docs(llm): fix stale references in protocols/shared.ts and gemini.ts

- subtractTokens JSDoc said the raw payload lives on Usage.native, but that field was renamed to providerMetadata earlier in this PR. - totalTokens JSDoc still described the abandoned "additive" first-pass contract where inputTokens/outputTokens were non-cached / visible only. We landed on inclusive totals; the fallback already covers cache and reasoning. - Removed a duplicate inline comment in Gemini's mapUsage — the function-level comment already explains the visible/reasoning sum and the undefined-when-incomplete rule.
test(llm): re-record golden scenarios against live providers
2026-05-10 22:08:12 -04:00 · 2026-05-10 22:03:02 -04:00 · 2026-05-10 21:42:09 -04:00 · 2026-05-10 20:39:22 -04:00 · 2026-05-10 19:29:41 -04:00 · 2026-05-10 19:15:46 -04:00
383 changed files with 11691 additions and 18786 deletions
@@ -23,7 +23,7 @@ runs:
        fi
    - name: Setup Bun
-      uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
+      uses: oven-sh/setup-bun@v2
      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@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
+      uses: actions/cache@v4
      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@fee1f7d63c2ff003460e3d139729b119787bc349 # v2.2.2
+      uses: actions/create-github-app-token@v2
      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@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        uses: actions/checkout@v4
        with:
          fetch-depth: 0
@@ -12,9 +12,9 @@ jobs:
      contents: read
      issues: write
    steps:
-      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+      - uses: actions/checkout@v4
-      - uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
+      - uses: oven-sh/setup-bun@v2
        with:
          bun-version: latest
@@ -21,7 +21,7 @@ jobs:
    timeout-minutes: 15
    steps:
      - name: Close inactive PRs
-        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+        uses: actions/github-script@v8
        with:
          github-token: ${{ secrets.GITHUB_TOKEN }}
          script: |
@@ -16,7 +16,7 @@ jobs:
    runs-on: ubuntu-latest
    steps:
      - name: Close non-compliant issues and PRs after 2 hours
-        uses: actions/github-script@f28e40c7f34bde8b3046d885e986cb6290c5673b # v7.1.0
+        uses: actions/github-script@v7
        with:
          script: |
            const { data: items } = await github.rest.issues.listForRepo({
@@ -21,18 +21,18 @@ jobs:
      REGISTRY: ghcr.io/${{ github.repository_owner }}
      TAG: "24.04"
    steps:
-      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+      - uses: actions/checkout@v4
      - uses: ./.github/actions/setup-bun
      - name: Set up QEMU
-        uses: docker/setup-qemu-action@c7c53464625b32c7a7e944ae62b3e17d2b600130 # v3.7.0
+        uses: docker/setup-qemu-action@v3
      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@8d2750c68a42422c14e847fe6c8ac0403b4cbd6f # v3.12.0
+        uses: docker/setup-buildx-action@v3
      - name: Login to GHCR
-        uses: docker/login-action@c94ce9fb468520275223c153574b00df6fe4bcc9 # v3.7.0
+        uses: docker/login-action@v3
        with:
          registry: ghcr.io
          username: ${{ github.repository_owner }}
@@ -13,11 +13,11 @@ jobs:
  deploy:
    runs-on: ubuntu-latest
    steps:
-      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
+      - uses: actions/checkout@v3
      - uses: ./.github/actions/setup-bun
-      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
+      - uses: actions/setup-node@v4
        with:
          node-version: "24"
@@ -16,7 +16,7 @@ jobs:
      contents: write
    steps:
      - name: Checkout repository
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        uses: actions/checkout@v4
        with:
          persist-credentials: false
          fetch-depth: 0
@@ -18,7 +18,7 @@ jobs:
      pull-requests: write
    steps:
      - name: Checkout repository
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        uses: actions/checkout@v4
        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@2c14fc5586fe0b88e5c04732d2e846769cc35671 # latest
+        uses: sst/opencode/github@latest
        env:
          OPENCODE_API_KEY: ${{ secrets.OPENCODE_API_KEY }}
        with:
@@ -13,7 +13,7 @@ jobs:
      issues: write
    steps:
      - name: Checkout repository
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        uses: actions/checkout@v4
        with:
          fetch-depth: 1
@@ -125,7 +125,7 @@ jobs:
      issues: write
    steps:
      - name: Checkout repository
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        uses: actions/checkout@v4
        with:
          fetch-depth: 1
@@ -13,7 +13,7 @@ jobs:
      pull-requests: write
    steps:
      - name: Checkout repository
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        uses: actions/checkout@v4
      - name: Setup Bun
        uses: ./.github/actions/setup-bun
@@ -20,10 +20,10 @@ jobs:
    timeout-minutes: 15
    steps:
      - name: Checkout repository
-        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+        uses: actions/checkout@v6
      - name: Setup Nix
-        uses: nixbuild/nix-quick-install-action@2c9db80fb984ceb1bcaa77cdda3fdf8cfba92035 # v34
+        uses: nixbuild/nix-quick-install-action@v34
      - name: Evaluate flake outputs (all systems)
        run: |
@@ -41,10 +41,10 @@ jobs:
    steps:
      - name: Checkout repository
-        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+        uses: actions/checkout@v6
      - name: Setup Nix
-        uses: nixbuild/nix-quick-install-action@2c9db80fb984ceb1bcaa77cdda3fdf8cfba92035 # v34
+        uses: nixbuild/nix-quick-install-action@v34
      - name: Compute node_modules hash
        id: hash
@@ -72,7 +72,7 @@ jobs:
          echo "Computed hash for ${SYSTEM}: $HASH"
      - name: Upload hash
-        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        uses: actions/upload-artifact@v4
        with:
          name: hash-${{ matrix.system }}
          path: hash.txt
@@ -85,7 +85,7 @@ jobs:
    steps:
      - name: Checkout repository
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        uses: actions/checkout@v4
        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@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
+        uses: actions/download-artifact@v4
        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@24d166886aee4646d448c8a389ff9e1ebcab3682 # v1.20.0
+        uses: SethCohen/github-releases-to-discord@v1
        with:
          webhook_url: ${{ secrets.DISCORD_WEBHOOK }}
@@ -21,12 +21,12 @@ jobs:
      issues: read
    steps:
      - name: Checkout repository
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        uses: actions/checkout@v4
      - uses: ./.github/actions/setup-bun
      - name: Run opencode
-        uses: anomalyco/opencode/github@2c14fc5586fe0b88e5c04732d2e846769cc35671 # latest
+        uses: anomalyco/opencode/github@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@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        uses: actions/checkout@v4
        with:
          fetch-depth: 1
@@ -78,7 +78,7 @@ jobs:
      issues: write
    steps:
      - name: Add Contributor Label
-        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+        uses: actions/github-script@v8
        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@f28e40c7f34bde8b3046d885e986cb6290c5673b # v7.1.0
+        uses: actions/github-script@v7
        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@f28e40c7f34bde8b3046d885e986cb6290c5673b # v7.1.0
+        uses: actions/github-script@v7
        with:
          script: |
            const pr = context.payload.pull_request;
@@ -16,7 +16,7 @@ jobs:
  publish:
    runs-on: blacksmith-4vcpu-ubuntu-2404
    steps:
-      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
+      - uses: actions/checkout@v3
        with:
          fetch-depth: 0
@@ -15,7 +15,7 @@ jobs:
  publish:
    runs-on: blacksmith-4vcpu-ubuntu-2404
    steps:
-      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
+      - uses: actions/checkout@v3
        with:
          fetch-depth: 0
@@ -35,7 +35,7 @@ jobs:
    runs-on: blacksmith-4vcpu-ubuntu-2404
    if: github.repository == 'anomalyco/opencode'
    steps:
-      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
+      - uses: actions/checkout@v3
        with:
          fetch-depth: 0
@@ -72,7 +72,7 @@ jobs:
    runs-on: blacksmith-4vcpu-ubuntu-2404
    if: github.repository == 'anomalyco/opencode'
    steps:
-      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
+      - uses: actions/checkout@v3
        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@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+      - uses: actions/upload-artifact@v4
        with:
          name: opencode-cli
          path: |
            packages/opencode/dist/opencode-darwin*
            packages/opencode/dist/opencode-linux*
-      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+      - uses: actions/upload-artifact@v4
        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@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
+      - uses: actions/checkout@v3
-      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
+      - uses: actions/download-artifact@v4
        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@a457da9ea143d694b1b9c7c869ebb04ebe844ef5 # v2.3.0
+        uses: azure/login@v2
        with:
          client-id: ${{ env.AZURE_CLIENT_ID }}
          tenant-id: ${{ env.AZURE_TENANT_ID }}
          subscription-id: ${{ env.AZURE_SUBSCRIPTION_ID }}
-      - uses: azure/artifact-signing-action@b443cf8ea4124818d2ea9f043cba29fc3ec47b16 # v1.2.0
+      - uses: azure/artifact-signing-action@v1
        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@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+      - uses: actions/upload-artifact@v4
        with:
          name: opencode-cli-signed-windows
          path: |
@@ -249,9 +249,9 @@ jobs:
            platform_flag: --linux
    runs-on: ${{ matrix.settings.host }}
    steps:
