as const
    }

@@ -420,12 +492,33 @@ const step = (state: ParserState, event: BedrockEvent) =>
        "Bedrock Converse tool delta is missing its tool call",
      )
      if (ToolStream.isError(result)) return yield* result
-      return [{ ...state, tools: result.tools }, result.event ? [result.event] : []] as const
+      const events: LLMEvent[] = []
+      const lifecycle = result.events.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
+      events.push(...result.events)
+      return [{ ...state, lifecycle, tools: result.tools }, events] as const
    }

    if (event.contentBlockStop) {
      const result = yield* ToolStream.finish(ADAPTER, state.tools, event.contentBlockStop.contentBlockIndex)
-      return [{ ...state, tools: result.tools }, result.event ? [result.event] : []] as const
+      const events: LLMEvent[] = []
+      const resultEvents = result.events ?? []
+      const lifecycle = resultEvents.length
+        ? Lifecycle.stepStart(state.lifecycle, events)
+        : Lifecycle.reasoningEnd(
+            Lifecycle.textEnd(state.lifecycle, events, `text-${event.contentBlockStop.contentBlockIndex}`),
+            events,
+            `reasoning-${event.contentBlockStop.contentBlockIndex}`,
+          )
+      events.push(...resultEvents)
+      return [
+        {
+          ...state,
+          hasToolCalls: resultEvents.some(LLMEvent.is.toolCall) ? true : state.hasToolCalls,
+          lifecycle,
+          tools: result.tools,
+        },
+        events,
+      ] as const
    }

    if (event.messageStop) {
@@ -449,16 +542,13 @@ const step = (state: ParserState, event: BedrockEvent) =>
        event.modelStreamErrorException?.message ??
        event.serviceUnavailableException?.message ??
        "Bedrock Converse stream error"
-      return [state, [{ type: "provider-error" as const, message, retryable: true }]] as const
+      return [state, [LLMEvent.providerError({ message, retryable: true })]] as const
    }

    if (event.validationException || event.throttlingException) {
      const message =
        event.validationException?.message ?? event.throttlingException?.message ?? "Bedrock Converse error"
-      return [
-        state,
-        [{ type: "provider-error" as const, message, retryable: event.throttlingException !== undefined }],
-      ] as const
+      return [state, [LLMEvent.providerError({ message, retryable: event.throttlingException !== undefined })]] as const
    }

    return [state, []] as const
@@ -468,7 +558,15 @@ const framing = BedrockEventStream.framing(ADAPTER)

 const onHalt = (state: ParserState): ReadonlyArray<LLMEvent> =>
  state.pendingFinish
-    ? [{ type: "request-finish", reason: state.pendingFinish.reason, usage: state.pendingFinish.usage }]
+    ? (() => {
+        const events: LLMEvent[] = []
+        Lifecycle.finish(state.lifecycle, events, {
+          reason:
+            state.pendingFinish.reason === "stop" && state.hasToolCalls ? "tool-calls" : state.pendingFinish.reason,
+          usage: state.pendingFinish.usage,
+        })
+        return events
+      })()
    : []

 // =============================================================================
@@ -486,7 +584,12 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: BedrockEvent,
-    initial: () => ({ tools: ToolStream.empty<number>(), pendingFinish: undefined }),
+    initial: () => ({
+      tools: ToolStream.empty<number>(),
+      pendingFinish: undefined,
+      hasToolCalls: false,
+      lifecycle: Lifecycle.initial(),
+    }),
    step,
    onHalt,
  },
@@ -5,9 +5,9 @@ import { Endpoint } from "../route/endpoint"
 import { Framing } from "../route/framing"
 import { Protocol } from "../route/protocol"
 import {
+  LLMEvent,
  Usage,
  type FinishReason,
-  type LLMEvent,
  type LLMRequest,
  type MediaPart,
  type TextPart,
@@ -16,6 +16,7 @@ import {
 } from "../schema"
 import { JsonObject, optionalArray, ProviderShared } from "./shared"
 import { GeminiToolSchema } from "./utils/gemini-tool-schema"
+import { Lifecycle } from "./utils/lifecycle"

 const ADAPTER = "gemini"
 export const DEFAULT_BASE_URL = "https://generativelanguage.googleapis.com/v1beta"
@@ -134,10 +135,9 @@ interface ParserState {
  readonly hasToolCalls: boolean
  readonly nextToolCallId: number
  readonly usage?: Usage
+  readonly lifecycle: Lifecycle.State
 }

-const invalid = ProviderShared.invalidRequest
-
 const mediaData = ProviderShared.mediaBytes

 // =============================================================================
@@ -281,15 +281,28 @@ const fromRequest = Effect.fn("Gemini.fromRequest")(function* (request: LLMReque
 // =============================================================================
 // Stream Parsing
 // =============================================================================
+// Gemini reports `promptTokenCount` (inclusive total) with a
+// `cachedContentTokenCount` subset. `candidatesTokenCount` is *exclusive*
+// of `thoughtsTokenCount` — visible-only, not a total — so we sum the two
+// to produce the inclusive `outputTokens` the rest of the contract expects.
 const mapUsage = (usage: GeminiUsage | undefined) => {
  if (!usage) return undefined
+  const cached = usage.cachedContentTokenCount
+  const nonCached = ProviderShared.subtractTokens(usage.promptTokenCount, cached)
+  // `candidatesTokenCount` is visible-only; sum with thoughts to produce the
+  // inclusive `outputTokens` the contract expects. Only compute the total
+  // when the visible component is reported — otherwise we'd fabricate an
+  // inclusive number from a partial breakdown.
+  const outputTokens =
+    usage.candidatesTokenCount !== undefined ? usage.candidatesTokenCount + (usage.thoughtsTokenCount ?? 0) : undefined
  return new Usage({
    inputTokens: usage.promptTokenCount,
-    outputTokens: usage.candidatesTokenCount,
+    outputTokens,
+    nonCachedInputTokens: nonCached,
+    cacheReadInputTokens: cached,
    reasoningTokens: usage.thoughtsTokenCount,
-    cacheReadInputTokens: usage.cachedContentTokenCount,
-    totalTokens: ProviderShared.totalTokens(usage.promptTokenCount, usage.candidatesTokenCount, usage.totalTokenCount),
-    native: usage,
+    totalTokens: ProviderShared.totalTokens(usage.promptTokenCount, outputTokens, usage.totalTokenCount),
+    providerMetadata: { google: usage },
  })
 }

@@ -311,7 +324,14 @@ const mapFinishReason = (finishReason: string | undefined, hasToolCalls: boolean

 const finish = (state: ParserState): ReadonlyArray<LLMEvent> =>
  state.finishReason || state.usage
-    ? [{ type: "request-finish", reason: mapFinishReason(state.finishReason, state.hasToolCalls), usage: state.usage }]
+    ? (() => {
+        const events: LLMEvent[] = []
+        Lifecycle.finish(state.lifecycle, events, {
+          reason: mapFinishReason(state.finishReason, state.hasToolCalls),
+          usage: state.usage,
+        })
+        return events
+      })()
    : []

 const step = (state: ParserState, event: GeminiEvent) => {
@@ -328,18 +348,22 @@ const step = (state: ParserState, event: GeminiEvent) => {

  const events: LLMEvent[] = []
  let hasToolCalls = nextState.hasToolCalls
+  let lifecycle = nextState.lifecycle
  let nextToolCallId = nextState.nextToolCallId

  for (const part of candidate.content.parts) {
    if ("text" in part && part.text.length > 0) {
-      events.push({ type: part.thought ? "reasoning-delta" : "text-delta", text: part.text })
+      lifecycle = part.thought
+        ? Lifecycle.reasoningDelta(lifecycle, events, "reasoning-0", part.text)
+        : Lifecycle.textDelta(lifecycle, events, "text-0", part.text)
      continue
    }

    if ("functionCall" in part) {
      const input = part.functionCall.args
      const id = `tool_${nextToolCallId++}`
-      events.push({ type: "tool-call", id, name: part.functionCall.name, input })
+      lifecycle = Lifecycle.stepStart(lifecycle, events)
+      events.push(LLMEvent.toolCall({ id, name: part.functionCall.name, input }))
      hasToolCalls = true
    }
  }
@@ -348,6 +372,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
    {
      ...nextState,
      hasToolCalls,
+      lifecycle,
      nextToolCallId,
      finishReason: candidate.finishReason ?? nextState.finishReason,
    },
@@ -371,7 +396,7 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: Protocol.jsonEvent(GeminiEvent),
-    initial: () => ({ hasToolCalls: false, nextToolCallId: 0 }),
+    initial: () => ({ hasToolCalls: false, nextToolCallId: 0, lifecycle: Lifecycle.initial() }),
    step,
    onHalt: finish,
  },
@@ -6,9 +6,9 @@ import { Framing } from "../route/framing"
 import { HttpTransport } from "../route/transport"
 import { Protocol } from "../route/protocol"
 import {
+  LLMEvent,
  Usage,
  type FinishReason,
-  type LLMEvent,
  type LLMRequest,
  type TextPart,
  type ToolCallPart,
@@ -16,6 +16,7 @@ import {
 } from "../schema"
 import { isRecord, JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared"
 import { OpenAIOptions } from "./utils/openai-options"
+import { Lifecycle } from "./utils/lifecycle"
 import { ToolStream } from "./utils/tool-stream"

 const ADAPTER = "openai-chat"
@@ -147,6 +148,7 @@ interface ParserState {
  readonly toolCallEvents: ReadonlyArray<LLMEvent>
  readonly usage?: Usage
  readonly finishReason?: FinishReason
+  readonly lifecycle: Lifecycle.State
 }

 const invalid = ProviderShared.invalidRequest
@@ -290,15 +292,24 @@ const mapFinishReason = (reason: string | null | undefined): FinishReason => {
  return "unknown"
 }