-      - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
+      - uses: actions/checkout@v3
-      - uses: apple-actions/import-codesign-certs@8f3fb608891dd2244cdab3d69cd68c0d37a7fe93 # v2.0.0
+      - uses: apple-actions/import-codesign-certs@v2
        if: runner.os == 'macOS'
        with:
          keychain: build
@@ -268,19 +268,19 @@ jobs:
      - name: Azure login
        if: runner.os == 'Windows'
-        uses: azure/login@a457da9ea143d694b1b9c7c869ebb04ebe844ef5 # v2.3.0
+        uses: azure/login@v2
        with:
          client-id: ${{ env.AZURE_CLIENT_ID }}
          tenant-id: ${{ env.AZURE_TENANT_ID }}
          subscription-id: ${{ env.AZURE_SUBSCRIPTION_ID }}
-      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
+      - uses: actions/setup-node@v4
        with:
          node-version: "24"
      - name: Cache apt packages
        if: contains(matrix.settings.host, 'ubuntu')
-        uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
+        uses: actions/cache@v4
        with:
          path: ~/apt-cache
          key: ${{ runner.os }}-${{ matrix.settings.target }}-apt-electron-${{ hashFiles('.github/workflows/publish.yml') }}
@@ -388,12 +388,12 @@ jobs:
            }
          }
-      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+      - uses: actions/upload-artifact@v4
        with:
          name: opencode-desktop-${{ matrix.settings.target }}
          path: packages/desktop/dist/*
-      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+      - uses: actions/upload-artifact@v4
        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@f43a0e5ff2bd294095638e18286ca9a3d1956744 # v3.6.0
+      - uses: actions/checkout@v3
      - uses: ./.github/actions/setup-bun
      - name: Login to GitHub Container Registry
-        uses: docker/login-action@c94ce9fb468520275223c153574b00df6fe4bcc9 # v3.7.0
+        uses: docker/login-action@v3
        with:
          registry: ghcr.io
          username: ${{ github.repository_owner }}
          password: ${{ secrets.GITHUB_TOKEN }}
      - name: Set up QEMU
-        uses: docker/setup-qemu-action@c7c53464625b32c7a7e944ae62b3e17d2b600130 # v3.7.0
+        uses: docker/setup-qemu-action@v3
      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@8d2750c68a42422c14e847fe6c8ac0403b4cbd6f # v3.12.0
+        uses: docker/setup-buildx-action@v3
-      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
+      - uses: actions/setup-node@v4
        with:
          node-version: "24"
          registry-url: "https://registry.npmjs.org"
-      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
+      - uses: actions/download-artifact@v4
        with:
          name: opencode-cli
          path: packages/opencode/dist
-      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
+      - uses: actions/download-artifact@v4
        with:
          name: opencode-cli-windows
          path: packages/opencode/dist
-      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
+      - uses: actions/download-artifact@v4
        with:
          name: opencode-cli-signed-windows
          path: packages/opencode/dist
-      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
+      - uses: actions/download-artifact@v4
        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@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
+        uses: actions/cache@v4
        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@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+      - uses: actions/checkout@v4
        with:
          fetch-depth: 0
@@ -25,7 +25,7 @@ jobs:
          fi
      - name: Checkout repository
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        uses: actions/checkout@v4
        with:
          fetch-depth: 1
@@ -16,7 +16,7 @@ jobs:
    steps:
      - name: Checkout
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        uses: actions/checkout@v4
      - 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@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        uses: actions/checkout@v4
      - 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@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+      - uses: actions/checkout@v4
        with:
          fetch-depth: 0
@@ -37,12 +37,12 @@ jobs:
        shell: bash
    steps:
      - name: Checkout repository
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        uses: actions/checkout@v4
        with:
          token: ${{ secrets.GITHUB_TOKEN }}
      - name: Setup Node
-        uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
+        uses: actions/setup-node@v4
        with:
          node-version: "24"
@@ -55,7 +55,7 @@ jobs:
          git config --global user.name "opencode"
      - name: Cache Turbo
-        uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
+        uses: actions/cache@v4
        with:
          path: node_modules/.cache/turbo
          key: turbo-${{ 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@bccf2e31636835cf0874589931c4116687171386 # v6.4.0
+        uses: mikepenz/action-junit-report@v6
        with:
          report_paths: packages/*/.artifacts/unit/junit.xml
          check_name: "unit results (${{ matrix.settings.name }})"
@@ -85,7 +85,7 @@ jobs:
      - name: Upload unit artifacts
        if: always()
-        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        uses: actions/upload-artifact@v4
        with:
          name: unit-${{ matrix.settings.name }}-${{ github.run_attempt }}
          include-hidden-files: true
@@ -111,12 +111,12 @@ jobs:
        shell: bash
    steps:
      - name: Checkout repository
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        uses: actions/checkout@v4
        with:
          token: ${{ secrets.GITHUB_TOKEN }}
      - name: Setup Node
-        uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
+        uses: actions/setup-node@v4
        with:
          node-version: "24"
@@ -131,7 +131,7 @@ jobs:
      - name: Cache Playwright browsers
        id: playwright-cache
-        uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
+        uses: actions/cache@v4
        with:
          path: ${{ github.workspace }}/.playwright-browsers
          key: ${{ runner.os }}-${{ runner.arch }}-playwright-${{ steps.playwright-version.outputs.version }}-chromium
@@ -155,7 +155,7 @@ jobs:
      - name: Upload Playwright artifacts
        if: always()
-        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        uses: actions/upload-artifact@v4
        with:
          name: playwright-${{ matrix.settings.name }}-${{ github.run_attempt }}
          if-no-files-found: ignore
@@ -12,7 +12,7 @@ jobs:
      issues: write
    steps:
      - name: Checkout repository
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        uses: actions/checkout@v4
        with:
          fetch-depth: 1
@@ -12,7 +12,7 @@ jobs:
    runs-on: blacksmith-4vcpu-ubuntu-2404
    steps:
      - name: Checkout repository
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        uses: actions/checkout@v4
      - name: Setup Bun
        uses: ./.github/actions/setup-bun
@@ -1,7 +1,11 @@
 {
  "$schema": "https://opencode.ai/config.json",
  "provider": {},
-  "permission": {},
+  "permission": {
    "edit": {
      "packages/opencode/migration/*": "ask",
    },
  },
  "mcp": {},
  "tools": {
    "github-triage": false,
@@ -1,37 +0,0 @@
 # Deepening
 How to deepen a cluster of shallow modules safely, given its dependencies. Assumes the vocabulary in [LANGUAGE.md](LANGUAGE.md) — **module**, **interface**, **seam**, **adapter**.
 ## Dependency categories
 When assessing a candidate for deepening, classify its dependencies. The category determines how the deepened module is tested across its seam.
 ### 1. In-process
 Pure computation, in-memory state, no I/O. Always deepenable — merge the modules and test through the new interface directly. No adapter needed.
 ### 2. Local-substitutable
 Dependencies that have local test stand-ins (PGLite for Postgres, in-memory filesystem). Deepenable if the stand-in exists. The deepened module is tested with the stand-in running in the test suite. The seam is internal; no port at the module's external interface.
 ### 3. Remote but owned (Ports & Adapters)
 Your own services across a network boundary (microservices, internal APIs). Define a **port** (interface) at the seam. The deep module owns the logic; the transport is injected as an **adapter**. Tests use an in-memory adapter. Production uses an HTTP/gRPC/queue adapter.
 Recommendation shape: _"Define a port at the seam, implement an HTTP adapter for production and an in-memory adapter for testing, so the logic sits in one deep module even though it's deployed across a network."_
 ### 4. True external (Mock)
 Third-party services (Stripe, Twilio, etc.) you don't control. The deepened module takes the external dependency as an injected port; tests provide a mock adapter.
 ## Seam discipline
 - **One adapter means a hypothetical seam. Two adapters means a real one.** Don't introduce a port unless at least two adapters are justified (typically production + test). A single-adapter seam is just indirection.
 - **Internal seams vs external seams.** A deep module can have internal seams (private to its implementation, used by its own tests) as well as the external seam at its interface. Don't expose internal seams through the interface just because tests use them.
 ## Testing strategy: replace, don't layer
 - Old unit tests on shallow modules become waste once tests at the deepened module's interface exist — delete them.
 - Write new tests at the deepened module's interface. The **interface is the test surface**.
 - Tests assert on observable outcomes through the interface, not internal state.
 - Tests should survive internal refactors — they describe behaviour, not implementation. If a test has to change when the implementation changes, it's testing past the interface.
@@ -1,44 +0,0 @@
 # Interface Design
 When the user wants to explore alternative interfaces for a chosen deepening candidate, use this parallel sub-agent pattern. Based on "Design It Twice" (Ousterhout) — your first idea is unlikely to be the best.