+// OpenAI Chat reports `prompt_tokens` (inclusive total) with a
+// `cached_tokens` subset, and `completion_tokens` (inclusive total) with
+// a `reasoning_tokens` subset. We pass the inclusive totals through and
+// derive the non-cached breakdown so the `LLM.Usage` contract is
+// satisfied on both sides.
 const mapUsage = (usage: OpenAIChatEvent["usage"]): Usage | undefined => {
  if (!usage) return undefined
+  const cached = usage.prompt_tokens_details?.cached_tokens
+  const reasoning = usage.completion_tokens_details?.reasoning_tokens
+  const nonCached = ProviderShared.subtractTokens(usage.prompt_tokens, cached)
  return new Usage({
    inputTokens: usage.prompt_tokens,
    outputTokens: usage.completion_tokens,
-    reasoningTokens: usage.completion_tokens_details?.reasoning_tokens,
-    cacheReadInputTokens: usage.prompt_tokens_details?.cached_tokens,
+    nonCachedInputTokens: nonCached,
+    cacheReadInputTokens: cached,
+    reasoningTokens: reasoning,
    totalTokens: ProviderShared.totalTokens(usage.prompt_tokens, usage.completion_tokens, usage.total_tokens),
-    native: usage,
+    providerMetadata: { openai: usage },
  })
 }

@@ -312,7 +323,9 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
    const toolDeltas = delta?.tool_calls ?? []
    let tools = state.tools

-    if (delta?.content) events.push({ type: "text-delta", text: delta.content })
+    let lifecycle = state.lifecycle
+
+    if (delta?.content) lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.content)

    for (const tool of toolDeltas) {
      const result = ToolStream.appendOrStart(
@@ -324,7 +337,8 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
      )
      if (ToolStream.isError(result)) return yield* result
      tools = result.tools
-      if (result.event) events.push(result.event)
+      if (result.events.length) lifecycle = Lifecycle.stepStart(lifecycle, events)
+      events.push(...result.events)
    }

    // Finalize accumulated tool inputs eagerly when finish_reason arrives so
@@ -340,18 +354,20 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
        toolCallEvents: finished?.events ?? state.toolCallEvents,
        usage,
        finishReason,
+        lifecycle,
      },
      events,
    ] as const
  })

 const finishEvents = (state: ParserState): ReadonlyArray<LLMEvent> => {
+  const events: LLMEvent[] = []
  const hasToolCalls = state.toolCallEvents.length > 0
  const reason = state.finishReason === "stop" && hasToolCalls ? "tool-calls" : state.finishReason
-  return [
-    ...state.toolCallEvents,
-    ...(reason ? ([{ type: "request-finish", reason, usage: state.usage }] satisfies ReadonlyArray<LLMEvent>) : []),
-  ]
+  const lifecycle = state.toolCallEvents.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
+  events.push(...state.toolCallEvents)
+  if (reason) Lifecycle.finish(lifecycle, events, { reason, usage: state.usage })
+  return events
 }

 // =============================================================================
@@ -371,7 +387,7 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: Protocol.jsonEvent(OpenAIChatEvent),
-    initial: () => ({ tools: ToolStream.empty<number>(), toolCallEvents: [] }),
+    initial: () => ({ tools: ToolStream.empty<number>(), toolCallEvents: [], lifecycle: Lifecycle.initial() }),
    step,
    onHalt: finishEvents,
  },
@@ -6,9 +6,9 @@ import { Framing } from "../route/framing"
 import { HttpTransport, WebSocketTransport } from "../route/transport"
 import { Protocol } from "../route/protocol"
 import {
+  LLMEvent,
  Usage,
  type FinishReason,
-  type LLMEvent,
  type LLMRequest,
  type ProviderMetadata,
  type TextPart,
@@ -17,6 +17,7 @@ import {
 } from "../schema"
 import { JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared"
 import { OpenAIOptions } from "./utils/openai-options"
+import { Lifecycle } from "./utils/lifecycle"
 import { ToolStream } from "./utils/tool-stream"

 const ADAPTER = "openai-responses"
@@ -165,6 +166,7 @@ type OpenAIResponsesEvent = Schema.Schema.Type<typeof OpenAIResponsesEvent>
 interface ParserState {
  readonly tools: ToolStream.State<string>
  readonly hasFunctionCall: boolean
+  readonly lifecycle: Lifecycle.State
 }

 const invalid = ProviderShared.invalidRequest
@@ -276,15 +278,23 @@ const fromRequest = Effect.fn("OpenAIResponses.fromRequest")(function* (request:
 // =============================================================================
 // Stream Parsing
 // =============================================================================
+// OpenAI Responses reports `input_tokens` (inclusive total) with a
+// `cached_tokens` subset, and `output_tokens` (inclusive total) with a
+// `reasoning_tokens` subset. Pass the totals through and derive the
+// non-cached breakdown.
 const mapUsage = (usage: OpenAIResponsesUsage | null | undefined) => {
  if (!usage) return undefined
+  const cached = usage.input_tokens_details?.cached_tokens
+  const reasoning = usage.output_tokens_details?.reasoning_tokens
+  const nonCached = ProviderShared.subtractTokens(usage.input_tokens, cached)
  return new Usage({
    inputTokens: usage.input_tokens,
    outputTokens: usage.output_tokens,
-    reasoningTokens: usage.output_tokens_details?.reasoning_tokens,
-    cacheReadInputTokens: usage.input_tokens_details?.cached_tokens,
+    nonCachedInputTokens: nonCached,
+    cacheReadInputTokens: cached,
+    reasoningTokens: reasoning,
    totalTokens: ProviderShared.totalTokens(usage.input_tokens, usage.output_tokens, usage.total_tokens),
-    native: usage,
+    providerMetadata: { openai: usage },
  })
 }

@@ -348,22 +358,20 @@ const hostedToolEvents = (
  const tool = HOSTED_TOOLS[item.type]
  const providerMetadata = openaiMetadata({ itemId: item.id })
  return [
-    {
-      type: "tool-call",
+    LLMEvent.toolCall({
      id: item.id,
      name: tool.name,
      input: tool.input(item),
      providerExecuted: true,
      providerMetadata,
-    },
-    {
-      type: "tool-result",
+    }),
+    LLMEvent.toolResult({
      id: item.id,
      name: tool.name,
      result: hostedToolResult(item),
      providerExecuted: true,
      providerMetadata,
-    },
+    }),
  ]
 }

@@ -379,33 +387,32 @@ const TERMINAL_TYPES = new Set(["response.completed", "response.incomplete", "re

 const onOutputTextDelta = (state: ParserState, event: OpenAIResponsesEvent): StepResult => {
  if (!event.delta) return [state, NO_EVENTS]
+  const events: LLMEvent[] = []
  return [
-    state,
-    [
-      {
-        type: "text-delta",
-        id: event.item_id,
-        text: event.delta,
-        ...(event.item_id ? { providerMetadata: openaiMetadata({ itemId: event.item_id }) } : {}),
-      },
-    ],
+    { ...state, lifecycle: Lifecycle.textDelta(state.lifecycle, events, event.item_id ?? "text-0", event.delta) },
+    events,
  ]
 }

 const onOutputItemAdded = (state: ParserState, event: OpenAIResponsesEvent): StepResult => {
  const item = event.item
  if (item?.type !== "function_call" || !item.id) return [state, NO_EVENTS]
+  const providerMetadata = openaiMetadata({ itemId: item.id })
+  const events: LLMEvent[] = []
+  const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
  return [
    {
+      ...state,
+      lifecycle,
      hasFunctionCall: state.hasFunctionCall,
      tools: ToolStream.start(state.tools, item.id, {
        id: item.call_id ?? item.id,
        name: item.name ?? "",
        input: item.arguments ?? "",
-        providerMetadata: openaiMetadata({ itemId: item.id }),
+        providerMetadata,
      }),
    },
-    NO_EVENTS,
+    [...events, LLMEvent.toolInputStart({ id: item.call_id ?? item.id, name: item.name ?? "", providerMetadata })],
  ]
 }

@@ -422,10 +429,10 @@ const onFunctionCallArgumentsDelta = Effect.fn("OpenAIResponses.onFunctionCallAr
    "OpenAI Responses tool argument delta is missing its tool call",
  )
  if (ToolStream.isError(result)) return yield* result
-  return [
-    { hasFunctionCall: state.hasFunctionCall, tools: result.tools },
-    result.event ? [result.event] : NO_EVENTS,
-  ] satisfies StepResult
+  const events: LLMEvent[] = []
+  const lifecycle = result.events.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
+  events.push(...result.events)
+  return [{ ...state, lifecycle, tools: result.tools }, events] satisfies StepResult
 })

 const onOutputItemDone = Effect.fn("OpenAIResponses.onOutputItemDone")(function* (
@@ -444,44 +451,55 @@ const onOutputItemDone = Effect.fn("OpenAIResponses.onOutputItemDone")(function*
      item.arguments === undefined
        ? yield* ToolStream.finish(ADAPTER, tools, item.id)
        : yield* ToolStream.finishWithInput(ADAPTER, tools, item.id, item.arguments)
+    const events: LLMEvent[] = []
+    const resultEvents = result.events ?? []
+    const lifecycle = resultEvents.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
+    events.push(...resultEvents)
    return [
-      { hasFunctionCall: result.event ? true : state.hasFunctionCall, tools: result.tools },
-      result.event ? [result.event] : NO_EVENTS,
+      {
+        ...state,
+        lifecycle,
+        hasFunctionCall: resultEvents.some(LLMEvent.is.toolCall) ? true : state.hasFunctionCall,
+        tools: result.tools,
+      },
+      events,
    ] satisfies StepResult
  }

-  if (isHostedToolItem(item)) return [state, hostedToolEvents(item)] satisfies StepResult
+  if (isHostedToolItem(item)) {
+    const events: LLMEvent[] = []
+    const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
+    events.push(...hostedToolEvents(item))
+    return [{ ...state, lifecycle }, events] satisfies StepResult
+  }

  return [state, NO_EVENTS] satisfies StepResult
 })