 Uses the vocabulary in [LANGUAGE.md](LANGUAGE.md) — **module**, **interface**, **seam**, **adapter**, **leverage**.
 ## Process
 ### 1. Frame the problem space
 Before spawning sub-agents, write a user-facing explanation of the problem space for the chosen candidate:
 - The constraints any new interface would need to satisfy
 - The dependencies it would rely on, and which category they fall into (see [DEEPENING.md](DEEPENING.md))
 - A rough illustrative code sketch to ground the constraints — not a proposal, just a way to make the constraints concrete
 Show this to the user, then immediately proceed to Step 2. The user reads and thinks while the sub-agents work in parallel.
 ### 2. Spawn sub-agents
 Spawn 3+ sub-agents in parallel using the Agent tool. Each must produce a **radically different** interface for the deepened module.
 Prompt each sub-agent with a separate technical brief (file paths, coupling details, dependency category from [DEEPENING.md](DEEPENING.md), what sits behind the seam). The brief is independent of the user-facing problem-space explanation in Step 1. Give each agent a different design constraint:
 - Agent 1: "Minimize the interface — aim for 1–3 entry points max. Maximise leverage per entry point."
 - Agent 2: "Maximise flexibility — support many use cases and extension."
 - Agent 3: "Optimise for the most common caller — make the default case trivial."
 - Agent 4 (if applicable): "Design around ports & adapters for cross-seam dependencies."
 Include both [LANGUAGE.md](LANGUAGE.md) vocabulary and CONTEXT.md vocabulary in the brief so each sub-agent names things consistently with the architecture language and the project's domain language.
 Each sub-agent outputs:
 1. Interface (types, methods, params — plus invariants, ordering, error modes)
 2. Usage example showing how callers use it
 3. What the implementation hides behind the seam
 4. Dependency strategy and adapters (see [DEEPENING.md](DEEPENING.md))
 5. Trade-offs — where leverage is high, where it's thin
 ### 3. Present and compare
 Present designs sequentially so the user can absorb each one, then compare them in prose. Contrast by **depth** (leverage at the interface), **locality** (where change concentrates), and **seam placement**.
 After comparing, give your own recommendation: which design you think is strongest and why. If elements from different designs would combine well, propose a hybrid. Be opinionated — the user wants a strong read, not a menu.
@@ -1,53 +0,0 @@
 # Language
 Shared vocabulary for every suggestion this skill makes. Use these terms exactly — don't substitute "component," "service," "API," or "boundary." Consistent language is the whole point.
 ## Terms
 **Module**
 Anything with an interface and an implementation. Deliberately scale-agnostic — applies equally to a function, class, package, or tier-spanning slice.
 _Avoid_: unit, component, service.
 **Interface**
 Everything a caller must know to use the module correctly. Includes the type signature, but also invariants, ordering constraints, error modes, required configuration, and performance characteristics.
 _Avoid_: API, signature (too narrow — those refer only to the type-level surface).
 **Implementation**
 What's inside a module — its body of code. Distinct from **Adapter**: a thing can be a small adapter with a large implementation (a Postgres repo) or a large adapter with a small implementation (an in-memory fake). Reach for "adapter" when the seam is the topic; "implementation" otherwise.
 **Depth**
 Leverage at the interface — the amount of behaviour a caller (or test) can exercise per unit of interface they have to learn. A module is **deep** when a large amount of behaviour sits behind a small interface. A module is **shallow** when the interface is nearly as complex as the implementation.
 **Seam** _(from Michael Feathers)_
 A place where you can alter behaviour without editing in that place. The _location_ at which a module's interface lives. Choosing where to put the seam is its own design decision, distinct from what goes behind it.
 _Avoid_: boundary (overloaded with DDD's bounded context).
 **Adapter**
 A concrete thing that satisfies an interface at a seam. Describes _role_ (what slot it fills), not substance (what's inside).
 **Leverage**
 What callers get from depth. More capability per unit of interface they have to learn. One implementation pays back across N call sites and M tests.
 **Locality**
 What maintainers get from depth. Change, bugs, knowledge, and verification concentrate at one place rather than spreading across callers. Fix once, fixed everywhere.
 ## Principles
 - **Depth is a property of the interface, not the implementation.** A deep module can be internally composed of small, mockable, swappable parts — they just aren't part of the interface. A module can have **internal seams** (private to its implementation, used by its own tests) as well as the **external seam** at its interface.
 - **The deletion test.** Imagine deleting the module. If complexity vanishes, the module wasn't hiding anything (it was a pass-through). If complexity reappears across N callers, the module was earning its keep.
 - **The interface is the test surface.** Callers and tests cross the same seam. If you want to test _past_ the interface, the module is probably the wrong shape.
 - **One adapter means a hypothetical seam. Two adapters means a real one.** Don't introduce a seam unless something actually varies across it.
 ## Relationships
 - A **Module** has exactly one **Interface** (the surface it presents to callers and tests).
 - **Depth** is a property of a **Module**, measured against its **Interface**.
 - A **Seam** is where a **Module**'s **Interface** lives.
 - An **Adapter** sits at a **Seam** and satisfies the **Interface**.
 - **Depth** produces **Leverage** for callers and **Locality** for maintainers.
 ## Rejected framings
 - **Depth as ratio of implementation-lines to interface-lines** (Ousterhout): rewards padding the implementation. We use depth-as-leverage instead.
 - **"Interface" as the TypeScript `interface` keyword or a class's public methods**: too narrow — interface here includes every fact a caller must know.
 - **"Boundary"**: overloaded with DDD's bounded context. Say **seam** or **interface**.
@@ -1,71 +0,0 @@
 ---
 name: improve-codebase-architecture
 description: Find deepening opportunities in a codebase, informed by the domain language in CONTEXT.md and the decisions in docs/adr/. Use when the user wants to improve architecture, find refactoring opportunities, consolidate tightly-coupled modules, or make a codebase more testable and AI-navigable.
 ---
 # Improve Codebase Architecture
 Surface architectural friction and propose **deepening opportunities** — refactors that turn shallow modules into deep ones. The aim is testability and AI-navigability.
 ## Glossary
 Use these terms exactly in every suggestion. Consistent language is the point — don't drift into "component," "service," "API," or "boundary." Full definitions in [LANGUAGE.md](LANGUAGE.md).
 - **Module** — anything with an interface and an implementation (function, class, package, slice).
 - **Interface** — everything a caller must know to use the module: types, invariants, error modes, ordering, config. Not just the type signature.
 - **Implementation** — the code inside.
 - **Depth** — leverage at the interface: a lot of behaviour behind a small interface. **Deep** = high leverage. **Shallow** = interface nearly as complex as the implementation.
 - **Seam** — where an interface lives; a place behaviour can be altered without editing in place. (Use this, not "boundary.")
 - **Adapter** — a concrete thing satisfying an interface at a seam.
 - **Leverage** — what callers get from depth.
 - **Locality** — what maintainers get from depth: change, bugs, knowledge concentrated in one place.
 Key principles (see [LANGUAGE.md](LANGUAGE.md) for the full list):
 - **Deletion test**: imagine deleting the module. If complexity vanishes, it was a pass-through. If complexity reappears across N callers, it was earning its keep.
 - **The interface is the test surface.**
 - **One adapter = hypothetical seam. Two adapters = real seam.**
 This skill is _informed_ by the project's domain model. The domain language gives names to good seams; ADRs record decisions the skill should not re-litigate.
 ## Process
 ### 1. Explore
 Read the project's domain glossary and any ADRs in the area you're touching first.
 Then use the Agent tool with `subagent_type=Explore` to walk the codebase. Don't follow rigid heuristics — explore organically and note where you experience friction:
 - Where does understanding one concept require bouncing between many small modules?
 - Where are modules **shallow** — interface nearly as complex as the implementation?
 - Where have pure functions been extracted just for testability, but the real bugs hide in how they're called (no **locality**)?
 - Where do tightly-coupled modules leak across their seams?
 - Which parts of the codebase are untested, or hard to test through their current interface?
 Apply the **deletion test** to anything you suspect is shallow: would deleting it concentrate complexity, or just move it? A "yes, concentrates" is the signal you want.
 ### 2. Present candidates
 Present a numbered list of deepening opportunities. For each candidate:
 - **Files** — which files/modules are involved
 - **Problem** — why the current architecture is causing friction
 - **Solution** — plain English description of what would change
 - **Benefits** — explained in terms of locality and leverage, and also in how tests would improve
 **Use CONTEXT.md vocabulary for the domain, and [LANGUAGE.md](LANGUAGE.md) vocabulary for the architecture.** If `CONTEXT.md` defines "Order," talk about "the Order intake module" — not "the FooBarHandler," and not "the Order service."
 **ADR conflicts**: if a candidate contradicts an existing ADR, only surface it when the friction is real enough to warrant revisiting the ADR. Mark it clearly (e.g. _"contradicts ADR-0007 — but worth reopening because…"_). Don't list every theoretical refactor an ADR forbids.
 Do NOT propose interfaces yet. Ask the user: "Which of these would you like to explore?"
 ### 3. Grilling loop
 Once the user picks a candidate, drop into a grilling conversation. Walk the design tree with them — constraints, dependencies, the shape of the deepened module, what sits behind the seam, what tests survive.
 Side effects happen inline as decisions crystallize:
 - **Naming a deepened module after a concept not in `CONTEXT.md`?** Add the term to `CONTEXT.md` — same discipline as `/grill-with-docs` (see [CONTEXT-FORMAT.md](../grill-with-docs/CONTEXT-FORMAT.md)). Create the file lazily if it doesn't exist.
 - **Sharpening a fuzzy term during the conversation?** Update `CONTEXT.md` right there.
 - **User rejects the candidate with a load-bearing reason?** Offer an ADR, framed as: _"Want me to record this as an ADR so future architecture reviews don't re-suggest it?"_ Only offer when the reason would actually be needed by a future explorer to avoid re-suggesting the same thing — skip ephemeral reasons ("not worth it right now") and self-evident ones. See [ADR-FORMAT.md](../grill-with-docs/ADR-FORMAT.md).
 - **Want to explore alternative interfaces for the deepened module?** See [INTERFACE-DESIGN.md](INTERFACE-DESIGN.md).
@@ -29,7 +29,7 @@
    },
    "packages/app": {
      "name": "@opencode-ai/app",
-      "version": "1.14.48",
+      "version": "1.14.46",
      "dependencies": {
        "@kobalte/core": "catalog:",
        "@opencode-ai/core": "workspace:*",
@@ -85,7 +85,7 @@
    },
    "packages/console/app": {
      "name": "@opencode-ai/console-app",
-      "version": "1.14.48",
+      "version": "1.14.46",
      "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": "1.14.48",
+      "version": "1.14.46",
      "dependencies": {
        "@aws-sdk/client-sts": "3.782.0",
        "@jsx-email/render": "1.1.1",
@@ -147,15 +147,17 @@
    },
    "packages/console/function": {
      "name": "@opencode-ai/console-function",
-      "version": "1.14.48",
+      "version": "1.14.46",
      "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": {
@@ -169,7 +171,7 @@
    },
    "packages/console/mail": {
      "name": "@opencode-ai/console-mail",
-      "version": "1.14.48",
+      "version": "1.14.46",
      "dependencies": {
        "@jsx-email/all": "2.2.3",
        "@jsx-email/cli": "1.4.3",
@@ -193,7 +195,7 @@
    },
    "packages/core": {
      "name": "@opencode-ai/core",
-      "version": "1.14.48",
+      "version": "1.14.46",
      "bin": {
        "opencode": "./bin/opencode",
      },
@@ -227,7 +229,7 @@
    },
    "packages/desktop": {
      "name": "@opencode-ai/desktop",
-      "version": "1.14.48",
+      "version": "1.14.46",
      "dependencies": {
        "drizzle-orm": "catalog:",
        "effect": "catalog:",
@@ -281,7 +283,7 @@
    },
    "packages/enterprise": {
      "name": "@opencode-ai/enterprise",
-      "version": "1.14.48",
+      "version": "1.14.46",
      "dependencies": {
        "@opencode-ai/core": "workspace:*",
        "@opencode-ai/ui": "workspace:*",
@@ -311,7 +313,7 @@
    },
    "packages/function": {
      "name": "@opencode-ai/function",
-      "version": "1.14.48",
+      "version": "1.14.46",
      "dependencies": {
        "@octokit/auth-app": "8.0.1",
        "@octokit/rest": "catalog:",
@@ -327,7 +329,7 @@
    },
    "packages/http-recorder": {
      "name": "@opencode-ai/http-recorder",
-      "version": "1.14.48",
+      "version": "1.14.46",
      "dependencies": {
        "@effect/platform-node": "catalog:",
        "effect": "catalog:",
@@ -340,7 +342,7 @@
    },
    "packages/llm": {
      "name": "@opencode-ai/llm",
-      "version": "1.14.48",
+      "version": "1.14.46",
      "dependencies": {
        "@smithy/eventstream-codec": "4.2.14",
        "@smithy/util-utf8": "4.2.2",
@@ -358,7 +360,7 @@
    },
    "packages/opencode": {
      "name": "opencode",
-      "version": "1.14.48",
+      "version": "1.14.46",
      "bin": {
        "opencode": "./bin/opencode",
      },
@@ -396,7 +398,6 @@
        "@octokit/graphql": "9.0.2",
        "@octokit/rest": "catalog:",
        "@openauthjs/openauth": "catalog:",
        "@opencode-ai/llm": "workspace:*",
        "@opencode-ai/plugin": "workspace:*",
        "@opencode-ai/script": "workspace:*",
        "@opencode-ai/sdk": "workspace:*",
@@ -495,7 +496,7 @@
    },
    "packages/plugin": {
      "name": "@opencode-ai/plugin",
-      "version": "1.14.48",
+      "version": "1.14.46",
      "dependencies": {
        "@opencode-ai/sdk": "workspace:*",
        "effect": "catalog:",
@@ -533,7 +534,7 @@
    },
    "packages/sdk/js": {
      "name": "@opencode-ai/sdk",
-      "version": "1.14.48",
+      "version": "1.14.46",
      "dependencies": {
        "cross-spawn": "catalog:",
      },
@@ -548,7 +549,7 @@
    },
    "packages/slack": {
      "name": "@opencode-ai/slack",
-      "version": "1.14.48",
+      "version": "1.14.46",
      "dependencies": {
        "@opencode-ai/sdk": "workspace:*",
        "@slack/bolt": "^3.17.1",
@@ -583,7 +584,7 @@
    },
    "packages/ui": {
      "name": "@opencode-ai/ui",
-      "version": "1.14.48",
+      "version": "1.14.46",
      "dependencies": {
        "@kobalte/core": "catalog:",
        "@opencode-ai/core": "workspace:*",
@@ -632,7 +633,7 @@
    },
    "packages/web": {
      "name": "@opencode-ai/web",
-      "version": "1.14.48",
+      "version": "1.14.46",
      "dependencies": {
        "@astrojs/cloudflare": "12.6.3",
        "@astrojs/markdown-remark": "6.3.1",
@@ -685,8 +686,8 @@
  },
  "catalog": {
    "@cloudflare/workers-types": "4.20251008.0",
-    "@effect/opentelemetry": "4.0.0-beta.65",
+    "@effect/opentelemetry": "4.0.0-beta.57",
-    "@effect/platform-node": "4.0.0-beta.65",
+    "@effect/platform-node": "4.0.0-beta.57",
    "@hono/zod-validator": "0.4.2",
    "@kobalte/core": "0.13.11",
    "@lydell/node-pty": "1.2.0-beta.10",
@@ -719,7 +720,7 @@
    "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.65",
+    "effect": "4.0.0-beta.59",
    "fuzzysort": "3.1.0",
    "hono": "4.10.7",
    "hono-openapi": "1.1.2",
@@ -1080,11 +1081,11 @@
    "@drizzle-team/brocli": ["@drizzle-team/brocli@0.11.0", "", {}, "sha512-hD3pekGiPg0WPCCGAZmusBBJsDqGUR66Y452YgQsZOnkdQ7ViEPKuyP4huUGEZQefp8g34RRodXYmJ2TbCH+tg=="],
-    "@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/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/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": ["@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-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=="],
+    "@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=="],
    "@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=="],
@@ -1234,6 +1235,8 @@
    "@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=="],
@@ -3038,7 +3041,7 @@
    "ee-first": ["ee-first@1.1.1", "", {}, "sha512-WMwm9LhRUo+WUaRN+vRuETqG89IgZphVSNkdFgeb6sS/E4OrDIN7t48CAewSHXc6C8lefD8KKfr5vY61brQlow=="],
-    "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=="],
+    "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=="],
    "ejs": ["ejs@3.1.10", "", { "dependencies": { "jake": "^10.8.5" }, "bin": { "ejs": "bin/cli.js" } }, "sha512-UeJmFfOrAQS8OJWPZ4qtgHyWExa088/MtK5UEyoJGFH67cDEXkZSviOiKRCZ4Xij0zxI3JECgYs3oKx+AizQBA=="],
@@ -5466,6 +5469,8 @@
    "@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=="],
@@ -1,8 +1,6 @@
 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`,
@@ -27,16 +25,6 @@ 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" },
@@ -44,26 +32,21 @@ const modelHttpErrorsQuery = (product: "go" | "zen") => {
    { column: "user_agent", op: "contains", value: "opencode" },
    { column: "isGoTier", op: "=", value: product === "go" ? "true" : "false" },
  ]
-  const failedHttpStatus = calculatedField({
+
  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($status, "429"))), 1, 0)`
-        : `IF(AND(EQUALS($status, "429"), $isFreeTier), 0, AND(GTE($status, "400"), NOT(EQUALS($status, "401"))), 1, 0)`,
+            : `IF(AND(GTE($status, "400"), NOT(EQUALS($status, "401"))), 1, 0)`,
-  })
+      },
-
+    ],
  return honeycomb.getQuerySpecificationOutput({
    breakdowns: ["model"],
    calculatedFields: [failedHttpStatus],
    calculations: [
      { op: "COUNT", name: "TOTAL", filterCombination: "AND", filters },
-      {
+      { op: "SUM", name: "FAILED", column: "is_failed_http_status", filterCombination: "AND", filters },
        op: "SUM",
        name: "FAILED",
        column: failedHttpStatus.name,
        filterCombination: "AND",
        filters,
      },
    ],
    formulas: [{ name: "ERROR", expression: "IF(GTE($TOTAL, 100), DIV($FAILED, $TOTAL), 0)" }],
    timeRange: 900,
@@ -76,30 +59,31 @@ const providerHttpErrorsQuery = (product: "go" | "zen") => {