-const onResponseFinish = (state: ParserState, event: OpenAIResponsesEvent): StepResult => [
-  state,
-  [
-    {
-      type: "request-finish",
-      reason: mapFinishReason(event, state.hasFunctionCall),
-      usage: mapUsage(event.response?.usage),
-      ...(event.response?.id || event.response?.service_tier
-        ? {
-            providerMetadata: openaiMetadata({
-              responseId: event.response.id,
-              serviceTier: event.response.service_tier,
-            }),
-          }
-        : {}),
-    },
-  ],
-]
+const onResponseFinish = (state: ParserState, event: OpenAIResponsesEvent): StepResult => {
+  const events: LLMEvent[] = []
+  const lifecycle = Lifecycle.finish(state.lifecycle, events, {
+    reason: mapFinishReason(event, state.hasFunctionCall),
+    usage: mapUsage(event.response?.usage),
+    providerMetadata:
+      event.response?.id || event.response?.service_tier
+        ? openaiMetadata({
+            responseId: event.response.id,
+            serviceTier: event.response.service_tier,
+          })
+        : undefined,
+  })
+  return [{ ...state, lifecycle }, events]
+}

 const onResponseFailed = (state: ParserState, event: OpenAIResponsesEvent): StepResult => [
  state,
-  [{ type: "provider-error", message: event.message ?? event.code ?? "OpenAI Responses response failed" }],
+  [LLMEvent.providerError({ message: event.message ?? event.code ?? "OpenAI Responses response failed" })],
 ]

 const onError = (state: ParserState, event: OpenAIResponsesEvent): StepResult => [
  state,
-  [{ type: "provider-error", message: event.message ?? event.code ?? "OpenAI Responses stream error" }],
+  [LLMEvent.providerError({ message: event.message ?? event.code ?? "OpenAI Responses stream error" })],
 ]

 const step = (state: ParserState, event: OpenAIResponsesEvent) => {
@@ -512,7 +530,7 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: Protocol.jsonEvent(OpenAIResponsesEvent),
-    initial: () => ({ hasFunctionCall: false, tools: ToolStream.empty<string>() }),
+    initial: () => ({ hasFunctionCall: false, tools: ToolStream.empty<string>(), lifecycle: Lifecycle.initial() }),
    step,
    terminal: (event) => TERMINAL_TYPES.has(event.type),
  },
@@ -42,6 +42,13 @@ export interface ToolAccumulator {
 * supplied total; otherwise falls back to `inputTokens + outputTokens` only
 * when at least one is defined. Returns `undefined` when neither input nor
 * output is known so routes don't publish a misleading `0`.
+ *
+ * Under the additive `LLM.Usage` contract, `inputTokens` and `outputTokens`
+ * are the non-cached input and visible output only. The provider-supplied
+ * `total` is the source of truth when present; the computed fallback
+ * under-counts cache and reasoning by design and exists mainly so
+ * Anthropic-style providers (which don't surface a total) still get a
+ * sensible aggregate on the input + output axes.
 */
 export const totalTokens = (
  inputTokens: number | undefined,
@@ -53,6 +60,35 @@ export const totalTokens = (
  return (inputTokens ?? 0) + (outputTokens ?? 0)
 }

+/**
+ * Subtract `subtrahend` from `total`, clamping to zero if the provider
+ * reports a non-sensical breakdown (e.g. `cached_tokens > prompt_tokens`).
+ * Used by protocol mappers when deriving a non-overlapping breakdown field
+ * from a provider's inclusive total — `nonCachedInputTokens` from
+ * `inputTokens - cacheReadInputTokens - cacheWriteInputTokens`.
+ *
+ * If `total` is `undefined`, returns `undefined` (we don't fabricate
+ * counts). If `subtrahend` is `undefined`, returns `total` unchanged. The
+ * provider-native breakdown stays available on `Usage.native` for debugging.
+ */
+export const subtractTokens = (total: number | undefined, subtrahend: number | undefined): number | undefined => {
+  if (total === undefined) return undefined
+  if (subtrahend === undefined) return total
+  return Math.max(0, total - subtrahend)
+}
+
+/**
+ * Sum a list of optional token counts, returning `undefined` only when
+ * every value is `undefined` (so we don't fabricate a `0`). Used by
+ * protocol mappers to derive the inclusive `inputTokens` total from a
+ * provider that natively reports a non-overlapping breakdown
+ * (e.g. Anthropic, whose `input_tokens` is already non-cached only).
+ */
+export const sumTokens = (...values: ReadonlyArray<number | undefined>): number | undefined => {
+  if (values.every((value) => value === undefined)) return undefined
+  return values.reduce<number>((acc, value) => acc + (value ?? 0), 0)
+}
+
 export const eventError = (route: string, message: string, raw?: string) =>
  new LLMError({
    module: "ProviderShared",
@@ -1,20 +1,37 @@
 import { Schema } from "effect"
 import type { CacheHint } from "../../schema"
+import { newBreakpoints, ttlBucket, type Breakpoints } from "./cache"

 // Bedrock cache markers are positional: emit a `cachePoint` block immediately
-// after the content the caller wants treated as a cacheable prefix.
+// after the content the caller wants treated as a cacheable prefix. Bedrock
+// accepts optional `ttl: "5m" | "1h"` on cachePoint, mirroring Anthropic.
 export const CachePointBlock = Schema.Struct({
-  cachePoint: Schema.Struct({ type: Schema.tag("default") }),
+  cachePoint: Schema.Struct({
+    type: Schema.tag("default"),
+    ttl: Schema.optional(Schema.Literals(["5m", "1h"])),
+  }),
 })
 export type CachePointBlock = Schema.Schema.Type<typeof CachePointBlock>

-// Bedrock recently added optional `ttl: "5m" | "1h"` on cachePoint. Map
-// `CacheHint.ttlSeconds` here once a recorded cassette validates the wire shape.
-const DEFAULT: CachePointBlock = { cachePoint: { type: "default" } }
+// Bedrock-Claude enforces the same 4-breakpoint cap as the Anthropic Messages
+// API. Callers pass a shared counter through every `block()` call site so the
+// budget is respected across `system`, `messages`, and `tools`.
+export const BEDROCK_BREAKPOINT_CAP = 4

-export const block = (cache: CacheHint | undefined): CachePointBlock | undefined => {
+export type { Breakpoints } from "./cache"
+export const breakpoints = () => newBreakpoints(BEDROCK_BREAKPOINT_CAP)
+
+const DEFAULT_5M: CachePointBlock = { cachePoint: { type: "default" } }
+const DEFAULT_1H: CachePointBlock = { cachePoint: { type: "default", ttl: "1h" } }
+
+export const block = (breakpoints: Breakpoints, cache: CacheHint | undefined): CachePointBlock | undefined => {
  if (cache?.type !== "ephemeral" && cache?.type !== "persistent") return undefined
-  return DEFAULT
+  if (breakpoints.remaining <= 0) {
+    breakpoints.dropped += 1
+    return undefined
+  }
+  breakpoints.remaining -= 1
+  return ttlBucket(cache.ttlSeconds) === "1h" ? DEFAULT_1H : DEFAULT_5M
 }

 export * as BedrockCache from "./bedrock-cache"
@@ -0,0 +1,16 @@
+// Shared helpers for provider cache-marker lowering. Anthropic and Bedrock
+// both enforce a 4-breakpoint cap per request and accept the same `5m`/`1h`
+// TTL buckets, so the counter and TTL mapping live here.
+
+export interface Breakpoints {
+  remaining: number
+  dropped: number
+}
+
+export const newBreakpoints = (cap: number): Breakpoints => ({ remaining: cap, dropped: 0 })
+
+// Returns `"1h"` for any `ttlSeconds >= 3600`, otherwise `undefined` (the
+// provider default 5m). Anthropic & Bedrock both treat anything shorter than
+// an hour as 5m.
+export const ttlBucket = (ttlSeconds: number | undefined): "1h" | undefined =>
+  ttlSeconds !== undefined && ttlSeconds >= 3600 ? "1h" : undefined
@@ -0,0 +1,88 @@
+import { LLMEvent, type FinishReason, type ProviderMetadata, type Usage } from "../../schema"
+
+export interface State {
+  readonly stepStarted: boolean
+  readonly text: ReadonlySet<string>
+  readonly reasoning: ReadonlySet<string>
+}
+
+export const initial = (): State => ({ stepStarted: false, text: new Set(), reasoning: new Set() })
+
+export const stepStart = (state: State, events: LLMEvent[]): State => {
+  if (state.stepStarted) return state
+  events.push(LLMEvent.stepStart({ index: 0 }))
+  return { ...state, stepStarted: true }
+}
+
+export const textDelta = (state: State, events: LLMEvent[], id: string, text: string): State => {
+  const stepped = stepStart(state, events)
+  if (stepped.text.has(id)) {
+    events.push(LLMEvent.textDelta({ id, text }))
+    return stepped
+  }
+  events.push(LLMEvent.textStart({ id }), LLMEvent.textDelta({ id, text }))
+  return { ...stepped, text: new Set([...stepped.text, id]) }
+}
+
+export const reasoningDelta = (state: State, events: LLMEvent[], id: string, text: string): State => {
+  const stepped = stepStart(state, events)
+  if (stepped.reasoning.has(id)) {
+    events.push(LLMEvent.reasoningDelta({ id, text }))
+    return stepped
+  }
+  events.push(LLMEvent.reasoningStart({ id }), LLMEvent.reasoningDelta({ id, text }))
+  return { ...stepped, reasoning: new Set([...stepped.reasoning, id]) }
+}
+
+export const reasoningEnd = (
+  state: State,
+  events: LLMEvent[],
+  id: string,
+  providerMetadata?: ProviderMetadata,
+): State => {
+  if (!state.reasoning.has(id)) return state
+  const stepped = stepStart(state, events)
+  events.push(LLMEvent.reasoningEnd({ id, providerMetadata }))
+  const reasoning = new Set(stepped.reasoning)
+  reasoning.delete(id)
+  return { ...stepped, reasoning }
+}
+
+export const textEnd = (state: State, events: LLMEvent[], id: string, providerMetadata?: ProviderMetadata): State => {
+  if (!state.text.has(id)) return state
+  const stepped = stepStart(state, events)
+  events.push(LLMEvent.textEnd({ id, providerMetadata }))
+  const text = new Set(stepped.text)
+  text.delete(id)
+  return { ...stepped, text }
+}
+
+const closeOpenBlocks = (state: State, events: LLMEvent[]): State => {
+  for (const id of state.reasoning) events.push(LLMEvent.reasoningEnd({ id }))
+  for (const id of state.text) events.push(LLMEvent.textEnd({ id }))
+  return { ...state, text: new Set(), reasoning: new Set() }
+}
+
+export const finish = (
+  state: State,
+  events: LLMEvent[],
+  input: {
+    readonly reason: FinishReason
+    readonly usage?: Usage
+    readonly providerMetadata?: ProviderMetadata
+  },
+): State => {
+  const stepped = closeOpenBlocks(stepStart(state, events), events)
+  events.push(
+    LLMEvent.stepFinish({
+      index: 0,
+      reason: input.reason,
+      usage: input.usage,
+      providerMetadata: input.providerMetadata,
+    }),
+    LLMEvent.requestFinish(input),
+  )
+  return { ...stepped, stepStarted: false }
+}
+
+export * as Lifecycle from "./lifecycle"
@@ -1,5 +1,5 @@
 import { Effect } from "effect"
-import { LLMError, type ProviderMetadata, type ToolCall, type ToolInputDelta } from "../../schema"
+import { LLMError, LLMEvent, type ProviderMetadata, type ToolCall } from "../../schema"
 import { eventError, parseToolInput, type ToolAccumulator } from "../shared"