    { column: "user_agent", op: "contains", value: "opencode" },
    { column: "isGoTier", op: "=", value: product === "go" ? "true" : "false" },
  ]
  const successHttpStatus = calculatedField({
    name: "is_success_http_status",
    expression: `IF(AND(GTE($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: [successHttpStatus, failedProviderHttpStatus],
+    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)`,
      },
    ],
    calculations: [
      {
        op: "SUM",
        name: "SUCCESS",
-        column: successHttpStatus.name,
+        column: "is_success_http_status",
        filterCombination: "AND",
        filters: [...filters, { column: "event_type", op: "=", value: "completions" }],
      },
      {
        op: "SUM",
        name: "FAILED",
-        column: failedProviderHttpStatus.name,
+        column: "is_failed_provider_http_status",
        filterCombination: "AND",
        filters: [...filters, { column: "event_type", op: "=", value: "llm.error" }],
      },
@@ -111,6 +95,8 @@ 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-Q9r1S15YL9LQK7DRhuOpw3Fxi24BPovEM995GZJayKw=",
+    "x86_64-linux": "sha256-baGxh+hk/rPhg0xI/OdMDz6dPwncgercYNBdTPnLX9o=",
-    "aarch64-linux": "sha256-C0rRTLnxxuuEkCBc3JZbkR66TUVwpcPFif3BU9GRAuA=",
+    "aarch64-linux": "sha256-VTWKq679B3Q4ZnAoQzC4VSCYA09wWecNJ+JajvjNB1U=",
-    "aarch64-darwin": "sha256-1HvalOO/pOkRlYH8CZ93psapt90C+pYzui1JCadBE1Q=",
+    "aarch64-darwin": "sha256-orf2zIBMTiiQrt/6qCzE+o0oKhv6u8zXF9DH1Bo3lbo=",
-    "x86_64-darwin": "sha256-RrndyLWfhWm4mZ88XytFF2NI+ly8la550Z5LBN/g5u4="
+    "x86_64-darwin": "sha256-1MZC1fadRoY4lhkmjlcUQTLYH9Q8pDI1bxd5f94f1xU="
  }
 }
@@ -28,8 +28,8 @@
      "packages/slack"
    ],
    "catalog": {
-      "@effect/opentelemetry": "4.0.0-beta.65",
+      "@effect/opentelemetry": "4.0.0-beta.57",
-      "@effect/platform-node": "4.0.0-beta.65",
+      "@effect/platform-node": "4.0.0-beta.57",
      "@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.65",
+      "effect": "4.0.0-beta.59",
      "ai": "6.0.168",
      "cross-spawn": "7.0.6",
      "hono": "4.10.7",
@@ -1,6 +1,6 @@
 {
  "name": "@opencode-ai/app",
-  "version": "1.14.48",
+  "version": "1.14.46",
  "description": "",
  "type": "module",
  "exports": {
@@ -107,8 +107,7 @@ function createCommandEntries(props: {
  const allowed = createMemo(() => {
    if (props.filesOnly()) return []
    return props.command.options.filter(
-      (option) =>
+      (option) => !option.disabled && !option.id.startsWith("suggested.") && option.id !== "file.open",
        !option.disabled && !option.hidden && !option.id.startsWith("suggested.") && option.id !== "file.open",
    )
  })
@@ -6,8 +6,7 @@ import { Dialog } from "@opencode-ai/ui/dialog"
 import { List } from "@opencode-ai/ui/list"
 import { Switch } from "@opencode-ai/ui/switch"
 import { useLanguage } from "@/context/language"
-import { useQueryOptions } from "@/context/global-sync"
+import { mcpQueryKey } from "@/context/global-sync"
 import { pathKey } from "@/utils/path-key"
 const statusLabels = {
  connected: "mcp.status.connected",
@@ -21,7 +20,6 @@ export const DialogSelectMcp: Component = () => {
  const sdk = useSDK()
  const language = useLanguage()
  const queryClient = useQueryClient()
  const queryOptions = useQueryOptions()
  const items = createMemo(() =>
    Object.entries(sync.data.mcp ?? {})
@@ -34,7 +32,7 @@ export const DialogSelectMcp: Component = () => {
      if (sync.data.mcp[name]?.status === "connected") await sdk.client.mcp.disconnect({ name })
      else await sdk.client.mcp.connect({ name })
    },
-    onSuccess: () => queryClient.refetchQueries(queryOptions.mcp(pathKey(sync.directory))),
+    onSuccess: () => queryClient.refetchQueries({ queryKey: mcpQueryKey(sync.directory) }),
  }))
  const enabledCount = createMemo(() => items().filter((i) => i.status === "connected").length)
@@ -16,6 +16,7 @@ import {
 } from "@/context/prompt"
 import { useLayout } from "@/context/layout"
 import { useSDK } from "@/context/sdk"
 import { useGlobalSDK } from "@/context/global-sdk"
 import { useSync } from "@/context/sync"
 import { useComments } from "@/context/comments"
 import { Button } from "@opencode-ai/ui/button"
@@ -55,8 +56,7 @@ import { PromptDragOverlay } from "./prompt-input/drag-overlay"
 import { promptPlaceholder } from "./prompt-input/placeholder"
 import { ImagePreview } from "@opencode-ai/ui/image-preview"
 import { useQueries } from "@tanstack/solid-query"
-import { useQueryOptions } from "@/context/global-sync"
+import { loadAgentsQuery, loadProvidersQuery } from "@/context/global-sync/bootstrap"
 import { pathKey } from "@/utils/path-key"
 interface PromptInputProps {
  class?: string
@@ -103,7 +103,7 @@ const NON_EMPTY_TEXT = /[^\s\u200B]/
 export const PromptInput: Component<PromptInputProps> = (props) => {
  const sdk = useSDK()
-  const queryOptions = useQueryOptions()
+  const globalSDK = useGlobalSDK()
  const sync = useSync()
  const local = useLocal()
@@ -1256,9 +1256,9 @@ export const PromptInput: Component<PromptInputProps> = (props) => {
  const [agentsQuery, globalProvidersQuery, providersQuery] = useQueries(() => ({
    queries: [
-      queryOptions.agents(pathKey(sdk.directory)),
+      loadAgentsQuery(sdk.directory, sdk.client),
-      queryOptions.providers(null),
+      loadProvidersQuery(null, globalSDK.client),
-      queryOptions.providers(pathKey(sdk.directory)),
+      loadProvidersQuery(sdk.directory, sdk.client),
    ],
  }))
@@ -123,13 +123,11 @@ function listFor(command: CommandContext, map: KeybindMap, palette: string) {
  for (const opt of command.catalog) {
    if (opt.id.startsWith("suggested.")) continue
    if (opt.hidden) continue
    out.set(opt.id, { title: opt.title, group: groupFor(opt.id) })
  }
  for (const opt of command.options) {
    if (opt.id.startsWith("suggested.")) continue
    if (opt.hidden) continue
    out.set(opt.id, { title: opt.title, group: groupFor(opt.id) })
  }
@@ -15,8 +15,7 @@ import { useSDK } from "@/context/sdk"
 import { normalizeServerUrl, ServerConnection, useServer } from "@/context/server"
 import { useSync } from "@/context/sync"
 import { useCheckServerHealth, type ServerHealth } from "@/utils/server-health"
-import { useQueryOptions } from "@/context/global-sync"
+import { mcpQueryKey } from "@/context/global-sync"
 import { pathKey } from "@/utils/path-key"
 const pollMs = 10_000
@@ -140,14 +139,13 @@ const useMcpToggleMutation = () => {
  const sdk = useSDK()
  const language = useLanguage()
  const queryClient = useQueryClient()
  const queryOptions = useQueryOptions()
  return useMutation(() => ({
    mutationFn: async (name: string) => {
      const status = sync.data.mcp[name]
      await (status?.status === "connected" ? sdk.client.mcp.disconnect({ name }) : sdk.client.mcp.connect({ name }))
    },
-    onSuccess: () => queryClient.refetchQueries(queryOptions.mcp(pathKey(sync.directory))),
+    onSuccess: () => queryClient.refetchQueries({ queryKey: mcpQueryKey(sync.directory) }),
    onError: (err) => {
      showToast({
        variant: "error",
@@ -81,7 +81,6 @@ export interface CommandOption {
  slash?: string
  suggested?: boolean
  disabled?: boolean
  hidden?: boolean
  onSelect?: (source?: "palette" | "keybind" | "slash") => void
  onHighlight?: () => (() => void) | void
 }
@@ -94,7 +93,6 @@ export type CommandCatalogItem = {
  category?: string
  keybind?: KeybindConfig
  slash?: string
  hidden?: boolean
 }
 export type CommandRegistration = {
@@ -281,7 +279,6 @@ export const { use: useCommand, provider: CommandProvider } = createSimpleContex
      setCatalog(
        registered().reduce((acc, opt) => {
          const id = actionId(opt.id)
          if (opt.title)
          acc[id] = {
            title: opt.title,
            description: opt.description,
@@ -18,10 +18,8 @@ import {
  bootstrapDirectory,
  bootstrapGlobal,
  clearProviderRev,
  loadAgentsQuery,
  loadGlobalConfigQuery,
  loadPathQuery,
  loadProjectsQuery,
  loadProvidersQuery,
 } from "./global-sync/bootstrap"
 import { createChildStoreManager } from "./global-sync/child-store"
@@ -35,7 +33,6 @@ import { formatServerError } from "@/utils/server-errors"
 import { queryOptions, useMutation, useQueries, useQuery, useQueryClient } from "@tanstack/solid-query"
 import { createRefreshQueue } from "./global-sync/queue"
 import { directoryKey } from "./global-sync/utils"
 import { PathKey } from "@/utils/path-key"
 type GlobalStore = {
  ready: boolean
@@ -51,33 +48,24 @@ type GlobalStore = {
  reload: undefined | "pending" | "complete"
 }
 export const loadSessionsQueryKey = (directory: string) => [directory, "loadSessions"] as const