 type StreamKey = string | number
@@ -27,13 +27,13 @@ export type State<K extends StreamKey> = Partial<Record<K, PendingTool>>
 /**
 * Result of adding argument text to one pending tool call. It returns both the
 * next `tools` state and the updated `tool` because parsers often need the
- * current id/name immediately. `event` is present only when new text arrived;
- * metadata-only deltas update identity without emitting `tool-input-delta`.
+ * current id/name immediately. `events` contains lifecycle and delta events
+ * produced by the append; metadata-only deltas update identity without output.
 */
 export interface AppendOutcome<K extends StreamKey> {
  readonly tools: State<K>
  readonly tool: PendingTool
-  readonly event?: ToolInputDelta
+  readonly events: ReadonlyArray<LLMEvent>
 }

 /** Create empty accumulator state for one provider stream. */
@@ -49,34 +49,31 @@ const withoutTool = <K extends StreamKey>(tools: State<K>, key: K): State<K> =>
  return next
 }

-const inputDelta = (tool: PendingTool, text: string): ToolInputDelta => ({
-  type: "tool-input-delta",
-  id: tool.id,
-  name: tool.name,
-  text,
-  ...(tool.providerMetadata ? { providerMetadata: tool.providerMetadata } : {}),
-})
+const inputStart = (tool: PendingTool) =>
+  LLMEvent.toolInputStart({
+    id: tool.id,
+    name: tool.name,
+    providerMetadata: tool.providerMetadata,
+  })
+
+const inputDelta = (tool: PendingTool, text: string) =>
+  LLMEvent.toolInputDelta({
+    id: tool.id,
+    name: tool.name,
+    text,
+  })

 const toolCall = (route: string, tool: PendingTool, inputOverride?: string) =>
  parseToolInput(route, tool.name, inputOverride ?? tool.input).pipe(
    Effect.map(
      (input): ToolCall =>
-        tool.providerExecuted
-          ? {
-              type: "tool-call",
-              id: tool.id,
-              name: tool.name,
-              input,
-              providerExecuted: true,
-              ...(tool.providerMetadata ? { providerMetadata: tool.providerMetadata } : {}),
-            }
-          : {
-              type: "tool-call",
-              id: tool.id,
-              name: tool.name,
-              input,
-              ...(tool.providerMetadata ? { providerMetadata: tool.providerMetadata } : {}),
-            },
+        LLMEvent.toolCall({
+          id: tool.id,
+          name: tool.name,
+          input,
+          providerExecuted: tool.providerExecuted ? true : undefined,
+          providerMetadata: tool.providerMetadata,
+        }),
    ),
  )

@@ -86,11 +83,16 @@ const appendTool = <K extends StreamKey>(
  key: K,
  tool: PendingTool,
  text: string,
-): AppendOutcome<K> => ({
-  tools: withTool(tools, key, tool),
-  tool,
-  event: text.length === 0 ? undefined : inputDelta(tool, text),
-})
+): AppendOutcome<K> => {
+  const events: LLMEvent[] = []
+  if (!tools[key]) events.push(inputStart(tool))
+  if (text.length > 0) events.push(inputDelta(tool, text))
+  return {
+    tools: withTool(tools, key, tool),
+    tool,
+    events,
+  }
+}

 export const isError = <K extends StreamKey>(result: AppendOutcome<K> | LLMError): result is LLMError =>
  result instanceof LLMError
@@ -131,7 +133,8 @@ export const appendOrStart = <K extends StreamKey>(
    providerExecuted: current?.providerExecuted,
    providerMetadata: current?.providerMetadata,
  }
-  if (current && delta.text.length === 0 && current.id === id && current.name === name) return { tools, tool: current }
+  if (current && delta.text.length === 0 && current.id === id && current.name === name)
+    return { tools, tool: current, events: [] }
  return appendTool(tools, key, tool, delta.text)
 }

@@ -149,7 +152,7 @@ export const appendExisting = <K extends StreamKey>(
 ): AppendOutcome<K> | LLMError => {
  const current = tools[key]
  if (!current) return eventError(route, missingToolMessage)
-  if (text.length === 0) return { tools, tool: current }
+  if (text.length === 0) return { tools, tool: current, events: [] }
  return appendTool(tools, key, { ...current, input: `${current.input}${text}` }, text)
 }

@@ -162,7 +165,13 @@ export const finish = <K extends StreamKey>(route: string, tools: State<K>, key:
  Effect.gen(function* () {
    const tool = tools[key]
    if (!tool) return { tools }
-    return { tools: withoutTool(tools, key), event: yield* toolCall(route, tool) }
+    return {
+      tools: withoutTool(tools, key),
+      events: [
+        LLMEvent.toolInputEnd({ id: tool.id, name: tool.name, providerMetadata: tool.providerMetadata }),
+        yield* toolCall(route, tool),
+      ],
+    }
  })

 /**
@@ -174,7 +183,13 @@ export const finishWithInput = <K extends StreamKey>(route: string, tools: State
  Effect.gen(function* () {
    const tool = tools[key]
    if (!tool) return { tools }
-    return { tools: withoutTool(tools, key), event: yield* toolCall(route, tool, input) }
+    return {
+      tools: withoutTool(tools, key),
+      events: [
+        LLMEvent.toolInputEnd({ id: tool.id, name: tool.name, providerMetadata: tool.providerMetadata }),
+        yield* toolCall(route, tool, input),
+      ],
+    }
  })

 /**
@@ -189,7 +204,14 @@ export const finishAll = <K extends StreamKey>(route: string, tools: State<K>) =
    )
    return {
      tools: empty<K>(),
-      events: yield* Effect.forEach(pending, (tool) => toolCall(route, tool)),
+      events: yield* Effect.forEach(pending, (tool) =>
+        toolCall(route, tool).pipe(
+          Effect.map((call) => [
+            LLMEvent.toolInputEnd({ id: tool.id, name: tool.name, providerMetadata: tool.providerMetadata }),
+            call,
+          ]),
+        ),
+      ).pipe(Effect.map((events) => events.flat())),
    }
  })

@@ -8,6 +8,7 @@ import type { Transport, TransportRuntime } from "./transport"
 import { WebSocketExecutor } from "./transport"
 import type { Service as WebSocketExecutorService } from "./transport/websocket"
 import type { Protocol } from "./protocol"
+import { applyCachePolicy } from "../cache-policy"
 import * as ProviderShared from "../protocols/shared"
 import * as ToolRuntime from "../tool-runtime"
 import type { Tools } from "../tool"
@@ -400,7 +401,7 @@ export function make<Body, Prepared, Frame, Event, State>(
 // validated provider body plus transport-private prepared data, but does not
 // execute transport.
 const compile = Effect.fn("LLM.compile")(function* (request: LLMRequest) {
-  const resolved = resolveRequestOptions(request)
+  const resolved = applyCachePolicy(resolveRequestOptions(request))
  const route = registeredRoute(resolved.model.route)
  if (!route) return yield* noRoute(resolved.model)

@@ -1,73 +1,155 @@
 import { Schema } from "effect"
-import { FinishReason, ProtocolID, ProviderMetadata, RouteID } from "./ids"
+import { ContentBlockID, FinishReason, ProtocolID, ProviderMetadata, ResponseID, RouteID, ToolCallID } from "./ids"
 import { ModelRef } from "./options"
 import { ToolResultValue } from "./messages"