 export const mcpQueryKey = (directory: string) => [directory, "mcp"] as const
 export const loadMcpQuery = (directory: string, sdk: OpencodeClient) =>
  queryOptions({
-    queryKey: [directory, "mcp"] as const,
+    queryKey: mcpQueryKey(directory),
    queryFn: () => sdk.mcp.status().then((r) => r.data ?? {}),
  })
 export const lspQueryKey = (directory: string) => [directory, "lsp"] as const
 export const loadLspQuery = (directory: string, sdk: OpencodeClient) =>
  queryOptions({
-    queryKey: [directory, "lsp"] as const,
+    queryKey: lspQueryKey(directory),
    queryFn: () => sdk.lsp.status().then((r) => r.data ?? []),
  })
 function makeQueryOptionsApi(globalSDK: () => OpencodeClient, sdkFor: (dir: PathKey) => OpencodeClient) {
  return {
    globalConfig: () => loadGlobalConfigQuery(globalSDK()),
    projects: () => loadProjectsQuery(globalSDK()),
    providers: (directory: PathKey | null) =>
      loadProvidersQuery(directory, directory === null ? globalSDK() : sdkFor(directory)),
    path: (directory: PathKey | null) => loadPathQuery(directory, directory === null ? globalSDK() : sdkFor(directory)),
    agents: (directory: PathKey) => loadAgentsQuery(directory, sdkFor(directory)),
    mcp: (directory: PathKey) => loadMcpQuery(directory, sdkFor(directory)),
    lsp: (directory: PathKey) => loadLspQuery(directory, sdkFor(directory)),
    sessions: (directory: PathKey) => ({ queryKey: [directory, "loadSessions"] as const }),
  }
 }
 export type QueryOptionsApi = ReturnType<typeof makeQueryOptionsApi>
 function createGlobalSync() {
  const globalSDK = useGlobalSDK()
  const language = useLanguage()
@@ -89,22 +77,12 @@ function createGlobalSync() {
  const sessionLoads = new Map<string, Promise<void>>()
  const sessionMeta = new Map<string, { limit: number }>()
  const sdkFor = (directory: string) => {
    const key = directoryKey(directory)
    const cached = sdkCache.get(key)
    if (cached) return cached
    const sdk = globalSDK.createClient({
      directory,
      throwOnError: true,
    })
    sdkCache.set(key, sdk)
    return sdk
  }
  const queryOptionsApi = makeQueryOptionsApi(() => globalSDK.client, sdkFor)
  const [configQuery, providerQuery, pathQuery] = useQueries(() => ({
-    queries: [queryOptionsApi.globalConfig(), queryOptionsApi.providers(null), queryOptionsApi.path(null)],
+    queries: [
      loadGlobalConfigQuery(globalSDK.client),
      loadProvidersQuery(null, globalSDK.client),
      loadPathQuery(null, globalSDK.client),
    ],
  }))
  const [globalStore, setGlobalStore] = createStore<GlobalStore>({
@@ -203,6 +181,18 @@ function createGlobalSync() {
    bootstrapInstance,
  })
  const sdkFor = (directory: string) => {
    const key = directoryKey(directory)
    const cached = sdkCache.get(key)
    if (cached) return cached
    const sdk = globalSDK.createClient({
      directory,
      throwOnError: true,
    })
    sdkCache.set(key, sdk)
    return sdk
  }
  const children = createChildStoreManager({
    owner,
    isBooting: (directory) => booting.has(directory),
@@ -219,7 +209,7 @@ function createGlobalSync() {
      clearSessionPrefetchDirectory(key)
    },
    translate: language.t,
-    queryOptions: queryOptionsApi,
+    getSdk: sdkFor,
    global: {
      provider: globalStore.provider,
    },
@@ -249,7 +239,7 @@ function createGlobalSync() {
    const limit = Math.max(store.limit + SESSION_RECENT_LIMIT, SESSION_RECENT_LIMIT)
    const promise = queryClient
      .fetchQuery({
-        ...queryOptionsApi.sessions(key),
+        queryKey: loadSessionsQueryKey(key),
        queryFn: () =>
          loadRootSessionsWithFallback({
            directory,
@@ -378,7 +368,7 @@ function createGlobalSync() {
      setSessionTodo,
      vcsCache: children.vcsCache.get(key),
      loadLsp: () => {
-        void queryClient.fetchQuery(queryOptionsApi.lsp(key))
+        void queryClient.fetchQuery(loadLspQuery(key, sdkFor(directory)))
      },
    })
  })
@@ -436,7 +426,6 @@ function createGlobalSync() {
    },
    child: children.child,
    peek: children.peek,
    queryOptions: queryOptionsApi,
    // bootstrap,
    updateConfig: updateConfigMutation.mutateAsync,
    project: projectApi,
@@ -458,7 +447,3 @@ export function useGlobalSync() {
  if (!context) throw new Error("useGlobalSync must be used within GlobalSyncProvider")
  return context
 }
 export function useQueryOptions() {
  return useGlobalSync().queryOptions
 }
@@ -22,7 +22,7 @@ describe("createChildStoreManager", () => {
      onBootstrap() {},
      onDispose() {},
      translate: (key) => key,
-      queryOptions: {} as any,
+      getSdk: () => null!,
      global: { provider: null! },
    })
@@ -1,7 +1,7 @@
 import { createRoot, getOwner, onCleanup, runWithOwner, type Owner } from "solid-js"
 import { createStore, type SetStoreFunction, type Store } from "solid-js/store"
 import { Persist, persisted } from "@/utils/persist"
-import type { ProviderListResponse, VcsInfo } from "@opencode-ai/sdk/v2/client"
+import type { OpencodeClient, ProviderListResponse, VcsInfo } from "@opencode-ai/sdk/v2/client"
 import {
  DIR_IDLE_TTL_MS,
  MAX_DIR_STORES,
@@ -15,7 +15,8 @@ import {
 } from "./types"
 import { canDisposeDirectory, pickDirectoriesToEvict } from "./eviction"
 import { useQueries } from "@tanstack/solid-query"
-import { QueryOptionsApi } from "../global-sync"
+import { loadPathQuery, loadProvidersQuery } from "./bootstrap"
 import { loadLspQuery, loadMcpQuery } from "../global-sync"
 import { directoryKey, type DirectoryKey } from "./utils"
 export function createChildStoreManager(input: {
@@ -25,7 +26,7 @@ export function createChildStoreManager(input: {
  onBootstrap: (directory: string) => void
  onDispose: (directory: string) => void
  translate: (key: string, vars?: Record<string, string | number>) => string
-  queryOptions: QueryOptionsApi
+  getSdk: (directory: string) => OpencodeClient
  global: {
    provider: ProviderListResponse
  }
@@ -170,15 +171,17 @@ export function createChildStoreManager(input: {
      const init = () =>
        createRoot((dispose) => {
          const sdk = input.getSdk(directory)
          const initialMeta = meta[0].value
          const initialIcon = icon[0].value
          const [pathQuery, mcpQuery, lspQuery, providerQuery] = useQueries(() => ({
            queries: [
-              input.queryOptions.path(key),
+              loadPathQuery(key, sdk),
-              input.queryOptions.mcp(key),
+              loadMcpQuery(key, sdk),
-              input.queryOptions.lsp(key),
+              loadLspQuery(key, sdk),
-              input.queryOptions.providers(key),
+              loadProvidersQuery(key, sdk),
            ],
          }))
@@ -228,7 +231,6 @@ export function createChildStoreManager(input: {
            limit: 5,
            message: {},
            part: {},
            part_text_accum_delta: {},
          })
          children[key] = child
          disposers.set(key, dispose)
@@ -81,7 +81,6 @@ const baseState = (input: Partial<State> = {}) =>
    limit: 10,
    message: {},
    part: {},
    part_text_accum_delta: {},
    ...input,
  }) as State
@@ -211,12 +211,6 @@ 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]
        }),
      )
@@ -225,11 +219,6 @@ 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])
@@ -251,11 +240,6 @@ 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)
@@ -279,7 +263,6 @@ 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,
@@ -39,7 +39,6 @@ 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: [] },
@@ -48,14 +47,12 @@ 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()
@@ -73,7 +70,6 @@ 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: {},
@@ -82,7 +78,6 @@ describe("app session cache", () => {
      part: { [m.id]: [part("prt_1", "ses_1", m.id)] },
      permission: {},
      question: {},
      part_text_accum_delta: {},
    }
    dropSessionCaches(store, ["ses_1"])
@@ -18,7 +18,6 @@ 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>) {
@@ -28,9 +27,6 @@ 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]
  }
@@ -72,9 +72,6 @@ export type State = {
  part: {
    [messageID: string]: Part[]
  }
  part_text_accum_delta: {
    [partID: string]: string
  }
 }
 export type VcsCache = {
@@ -43,7 +43,6 @@ type SessionView = {
  reviewOpen?: string[]
  pendingMessage?: string
  pendingMessageAt?: number
  todoCollapsed?: boolean
 }
 type TabHandoff = {
@@ -760,18 +759,6 @@ 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() {
@@ -44,7 +44,7 @@ const migrate = (value: unknown) => {
 }
 const clone = (value: State | undefined) => {
-  if (!value) return
+  if (!value) return undefined
  return {
    ...value,