+/**
+ * Token usage reported by an LLM provider.
+ *
+ * **Inclusive totals** (match AI SDK / OpenAI / LangChain convention — a
+ * reader from any of those ecosystems sees the number they expect):
+ *
+ * - `inputTokens` — total prompt tokens, *including* cached reads/writes.
+ * - `outputTokens` — total output tokens, *including* reasoning.
+ * - `totalTokens` — provider-supplied total, or `inputTokens + outputTokens`.
+ *
+ * **Non-overlapping breakdown** (every field is independently meaningful;
+ * consumers never have to subtract):
+ *
+ * - `nonCachedInputTokens` — the "fresh" portion of the prompt.
+ * - `cacheReadInputTokens` — input tokens served from cache.
+ * - `cacheWriteInputTokens` — input tokens written to cache.
+ * - `reasoningTokens` — subset of `outputTokens` spent on hidden reasoning.
+ *
+ * **Invariant**: `nonCachedInputTokens + cacheReadInputTokens +
+ * cacheWriteInputTokens = inputTokens`, and `reasoningTokens ≤ outputTokens`.
+ * Each protocol mapper computes whichever side it doesn't get natively,
+ * with `Math.max(0, …)` clamping for defense against provider bugs. Because
+ * every breakdown field is stored independently, downstream consumers can
+ * read whatever they need (cost-by-category, context-pressure, AI-SDK-style
+ * inclusive total) without ever subtracting — eliminating the underflow
+ * class of bug where a clamped difference would silently store the wrong
+ * value.
+ *
+ * **Semantics by provider**:
+ *
+ * - OpenAI Chat / Responses / Gemini / Bedrock: provider reports inclusive
+ *   `inputTokens` and an inclusive `outputTokens`; mapper subtracts to
+ *   derive the breakdown.
+ * - Anthropic: provider reports the breakdown natively (`input_tokens` is
+ *   non-cached only); mapper sums to derive the inclusive `inputTokens`.
+ *   Anthropic does *not* break extended-thinking out of `output_tokens`, so
+ *   `reasoningTokens` is `undefined` and `outputTokens` carries the
+ *   combined total — a documented limitation of the Anthropic API.
+ *
+ * `providerMetadata` always carries the provider's raw usage payload —
+ * keyed by provider name (`{ openai: ... }`, `{ anthropic: ... }`, etc.)
+ * — for fields we don't normalize and for billing-level audit trails.
+ * Matches the same escape-hatch field on `LLMEvent`.
+ */
 export class Usage extends Schema.Class<Usage>("LLM.Usage")({
  inputTokens: Schema.optional(Schema.Number),
  outputTokens: Schema.optional(Schema.Number),
-  reasoningTokens: Schema.optional(Schema.Number),
+  nonCachedInputTokens: Schema.optional(Schema.Number),
  cacheReadInputTokens: Schema.optional(Schema.Number),
  cacheWriteInputTokens: Schema.optional(Schema.Number),
+  reasoningTokens: Schema.optional(Schema.Number),
  totalTokens: Schema.optional(Schema.Number),
-  native: Schema.optional(Schema.Record(Schema.String, Schema.Unknown)),
-}) {}
+  providerMetadata: Schema.optional(ProviderMetadata),
+}) {
+  /**
+   * Visible output tokens — `outputTokens` minus `reasoningTokens`, clamped
+   * to zero. The one place subtraction happens in this contract; the clamp
+   * means a provider reporting `reasoningTokens > outputTokens` produces a
+   * harmless zero rather than a negative that crashes downstream schemas.
+   */
+  get visibleOutputTokens() {
+    return Math.max(0, (this.outputTokens ?? 0) - (this.reasoningTokens ?? 0))
+  }
+}

 export const RequestStart = Schema.Struct({
-  type: Schema.Literal("request-start"),
-  id: Schema.String,
+  type: Schema.tag("request-start"),
+  id: ResponseID,
  model: ModelRef,
 }).annotate({ identifier: "LLM.Event.RequestStart" })
 export type RequestStart = Schema.Schema.Type<typeof RequestStart>

 export const StepStart = Schema.Struct({
-  type: Schema.Literal("step-start"),
+  type: Schema.tag("step-start"),
  index: Schema.Number,
 }).annotate({ identifier: "LLM.Event.StepStart" })
 export type StepStart = Schema.Schema.Type<typeof StepStart>

 export const TextStart = Schema.Struct({
-  type: Schema.Literal("text-start"),
-  id: Schema.String,
+  type: Schema.tag("text-start"),
+  id: ContentBlockID,
  providerMetadata: Schema.optional(ProviderMetadata),
 }).annotate({ identifier: "LLM.Event.TextStart" })
 export type TextStart = Schema.Schema.Type<typeof TextStart>

 export const TextDelta = Schema.Struct({
-  type: Schema.Literal("text-delta"),
-  id: Schema.optional(Schema.String),
+  type: Schema.tag("text-delta"),
+  id: ContentBlockID,
  text: Schema.String,
-  providerMetadata: Schema.optional(ProviderMetadata),
 }).annotate({ identifier: "LLM.Event.TextDelta" })
 export type TextDelta = Schema.Schema.Type<typeof TextDelta>

 export const TextEnd = Schema.Struct({
-  type: Schema.Literal("text-end"),
-  id: Schema.String,
+  type: Schema.tag("text-end"),
+  id: ContentBlockID,
  providerMetadata: Schema.optional(ProviderMetadata),
 }).annotate({ identifier: "LLM.Event.TextEnd" })
 export type TextEnd = Schema.Schema.Type<typeof TextEnd>

-export const ReasoningDelta = Schema.Struct({
-  type: Schema.Literal("reasoning-delta"),
-  id: Schema.optional(Schema.String),
-  text: Schema.String,
+export const ReasoningStart = Schema.Struct({
+  type: Schema.tag("reasoning-start"),
+  id: ContentBlockID,
  providerMetadata: Schema.optional(ProviderMetadata),
+}).annotate({ identifier: "LLM.Event.ReasoningStart" })
+export type ReasoningStart = Schema.Schema.Type<typeof ReasoningStart>
+
+export const ReasoningDelta = Schema.Struct({
+  type: Schema.tag("reasoning-delta"),
+  id: ContentBlockID,
+  text: Schema.String,
 }).annotate({ identifier: "LLM.Event.ReasoningDelta" })
 export type ReasoningDelta = Schema.Schema.Type<typeof ReasoningDelta>

+export const ReasoningEnd = Schema.Struct({
+  type: Schema.tag("reasoning-end"),
+  id: ContentBlockID,
+  providerMetadata: Schema.optional(ProviderMetadata),
+}).annotate({ identifier: "LLM.Event.ReasoningEnd" })
+export type ReasoningEnd = Schema.Schema.Type<typeof ReasoningEnd>
+
+export const ToolInputStart = Schema.Struct({
+  type: Schema.tag("tool-input-start"),
+  id: ToolCallID,
+  name: Schema.String,
+  providerMetadata: Schema.optional(ProviderMetadata),
+}).annotate({ identifier: "LLM.Event.ToolInputStart" })
+export type ToolInputStart = Schema.Schema.Type<typeof ToolInputStart>
+
 export const ToolInputDelta = Schema.Struct({
-  type: Schema.Literal("tool-input-delta"),
-  id: Schema.String,
+  type: Schema.tag("tool-input-delta"),
+  id: ToolCallID,
  name: Schema.String,
  text: Schema.String,
-  providerMetadata: Schema.optional(ProviderMetadata),
 }).annotate({ identifier: "LLM.Event.ToolInputDelta" })
 export type ToolInputDelta = Schema.Schema.Type<typeof ToolInputDelta>

+export const ToolInputEnd = Schema.Struct({
+  type: Schema.tag("tool-input-end"),
+  id: ToolCallID,
+  name: Schema.String,
+  providerMetadata: Schema.optional(ProviderMetadata),
+}).annotate({ identifier: "LLM.Event.ToolInputEnd" })
+export type ToolInputEnd = Schema.Schema.Type<typeof ToolInputEnd>
+
 export const ToolCall = Schema.Struct({
-  type: Schema.Literal("tool-call"),
-  id: Schema.String,
+  type: Schema.tag("tool-call"),
+  id: ToolCallID,
  name: Schema.String,
  input: Schema.Unknown,
  providerExecuted: Schema.optional(Schema.Boolean),
@@ -76,8 +158,8 @@ export const ToolCall = Schema.Struct({
 export type ToolCall = Schema.Schema.Type<typeof ToolCall>

 export const ToolResult = Schema.Struct({
-  type: Schema.Literal("tool-result"),
-  id: Schema.String,
+  type: Schema.tag("tool-result"),
+  id: ToolCallID,
  name: Schema.String,
  result: ToolResultValue,
  providerExecuted: Schema.optional(Schema.Boolean),
@@ -86,8 +168,8 @@ export const ToolResult = Schema.Struct({
 export type ToolResult = Schema.Schema.Type<typeof ToolResult>

 export const ToolError = Schema.Struct({
-  type: Schema.Literal("tool-error"),
-  id: Schema.String,
+  type: Schema.tag("tool-error"),
+  id: ToolCallID,
  name: Schema.String,
  message: Schema.String,
  providerMetadata: Schema.optional(ProviderMetadata),
@@ -95,7 +177,7 @@ export const ToolError = Schema.Struct({
 export type ToolError = Schema.Schema.Type<typeof ToolError>

 export const StepFinish = Schema.Struct({
-  type: Schema.Literal("step-finish"),
+  type: Schema.tag("step-finish"),
  index: Schema.Number,
  reason: FinishReason,
  usage: Schema.optional(Usage),
@@ -104,7 +186,7 @@ export const StepFinish = Schema.Struct({
 export type StepFinish = Schema.Schema.Type<typeof StepFinish>

 export const RequestFinish = Schema.Struct({
-  type: Schema.Literal("request-finish"),
+  type: Schema.tag("request-finish"),
  reason: FinishReason,
  usage: Schema.optional(Usage),
  providerMetadata: Schema.optional(ProviderMetadata),
@@ -112,7 +194,7 @@ export const RequestFinish = Schema.Struct({
 export type RequestFinish = Schema.Schema.Type<typeof RequestFinish>

 export const ProviderErrorEvent = Schema.Struct({
-  type: Schema.Literal("provider-error"),
+  type: Schema.tag("provider-error"),
  message: Schema.String,
  retryable: Schema.optional(Schema.Boolean),
  providerMetadata: Schema.optional(ProviderMetadata),
@@ -125,8 +207,12 @@ const llmEventTagged = Schema.Union([
  TextStart,
  TextDelta,
  TextEnd,
+  ReasoningStart,
  ReasoningDelta,
+  ReasoningEnd,
+  ToolInputStart,
  ToolInputDelta,
+  ToolInputEnd,
  ToolCall,
  ToolResult,
  ToolError,
@@ -135,20 +221,52 @@ const llmEventTagged = Schema.Union([
  ProviderErrorEvent,
 ]).pipe(Schema.toTaggedUnion("type"))

+type WithID<Event extends { readonly id: unknown }, ID> = Omit<Event, "type" | "id"> & { readonly id: ID | string }
+
+const responseID = (value: ResponseID | string) => ResponseID.make(value)
+const contentBlockID = (value: ContentBlockID | string) => ContentBlockID.make(value)
+const toolCallID = (value: ToolCallID | string) => ToolCallID.make(value)
+
 /**
 * camelCase aliases for `LLMEvent.guards` (provided by `Schema.toTaggedUnion`).
 * Lets consumers write `events.filter(LLMEvent.is.toolCall)` instead of
 * `events.filter(LLMEvent.guards["tool-call"])`.
 */
 export const LLMEvent = Object.assign(llmEventTagged, {
+  requestStart: (input: WithID<RequestStart, ResponseID>) => RequestStart.make({ ...input, id: responseID(input.id) }),
+  stepStart: StepStart.make,
+  textStart: (input: WithID<TextStart, ContentBlockID>) => TextStart.make({ ...input, id: contentBlockID(input.id) }),
+  textDelta: (input: WithID<TextDelta, ContentBlockID>) => TextDelta.make({ ...input, id: contentBlockID(input.id) }),
+  textEnd: (input: WithID<TextEnd, ContentBlockID>) => TextEnd.make({ ...input, id: contentBlockID(input.id) }),
+  reasoningStart: (input: WithID<ReasoningStart, ContentBlockID>) =>
+    ReasoningStart.make({ ...input, id: contentBlockID(input.id) }),
+  reasoningDelta: (input: WithID<ReasoningDelta, ContentBlockID>) =>
+    ReasoningDelta.make({ ...input, id: contentBlockID(input.id) }),
+  reasoningEnd: (input: WithID<ReasoningEnd, ContentBlockID>) =>
+    ReasoningEnd.make({ ...input, id: contentBlockID(input.id) }),
+  toolInputStart: (input: WithID<ToolInputStart, ToolCallID>) =>
+    ToolInputStart.make({ ...input, id: toolCallID(input.id) }),
+  toolInputDelta: (input: WithID<ToolInputDelta, ToolCallID>) =>
+    ToolInputDelta.make({ ...input, id: toolCallID(input.id) }),
+  toolInputEnd: (input: WithID<ToolInputEnd, ToolCallID>) => ToolInputEnd.make({ ...input, id: toolCallID(input.id) }),
+  toolCall: (input: WithID<ToolCall, ToolCallID>) => ToolCall.make({ ...input, id: toolCallID(input.id) }),
+  toolResult: (input: WithID<ToolResult, ToolCallID>) => ToolResult.make({ ...input, id: toolCallID(input.id) }),
+  toolError: (input: WithID<ToolError, ToolCallID>) => ToolError.make({ ...input, id: toolCallID(input.id) }),
+  stepFinish: StepFinish.make,
+  requestFinish: RequestFinish.make,
+  providerError: ProviderErrorEvent.make,
  is: {
    requestStart: llmEventTagged.guards["request-start"],
    stepStart: llmEventTagged.guards["step-start"],
    textStart: llmEventTagged.guards["text-start"],
    textDelta: llmEventTagged.guards["text-delta"],
    textEnd: llmEventTagged.guards["text-end"],
+    reasoningStart: llmEventTagged.guards["reasoning-start"],
    reasoningDelta: llmEventTagged.guards["reasoning-delta"],
+    reasoningEnd: llmEventTagged.guards["reasoning-end"],
+    toolInputStart: llmEventTagged.guards["tool-input-start"],
    toolInputDelta: llmEventTagged.guards["tool-input-delta"],
+    toolInputEnd: llmEventTagged.guards["tool-input-end"],
    toolCall: llmEventTagged.guards["tool-call"],
    toolResult: llmEventTagged.guards["tool-result"],
    toolError: llmEventTagged.guards["tool-error"],
@@ -14,6 +14,15 @@ export type ModelID = typeof ModelID.Type
 export const ProviderID = Schema.String.pipe(Schema.brand("LLM.ProviderID"))
 export type ProviderID = typeof ProviderID.Type