    model: value.model ? { ...value.model } : undefined,
@@ -104,7 +104,7 @@ export const { use: useLocal, provider: LocalProvider } = createSimpleContext({
    const pickAgent = (name: string | undefined) => {
      const items = list()
-      if (items.length === 0) return
+      if (items.length === 0) return undefined
      return items.find((item) => item.name === name) ?? items[0]
    }
@@ -227,14 +227,14 @@ export const { use: useLocal, provider: LocalProvider } = createSimpleContext({
        () => agent.current()?.model,
        fallback,
      )
-      if (!item) return
+      if (!item) return undefined
      return models.find(item)
    }
    const configured = () => {
      const item = agent.current()
      const model = current()
-      if (!item || !model) return
+      if (!item || !model) return undefined
      return getConfiguredAgentVariant({
        agent: { model: item.model, variant: item.variant },
        model: { providerID: model.provider.id, modelID: model.id, variants: model.variants },
@@ -314,16 +314,11 @@ export const { use: useLocal, provider: LocalProvider } = createSimpleContext({
        configured,
        selected,
        current() {
-          const resolved = resolveModelVariant({
+          return resolveModelVariant({
            variants: this.list(),
            selected: this.selected(),
            configured: this.configured(),
          })
          if (resolved) return resolved
          const model = current()
          if (!model) return
          const saved = models.variant.get({ providerID: model.provider.id, modelID: model.id })
          if (saved && this.list().includes(saved)) return saved
        },
        list() {
          const item = current()
@@ -340,9 +335,6 @@ export const { use: useLocal, provider: LocalProvider } = createSimpleContext({
              variant: value ?? null,
            })
            write({ variant: value ?? null })
            if (model) {
              models.variant.set({ providerID: model.provider.id, modelID: model.id }, value ?? undefined)
            }
          })
        },
        cycle() {
@@ -25,7 +25,6 @@ export const dict = {
  "command.project.open": "Open project",
  "command.project.previous": "Previous project",
  "command.project.next": "Next project",
  "command.project.index": "Switch to project {{index}}",
  "command.provider.connect": "Connect provider",
  "command.server.switch": "Switch server",
  "command.settings.open": "Open settings",
@@ -960,15 +960,6 @@ export default function Layout(props: ParentProps) {
    void openProject(target.worktree)
  }
  function navigateToProjectIndex(index: number) {
    const projects = layout.projects.list()
    const target = projects[index]
    if (!target) return
    globalSync.child(target.worktree)
    void openProject(target.worktree)
  }
  function navigateSessionByUnseen(offset: number) {
    const sessions = currentSessions()
    if (sessions.length === 0) return
@@ -1049,19 +1040,6 @@ export default function Layout(props: ParentProps) {
        keybind: "mod+alt+arrowdown",
        onSelect: () => navigateProjectByOffset(1),
      },
      ...Array.from({ length: 9 }, (_, i) => {
        const index = i
        const number = index + 1
        return {
          id: `project.${number}`,
          category: language.t("command.category.project"),
          title: `Open Project {number}`,
          keybind: `mod+${number}`,
          disabled: layout.projects.list().length <= index,
          hidden: true,
          onSelect: () => navigateToProjectIndex(index),
        }
      }),
      {
        id: "provider.connect",
        title: language.t("command.provider.connect"),
@@ -1431,20 +1409,19 @@ export default function Layout(props: ParentProps) {
    const index = list.findIndex((x) => pathKey(x.worktree) === key)
    const active = pathKey(currentProject()?.worktree ?? "") === key
    if (index === -1) return
    const next = list[index + 1]
    if (!active) {
      layout.projects.close(directory)
      return
    }
-    if (list.length === 1) {
+    if (!next) {
      layout.projects.close(directory)
      navigate("/")
      return
    }
    const next = list[index + 1] ?? list[index - 1]
    navigateWithSidebarReset(`/${base64Encode(next.worktree)}/session`)
    layout.projects.close(directory)
    queueMicrotask(() => {
@@ -1957,7 +1934,7 @@ export default function Layout(props: ParentProps) {
    if (!created?.directory) return
-    setWorkspaceName(created.directory, created.branch ?? getFilename(created.directory), project.id, created.branch)
+    setWorkspaceName(created.directory, created.branch, project.id, created.branch)
    const local = project.worktree
    const key = pathKey(created.directory)
@@ -2119,7 +2096,6 @@ export default function Layout(props: ParentProps) {
              </div>
            </Show>
          }
          keyed
        >
          {(project) => (
            <>
@@ -2130,7 +2106,9 @@ export default function Layout(props: ParentProps) {
                      id={`project:${projectId()}`}
                      value={projectName}
                      onSave={(next) => {
-                        void renameProject(project, next)
+                        const item = project()
                        if (!item) return
                        void renameProject(item, next)
                      }}
                      class="text-14-medium text-text-strong truncate"
                      displayClass="text-14-medium text-text-strong truncate"
@@ -2172,7 +2150,9 @@ export default function Layout(props: ParentProps) {
                      <DropdownMenu.Content class="mt-1">
                        <DropdownMenu.Item
                          onSelect={() => {
-                            showEditProjectDialog(project)
+                            const item = project()
                            if (!item) return
                            showEditProjectDialog(item)
                          }}
                        >
                          <DropdownMenu.ItemLabel>{language.t("common.edit")}</DropdownMenu.ItemLabel>
@@ -2182,7 +2162,9 @@ export default function Layout(props: ParentProps) {
                          data-project={slug()}
                          disabled={!canToggle()}
                          onSelect={() => {
-                            toggleProjectWorkspaces(project)
+                            const item = project()
                            if (!item) return
                            toggleProjectWorkspaces(item)
                          }}
                        >
                          <DropdownMenu.ItemLabel>
@@ -2241,7 +2223,7 @@ export default function Layout(props: ParentProps) {
                      <div class="flex-1 min-h-0">
                        <LocalWorkspace
                          ctx={workspaceSidebarCtx}
-                          project={project}
+                          project={project()}
                          sortNow={sortNow}
                          mobile={panelProps.mobile}
                        />
@@ -2256,7 +2238,9 @@ export default function Layout(props: ParentProps) {
                        icon="plus-small"
                        class="w-full"
                        onClick={() => {
-                          void createWorkspace(project)
+                          const item = project()
                          if (!item) return
                          void createWorkspace(item)
                        }}
                      >
                        {language.t("workspace.new")}
@@ -2283,7 +2267,7 @@ export default function Layout(props: ParentProps) {
                                <SortableWorkspace
                                  ctx={workspaceSidebarCtx}
                                  directory={directory}
-                                  project={project}
+                                  project={project()}
                                  sortNow={sortNow}
                                  mobile={panelProps.mobile}
                                />
@@ -14,7 +14,7 @@ import { Spinner } from "@opencode-ai/ui/spinner"
 import { Tooltip } from "@opencode-ai/ui/tooltip"
 import { type Session } from "@opencode-ai/sdk/v2/client"
 import { type LocalProject } from "@/context/layout"
-import { useGlobalSync, useQueryOptions } from "@/context/global-sync"
+import { loadSessionsQueryKey, useGlobalSync } from "@/context/global-sync"
 import { useLanguage } from "@/context/language"
 import { pathKey } from "@/utils/path-key"
 import { NewSessionItem, SessionItem, SessionSkeleton } from "./sidebar-items"