+export const ResponseID = Schema.String
+export type ResponseID = Schema.Schema.Type<typeof ResponseID>
+
+export const ContentBlockID = Schema.String
+export type ContentBlockID = Schema.Schema.Type<typeof ContentBlockID>
+
+export const ToolCallID = Schema.String
+export type ToolCallID = Schema.Schema.Type<typeof ToolCallID>
+
 export const ReasoningEfforts = ["none", "minimal", "low", "medium", "high", "xhigh", "max"] as const
 export const ReasoningEffort = Schema.Literals(ReasoningEfforts)
 export type ReasoningEffort = Schema.Schema.Type<typeof ReasoningEffort>
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Kit LangtonandGitHub	613a8d1beb	test(provider): migrate more config-backed cases (#26984 )	2026-05-11 22:21:50 -04:00
Kit Langton	708557880b	test(provider): migrate provider tests to effect runner	2026-05-11 22:13:26 -04:00
Kit LangtonandGitHub	0f5d4ae648	test(project): stabilize VCS branch update test (#26979 )	2026-05-11 22:12:07 -04:00
Kit LangtonandGitHub	ce72020750	test(tool): migrate edit tests to Effect runner (#26977 )	2026-05-11 21:42:17 -04:00
Kit LangtonandGitHub	c43d606f8e	agent: use Effect schema for generated agent object (#26973 )	2026-05-11 21:42:04 -04:00
Kit LangtonandGitHub	1007630347	Migrate runtime validators to Effect Schema (#26975 )	2026-05-11 21:41:56 -04:00
Kit LangtonandGitHub	9e8274d2da	Remove internal Zod schemas (#26974 )	2026-05-11 21:40:44 -04:00
Kit LangtonandGitHub	74aa735e6a	fix(tui): guard prompt submit against concurrent invocation (#26972 )	2026-05-12 01:35:28 +00:00
Kit LangtonandGitHub	8030a6c187	Emit LLM stream lifecycle events (#26971 )	2026-05-11 21:31:48 -04:00
Kit LangtonandGitHub	e5aa5161f2	Remove effect-zod bridge (#26956 )	2026-05-11 21:14:55 -04:00
Kit LangtonandGitHub	abb1ee6278	docs(test): add Effect migration orchestration notes (#26963 )	2026-05-11 20:59:51 -04:00
Kit LangtonandGitHub	c4003579bb	test(project): migrate VCS tests to Effect runner (#26965 )	2026-05-11 20:59:34 -04:00
Kit LangtonandGitHub	0d9c534184	test(snapshot): migrate snapshot tests to Effect runner (#26964 )	2026-05-11 20:59:31 -04:00
Aiden ClineandGitHub	5773d43cbf	ci: GitHub Actions dependencies (#26962 )	2026-05-11 19:50:35 -05:00
opencode-agent[bot]	e0e9414cbd	chore: generate	2026-05-12 00:41:30 +00:00
Kit LangtonandGitHub	44edb639c2	test(session): migrate message pagination to Effect runner (#26957 )	2026-05-12 00:40:23 +00:00
Kit LangtonandGitHub	fbd52ca2f4	test(file): migrate file tests to Effect runner (#26959 )	2026-05-12 00:39:31 +00:00
opencode-agent[bot]	8015ff7ca5	chore: generate	2026-05-12 00:33:34 +00:00
Kit LangtonandGitHub	ec9584177f	docs(test): plan Effect test migration (#26954 )	2026-05-11 20:32:27 -04:00
Kit LangtonandGitHub	061efc6cf2	Fix run JSON output draining (#26955 )	2026-05-11 20:30:23 -04:00
Brendan AllanandGitHub	fe374aea46	feat(app): persist todo dock collapsed state (#26953 )	2026-05-11 23:59:38 +00:00
Kit LangtonandGitHub	46edc98f10	Validate TUI config with Effect Schema (#26952 )	2026-05-11 23:51:45 +00:00
Kit LangtonandGitHub	fdeb2748e1	test(agent): isolate plugin agent regression (#26948 )	2026-05-11 19:38:54 -04:00
Kit LangtonandGitHub	59e6967b8f	Generate config schema from Effect Schema (#26939 )	2026-05-11 19:24:58 -04:00
opencode-agent[bot]	0c619cb59c	chore: generate	2026-05-11 22:34:04 +00:00
Frank	0cf90109dc	zen: tps rate limit	2026-05-11 18:32:39 -04:00
Kit LangtonandGitHub	812668ae2f	Generate TUI schema from Effect Schema (#26945 )	2026-05-11 17:50:14 -04:00
Kit LangtonandGitHub	a5c35bf182	Avoid bootstrapping server plugins from TUI plugin runtime (#26938 )	2026-05-11 17:32:16 -04:00
opencode-agent[bot]	bdd5a80886	chore: update nix node_modules hashes	2026-05-11 21:28:49 +00:00
Kit LangtonandGitHub	fd65d29dcc	Drop unused opencode Zod statics (#26935 )	2026-05-11 17:14:18 -04:00
Kit LangtonandGitHub	d3caac5270	chore(deps): upgrade effect to 4.0.0-beta.65 (#26934 )	2026-05-11 21:09:57 +00:00
Kit LangtonandGitHub	fe7ca3421e	Drop Config Info Zod static (#26933 )	2026-05-11 16:49:22 -04:00
Kit LangtonandGitHub	cc95197d72	Drop prompt input Zod statics (#26923 )	2026-05-11 16:49:08 -04:00
James LongandGitHub	9067218b74	fix(core): always start worktrees as detached (#26931 )	2026-05-11 16:22:25 -04:00
Kit LangtonandGitHub	42a0453945	Drop small session Zod statics (#26921 )	2026-05-11 16:12:31 -04:00
Kit LangtonandGitHub	4d9eb6c320	Validate structured output tests with Effect Schema (#26919 )	2026-05-11 16:11:25 -04:00
Kit LangtonandGitHub	c060c436b6	Drop LSP config Zod statics (#26920 )	2026-05-11 16:11:11 -04:00
Kit LangtonandGitHub	45adfedd64	Drop unused domain Zod statics (#26927 )	2026-05-11 16:10:58 -04:00
opencode-agent[bot]	0bced8ec96	chore: update nix node_modules hashes	2026-05-11 17:32:56 +00:00
Kit LangtonandGitHub	8d9b9719be	Drop unused small ID Zod statics (#26908 )	2026-05-11 13:22:59 -04:00
Kit LangtonandGitHub	c7e084c32c	Simplify single-backend HttpApi exerciser (#26906 )	2026-05-11 13:22:31 -04:00
Kit LangtonandGitHub	6f2f759fbb	Clean up post-Hono references (#26903 )	2026-05-11 13:20:54 -04:00
opencode-agent[bot]	cddab63808	chore: generate	2026-05-11 17:17:38 +00:00
Shoubhit DashandGitHub	12583b18f0	feat(tui): pin, quick-switch, and cycle recent sessions (#26858 )	2026-05-11 22:46:27 +05:30
Kit LangtonandGitHub	df386bd651	feat(skill): enable customize-opencode by default, link full schema (#26899 )	2026-05-11 13:13:41 -04:00
opencode-agent[bot]	25b12ed754	chore: generate	2026-05-11 17:06:57 +00:00
Kit LangtonandGitHub	023e1c711e	refactor(llm): colocate per-type factories on their namespaces (#26799 )	2026-05-11 13:05:41 -04:00
Kit LangtonandGitHub	52f7ba7d4d	fix(llm): drop removed dispatch option from recorded cache tests (#26900 )	2026-05-11 11:48:30 -04:00
opencode-agent[bot]	19fce2bc6f	chore: generate	2026-05-11 15:21:20 +00:00
Shoubhit DashandGitHub	4bae84c8b0	feat(scout): autocomplete configured mentions (#26843 )	2026-05-11 20:50:03 +05:30
Kit LangtonandGitHub	f240bba8e7	chore(http-recorder): remove content-matching dispatch mode (#26792 )	2026-05-11 11:10:18 -04:00
Kit LangtonandGitHub	bcee247988	Define project update input with Effect Schema (#26803 )	2026-05-11 15:05:31 +00:00
Kit LangtonandGitHub	64c504212a	Parse command config with Effect Schema (#26801 )	2026-05-11 10:22:45 -04:00
opencode-agent[bot]	6592d80460	chore: generate	2026-05-11 14:13:39 +00:00
SebastianandGitHub	d821650b48	add default diff parser for markdown fenced code blocks (#26887 )	2026-05-11 14:12:32 +00:00
Victor NavarroandGitHub	6c9f12bb2d	chore: exclude status 429 from free model alerts (#26879 )	2026-05-11 16:04:54 +02:00
opencode-agent[bot]	fca89e5e4d	chore: generate	2026-05-11 13:44:39 +00:00
Frank	1eff01b6a5	sync	2026-05-11 09:43:12 -04:00
opencode-agent[bot]	e93b1f3a7f	chore: generate	2026-05-11 13:38:44 +00:00
Frank	8874d4a42f	zen: deekseek v4 flash free	2026-05-11 09:37:14 -04:00
Brendan AllanandGitHub	c933504d9c	fix(ui): better handle patch file support when rendering patch/edit tools (#26828 )	2026-05-11 17:21:59 +08:00
Shoubhit DashandGitHub	2d0d3d596e	feat(compaction): serialize compaction tail (#26830 )	2026-05-11 13:40:36 +05:30
opencode-agent[bot]	3bd98ea055	chore: generate	2026-05-11 07:28:32 +00:00
Shoubhit DashandGitHub	7e997cfba4	refactor(scout): resolve configured reference mentions (#26701 )	2026-05-11 12:57:26 +05:30
Brendan AllanandGitHub	5d6f2a1524	fix(ui): use part_text_accum_delta to prevent markdown cutoff during streaming (#26822 )	2026-05-11 15:15:03 +08:00
opencode-agent[bot]	b1cb71856e	chore: generate	2026-05-11 05:27:40 +00:00
Aiden ClineandGitHub	518264fcd9	fix(opencode): fix full session fork (#26811 )	2026-05-11 00:26:36 -05:00
DaxandGitHub	7235c9c9b8	Trace data migrations (#26809 )	2026-05-11 03:23:01 +00:00
Kit LangtonandGitHub	274033cd52	Validate prompt messages with Effect Schema (#26796 )	2026-05-11 02:59:20 +00:00
Kit LangtonandGitHub	9b369ee815	chore(llm): make cache: 'auto' the default (#26798 )	2026-05-10 22:46:01 -04:00
SebastianandGitHub	721ff5121e	fix prompt history behaviour and session line up/down commands (#26797 )	2026-05-11 02:27:46 +00:00
opencode-agent[bot]	cfbf5d1c6f	chore: update nix node_modules hashes	2026-05-11 02:20:35 +00:00
opencode-agent[bot]	02cb7e7b71	chore: generate	2026-05-11 02:11:07 +00:00
Kit LangtonandGitHub	942630eb4a	feat(llm): cache-policy auto-placement (#26786 )	2026-05-10 22:09:55 -04:00
opencode	ce66b191d1	sync release versions for v1.14.48	2026-05-11 02:07:48 +00:00
Kit LangtonandGitHub	6f1f5944ce	Delete unused opencode Zod helpers (#26793 )	2026-05-10 21:57:18 -04:00
opencode-agent[bot]	1c49b2ed67	chore: generate	2026-05-11 01:55:22 +00:00
opencode-agent[bot]	3c6be604ec	chore: update nix node_modules hashes	2026-05-11 01:54:13 +00:00
Kit LangtonandGitHub	ddc02c2893	Drop synchronous SyncEvent facades (#26789 )	2026-05-11 01:53:50 +00:00
Kit LangtonandGitHub	2703eff2e2	refactor(llm): normalize Usage as inclusive total + non-overlapping breakdown (#26735 )	2026-05-10 21:52:50 -04:00
opencode-agent[bot]	38e4540119	chore: generate	2026-05-11 01:41:26 +00:00
Dax Raad	4100fcbd17	disable image resizing	2026-05-10 21:40:12 -04:00
Kit LangtonandGitHub	83cb0f60ec	Drop EventV2 run facade (#26783 )	2026-05-10 21:26:28 -04:00
Kit LangtonandGitHub	effd96755e	Use SyncEvent service at event call sites (#26782 )	2026-05-10 21:20:13 -04:00
opencode-agent[bot]	5801cce1b5	chore: generate	2026-05-11 01:18:39 +00:00
Kit LangtonandGitHub	77e6c0d329	feat(llm): cache hint TTL, breakpoint cap, and tool placement (#26779 )	2026-05-10 21:17:38 -04:00
Kit LangtonandGitHub	fed716ada5	Clarify compaction test harness (#26777 )	2026-05-10 20:41:21 -04:00
opencode	426d92e352	sync release versions for v1.14.47	2026-05-11 00:23:17 +00:00
opencode-agent[bot]	d0412a213b	chore: generate	2026-05-11 00:18:03 +00:00
Kit LangtonandGitHub	16aa67086b	Effectify remaining compaction process tests (#26776 )	2026-05-10 20:17:01 -04:00
DaxandGitHub	7cea32ee08	Add background code migration service (#26652 )	2026-05-11 00:16:42 +00:00
Kit LangtonandGitHub	128d10d9e9	Simplify compaction test helpers (#26742 )	2026-05-10 20:03:11 -04:00
Kit LangtonandGitHub	5ef72e1101	Drop unused ID Zod statics (#26740 )	2026-05-10 19:58:21 -04:00
Kit LangtonandGitHub	d8060dc9ad	Drop compaction create facade (#26739 )	2026-05-10 19:54:16 -04:00
Kit LangtonandGitHub	f71fb18d3d	Replace compaction create test fixtures (#26738 )	2026-05-10 19:53:22 -04:00
SebastianandGitHub	5654dd2aad	restore managed textarea keymap handling (#26771 )	2026-05-11 01:45:59 +02:00
Kit LangtonandGitHub	64dde0cb15	Migrate compaction tool-call test (#26737 )	2026-05-10 19:44:42 -04:00
Kit LangtonandGitHub	a658e43eeb	Migrate compaction plugin test (#26736 )	2026-05-10 19:42:44 -04:00
Kit LangtonandGitHub	ca8a42c973	Migrate LLM compaction tail tests (#26734 )	2026-05-10 19:41:28 -04:00
Kit LangtonandGitHub	8f5f75db12	Migrate compaction LLM test (#26733 )	2026-05-10 19:20:44 -04:00
Kit LangtonandGitHub	4ff0d07b1d	Migrate configurable compaction tests (#26732 )	2026-05-10 19:18:02 -04:00
Kit LangtonandGitHub	2a571b3cee	Migrate compaction overflow tests (#26731 )	2026-05-10 19:12:49 -04:00
DaxandGitHub	0fffcdfe46	Persist session model switches outside event flag (#26765 )	2026-05-10 18:53:14 -04:00
DaxandGitHub	2ba2ee52e8	docs: document bun dev tmux capture (#26764 )	2026-05-10 17:41:20 -04:00
Frank	56d818fc34	zen: fix reasoning token for openai compatible endpoint	2026-05-10 13:36:02 -04:00
Kit LangtonandGitHub	9c8da69196	Use Effect timeout in compaction test (#26728 )	2026-05-10 12:45:54 -04:00
opencode-agent[bot]	a78018697c	chore: generate	2026-05-10 16:44:40 +00:00
Kit LangtonandGitHub	e45b6ef1de	refactor(http-recorder): use Schema.TaggedErrorClass for cassette errors (#26729 )	2026-05-10 16:43:33 +00:00
opencode-agent[bot]	b616543ac2	chore: generate	2026-05-10 16:30:55 +00:00
Kit LangtonandGitHub	2bd3d9a696	refactor(http-recorder): hide cassette format behind Cassette seam (#26725 )	2026-05-10 12:29:55 -04:00
Kit LangtonandGitHub	fa15dbc5ec	Migrate compaction process tests (#26723 )	2026-05-10 12:25:44 -04:00
opencode-agent[bot]	312e5c7a7c	chore: generate	2026-05-10 16:22:29 +00:00
Kit LangtonandGitHub	049502fac6	fix(server): return diagnosable body for schema rejections (#26631 )	2026-05-10 16:21:32 +00:00
opencode-agent[bot]	cc2915be16	chore: generate	2026-05-10 16:20:16 +00:00
Kit LangtonandGitHub	ce061bf661	Add explicit LLM stream lifecycle events (#26722 )	2026-05-10 12:19:13 -04:00
Frank	3b8790e034	zen: fix usage css on mobile	2026-05-10 12:14:11 -04:00
Kit LangtonandGitHub	a4f3cedcdf	Start effect-style compaction tests	2026-05-10 16:12:00 +00:00
opencode-agent[bot]	1c9a2eb239	chore: generate	2026-05-10 16:06:18 +00:00
Kit LangtonandGitHub	4fb417d3b5	feat(http-recorder): default mode to "auto" (#26719 )	2026-05-10 16:05:11 +00:00
Kit LangtonandGitHub	11030c627b	Scope boolean query overrides	2026-05-10 11:57:52 -04:00
opencode-agent[bot]	c104098a66	chore: generate	2026-05-10 15:55:49 +00:00
Kit LangtonandGitHub	49ee3ba85a	Source diff message query pattern (#26638 )	2026-05-10 11:54:54 -04:00
opencode-agent[bot]	4fc538378d	chore: generate	2026-05-10 14:50:21 +00:00
Kit LangtonandGitHub	d28b5ad2f4	refactor(http-recorder): Redactor + Recorder seams, README (#26636 )	2026-05-10 10:49:22 -04:00
opencode-agent[bot]	6589a66822	chore: generate	2026-05-10 12:28:11 +00:00
Shoubhit DashandGitHub	5cf9abe743	feat(scout): materialize configured reference repos (#26692 )	2026-05-10 17:57:11 +05:30
Frank	903d81819d	Zen: add Ring 2.6 1T	2026-05-10 03:51:34 -04:00
opencode-agent[bot]	472f9e64a6	chore: update nix node_modules hashes	2026-05-10 07:06:30 +00:00
Frank	c04fa9e253	sync: revert This reverts commit `3a7f617098`.	2026-05-10 02:58:46 -04:00
opencode-agent[bot]	3a78fb1f42	chore: generate	2026-05-10 06:49:21 +00:00
Aiden ClineandGitHub	85ce6a5f95	feat: better image handling (auto resize & max size constraints) (#26401 )	2026-05-10 01:48:19 -05:00
opencode-agent[bot]	5217e6c1af	chore: generate	2026-05-10 06:39:09 +00:00
Frank	3a7f617098	go: add tencent icon	2026-05-10 02:37:50 -04:00
opencode-agent[bot]	d9150413cb	chore: generate	2026-05-10 06:24:35 +00:00
JackandGitHub	bcbc1dba22	Go add hy3 preview (#26533 )	2026-05-10 02:23:34 -04:00
Frank	ce3235e115	sync	2026-05-10 02:17:32 -04:00
opencode-agent[bot]	a9a2a597d5	chore: generate	2026-05-10 04:30:04 +00:00
DaxandGitHub	3753601f87	Format TUI paths relative to session directory (#26648 )	2026-05-10 04:29:02 +00:00
Kit LangtonandGitHub	fb4bab8a66	Remove redundant ID Zod overrides (#26633 )	2026-05-09 23:12:21 -04:00
opencode-agent[bot]	b3526f6ce9	chore: generate	2026-05-10 03:03:37 +00:00
Kit LangtonandGitHub	f220f02a2f	Source workspace path pattern (#26632 )	2026-05-09 23:02:31 -04:00
opencode-agent[bot]	235a86fb60	chore: generate	2026-05-10 02:59:46 +00:00
Kit LangtonandGitHub	67b9c9c027	Source HTTP API ID path patterns (#26623 )	2026-05-09 22:58:47 -04:00
opencode	2f11c9f7ed	sync release versions for v1.14.46	2026-05-10 02:34:36 +00:00
opencode-agent[bot]	e1c1193f3e	chore: generate	2026-05-10 02:11:45 +00:00
Kit LangtonandGitHub	29250a0efb	fix(session): loosen remaining stored numeric schemas to tolerate legacy data (#26622 )	2026-05-09 22:10:48 -04:00
Kit LangtonandGitHub	c6e6bdf59f	fix(session): tolerate negative token counts in stored parts (#26620 )	2026-05-09 22:10:44 -04:00
opencode-agent[bot]	d80e1199ca	chore: generate	2026-05-10 02:06:39 +00:00
Kit LangtonandGitHub	10ea59066f	feat(skill): built-in opencode-meta skill (#26617 )	2026-05-09 22:05:37 -04:00
Kit LangtonandGitHub	79d6b10d7c	fix(mcp): tolerate output schema ref failures (#26614 )	2026-05-09 22:03:59 -04:00
Kit LangtonandGitHub	6e78f36a0f	Narrow HTTP API numeric query overrides (#26618 )	2026-05-09 22:02:51 -04:00
Kit LangtonandGitHub	16866e1180	Share HTTP API boolean query schema (#26615 )	2026-05-09 21:41:15 -04:00
opencode-agent[bot]	6d130e5deb	chore: generate	2026-05-10 01:13:41 +00:00
Kit LangtonandGitHub	e30d8173c1	Fix OpenAPI workspace query drift (#26609 )	2026-05-09 21:12:34 -04:00
opencode	7a79f3a5ea	sync release versions for v1.14.45	2026-05-10 00:07:24 +00:00
Kit Langton GitHub Developer	b8ca71d309	fix(task): subagent inherits parent agent's deny rules (Plan Mode security bypass) (#26597 ) Co-authored-by: Developer <temp@example.com>	2026-05-09 23:51:55 +00:00
Kit Langton GitHub Developer	6849f96825	refactor(provider): share model status schema (#26595 ) Co-authored-by: Developer <temp@example.com>	2026-05-09 19:20:31 -04:00
Kit Langton GitHub Developer	00c3248295	fix(config): allow active provider model status (#26592 ) Co-authored-by: Developer <temp@example.com>	2026-05-09 18:59:05 -04:00
opencode-agent[bot]	818b56dbd0	chore: generate	2026-05-09 22:47:54 +00:00
Kit LangtonandGitHub	29b5b24787	fix(tui): aggregate bootstrap request failures TUI bootstrap now reports all parallel fetch failures together instead of losing sibling failures after the first rejection.	2026-05-09 18:46:52 -04:00
Kit LangtonandGitHub	11363170ca	fix(sdk): wrap thrown error bodies in Error SDK throwOnError paths now convert structured response bodies into real Error instances while preserving the original body and status in cause.	2026-05-09 18:46:43 -04:00
Kit LangtonandGitHub	ba9e4b67ed	fix(tool/read): match permission patterns against worktree-relative path Read permission checks now use the same worktree-relative path basis as edit/write/apply_patch, so configured patterns apply consistently.	2026-05-09 18:46:29 -04:00
Kit LangtonandGitHub	bd1029b19f	test(server): cover HttpApi context inheritance Adds regression coverage for request context inheritance in promptAsync and explicit context provisioning in stream bodies.	2026-05-09 18:46:10 -04:00
Kit Langton GitHub Developer	43b51f09d0	fix(httpapi): align runtime query schemas with workspace routing params (#26581 ) Co-authored-by: Developer <temp@example.com>	2026-05-09 16:50:30 -04:00
opencode-agent[bot]	c61ab51886	chore: generate	2026-05-09 20:45:29 +00:00
Kit Langton GitHub Developer	d373c562f2	fix(session): accept legacy summary diffs (#26579 ) Co-authored-by: Developer <temp@example.com>	2026-05-09 16:44:24 -04:00
opencode-agent[bot]	5fa5d876fc	chore: generate	2026-05-09 20:31:32 +00:00
Kit Langton GitHub Developer	805af011c9	test(session): regression test for #26574 + mirror loosening on Vcs.FileDiff (#26578 ) Co-authored-by: Developer <temp@example.com>	2026-05-09 16:30:31 -04:00
opencode-agent[bot]	480aa8b23b	chore: generate	2026-05-09 20:24:28 +00:00
OpeOginniandGitHub	d62442bb5d	fix(sessions): allow optional patch field in diff for migrated sessions (#26574 )	2026-05-09 16:23:28 -04:00
Kit LangtonandGitHub	8602937a37	test(session): cover workspace-routed messages (#26576 )	2026-05-09 16:19:06 -04:00
Kit Langton GitHub Developer	77da433e0a	fix(session): accept routing params in message list (#26569 ) Co-authored-by: Developer <temp@example.com>	2026-05-09 16:02:35 -04:00
opencode-agent[bot]	ad79f3e0cf	chore: generate	2026-05-09 19:59:00 +00:00
Kit Langton GitHub Developer	6c2dfd2f52	fix(tui): guard messages.data in session.sync against undefined (#26566 ) Co-authored-by: Developer <temp@example.com>	2026-05-09 19:58:03 +00:00
SebastianandGitHub	9a8b54fe62	Plugin command API shim (#26564 )	2026-05-09 21:56:49 +02:00
DaxandGitHub	dcdbdb218f	Move schema utilities into core (#26565 )	2026-05-09 19:51:09 +00:00
Kit Langton GitHub Developer	5e49029e70	fix(provider): isolate plugin model mutations (#26561 ) Co-authored-by: Developer <temp@example.com>	2026-05-09 15:47:07 -04:00
opencode	19abadaf27	sync release versions for v1.14.44	2026-05-09 19:33:26 +00:00