@@ -300,7 +300,6 @@ export const SortableWorkspace = (props: {
  const navigate = useNavigate()
  const params = useParams()
  const globalSync = useGlobalSync()
  const queryOptions = useQueryOptions()
  const language = useLanguage()
  const sortable = createSortable(props.directory)
  const [workspaceStore, setWorkspaceStore] = globalSync.child(props.directory, { bootstrap: false })
@@ -321,7 +320,7 @@ export const SortableWorkspace = (props: {
  const boot = createMemo(() => open() || active())
  const count = createMemo(() => sessions()?.length ?? 0)
  const hasMore = createMemo(() => workspaceStore.sessionTotal > count())
-  const fetching = useIsFetching(() => queryOptions.sessions(pathKey(props.directory)))
+  const fetching = useIsFetching(() => ({ queryKey: loadSessionsQueryKey(props.directory) }))
  const busy = createMemo(() => props.ctx.isBusy(props.directory))
  const loading = () => fetching() > 0 && count() === 0
  const touch = createMediaQuery("(hover: none)")
@@ -447,7 +446,6 @@ export const LocalWorkspace = (props: {
  mobile?: boolean
 }): JSX.Element => {
  const globalSync = useGlobalSync()
  const queryOptions = useQueryOptions()
  const language = useLanguage()
  const workspace = createMemo(() => {
    const [store, setStore] = globalSync.child(props.project.worktree)
@@ -456,7 +454,7 @@ export const LocalWorkspace = (props: {
  const slug = createMemo(() => base64Encode(props.project.worktree))
  const sessions = createMemo(() => sortedRootSessions(workspace().store, props.sortNow()))
  const count = createMemo(() => sessions()?.length ?? 0)
-  const fetching = useIsFetching(() => queryOptions.sessions(pathKey(props.project.worktree)))
+  const fetching = useIsFetching(() => ({ queryKey: loadSessionsQueryKey(props.project.worktree) }))
  const hasMore = createMemo(() => workspace().store.sessionTotal > count())
  const loading = () => fetching() > 0 && count() === 0
  const loadMore = async () => {
@@ -2,7 +2,6 @@ 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"
@@ -47,12 +46,10 @@ 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))
@@ -210,8 +207,6 @@ 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()}
@@ -42,17 +42,18 @@ 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() }))
@@ -71,7 +72,7 @@ export function SessionTodoDock(props: {
  )
  const preview = createMemo(() => active()?.content ?? "")
-  const collapse = useSpring(() => (props.collapsed ? 1 : 0), { visualDuration: 0.3, bounce: 0 })
+  const collapse = useSpring(() => (store.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())))
@@ -106,11 +107,11 @@ export function SessionTodoDock(props: {
          class="pl-3 pr-2 py-2 flex items-center gap-2 overflow-visible"
          role="button"
          tabIndex={0}
-          onClick={props.onToggle}
+          onClick={toggle}
          onKeyDown={(event) => {
            if (event.key !== "Enter" && event.key !== " ") return
            event.preventDefault()
-            props.onToggle()
+            toggle()
          }}
        >
          <span
@@ -147,7 +148,7 @@ export function SessionTodoDock(props: {
          >
            <TextReveal
              class="text-14-regular text-text-base cursor-default"
-              text={props.collapsed ? preview() : undefined}
+              text={store.collapsed ? preview() : undefined}
              duration={600}
              travel={25}
              edge={17}
@@ -160,7 +161,7 @@ export function SessionTodoDock(props: {
          <div class="ml-auto">
            <IconButton
              data-action="session-todo-toggle-button"
-              data-collapsed={props.collapsed ? "true" : "false"}
+              data-collapsed={store.collapsed ? "true" : "false"}
              icon="chevron-down"
              size="normal"
              variant="ghost"
@@ -171,16 +172,16 @@ export function SessionTodoDock(props: {
              }}
              onClick={(event) => {
                event.stopPropagation()
-                props.onToggle()
+                toggle()
              }}
-              aria-label={props.collapsed ? props.expandLabel : props.collapseLabel}
+              aria-label={store.collapsed ? props.expandLabel : props.collapseLabel}
            />
          </div>
        </div>
        <div
          data-slot="session-todo-list"
-          aria-hidden={props.collapsed || off()}
+          aria-hidden={store.collapsed || off()}
          classList={{
            "pointer-events-none": hide() > 0.1,
          }}
@@ -1,6 +1,6 @@
 {
  "name": "@opencode-ai/console-app",
-  "version": "1.14.48",
+  "version": "1.14.46",
  "type": "module",
  "license": "MIT",
  "scripts": {
@@ -53,7 +53,7 @@ export function UsageSection() {
  }
  const calculateTotalOutputTokens = (u: Awaited<ReturnType<typeof getUsageInfo>>[0]) => {
-    return u.outputTokens
+    return u.outputTokens + (u.reasoningTokens ?? 0)
  }
  const goPrev = async () => {
@@ -47,7 +47,6 @@ 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 = {
@@ -130,8 +129,6 @@ 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(
@@ -145,7 +142,6 @@ export async function handler(
        retry,
        stickyProvider,
        modelTpmLimits,
        modelTpsLimits,
      )
      validateModelSettings(billingSource, authInfo)
      updateProviderKey(authInfo, providerInfo)
@@ -298,17 +294,14 @@ 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": timestampLastByte,
+                  "timestamp.last_byte": Date.now(),
                })
                dataDumper?.flush()
                await rateLimiter?.track()
@@ -318,13 +311,6 @@ 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)
@@ -335,10 +321,10 @@ export async function handler(
              }
              if (responseLength === 0) {
-                timestampFirstByte = Date.now()
+                const now = Date.now()
                logger.metric({
-                  time_to_first_byte: timestampFirstByte - startTimestamp,
+                  time_to_first_byte: now - startTimestamp,
-                  "timestamp.first_byte": timestampFirstByte,
+                  "timestamp.first_byte": now,
                })
              }
@@ -492,7 +478,6 @@ 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
@@ -524,11 +509,6 @@ 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
@@ -909,6 +889,10 @@ 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
@@ -925,11 +909,17 @@ export async function handler(
      return modelCost.cacheWrite1h * cacheWrite1hTokens * 100
    })()
    const totalCostInCent =
-      inputCost + outputCost + (cacheReadCost ?? 0) + (cacheWrite5mCost ?? 0) + (cacheWrite1hCost ?? 0)
+      inputCost +
      outputCost +
      (reasoningCost ?? 0) +
      (cacheReadCost ?? 0) +
      (cacheWrite5mCost ?? 0) +
      (cacheWrite1hCost ?? 0)
    return {
      totalCostInCent,
      inputCost,
      outputCost,
      reasoningCost,
      cacheReadCost,
      cacheWrite5mCost,
      cacheWrite1hCost,
@@ -951,7 +941,8 @@ export async function handler(
  ) {
    const { inputTokens, outputTokens, reasoningTokens, cacheReadTokens, cacheWrite5mTokens, cacheWrite1hTokens } =
      usageInfo
-    const { totalCostInCent, inputCost, outputCost, cacheReadCost, cacheWrite5mCost, cacheWrite1hCost } = costInfo
+    const { totalCostInCent, inputCost, outputCost, reasoningCost, cacheReadCost, cacheWrite5mCost, cacheWrite1hCost } =
      costInfo
    logger.metric({
      "tokens.input": inputTokens,
@@ -962,12 +953,14 @@ 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,
@@ -1,89 +0,0 @@
 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} + ${unqualify}`,
            },
          }),
      )
    },
  }
 }
@@ -50,7 +50,7 @@ export const openaiHelper: ProviderHelper = ({ workspaceID }) => ({
    const cacheReadTokens = usage.input_tokens_details?.cached_tokens ?? undefined
    return {
      inputTokens: inputTokens - (cacheReadTokens ?? 0),
-      outputTokens,
+      outputTokens: outputTokens - (reasoningTokens ?? 0),
      reasoningTokens,
      cacheReadTokens,
      cacheWrite5mTokens: undefined,
@@ -1,7 +0,0 @@
 CREATE TABLE `model_tps_rate_limit` (
 	`id` varchar(255) NOT NULL,
 	`interval` bigint NOT NULL,
 	`qualify` int NOT NULL,
 	`unqualify` int NOT NULL,
 	CONSTRAINT PRIMARY KEY(`id`,`interval`)
 );
@@ -1,7 +1,7 @@
 {
  "$schema": "https://json.schemastore.org/package.json",
  "name": "@opencode-ai/console-core",
-  "version": "1.14.48",
+  "version": "1.14.46",
  "private": true,
  "type": "module",
  "license": "MIT",
@@ -36,7 +36,6 @@ 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(),
@@ -40,14 +40,3 @@ 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": "1.14.48",
+  "version": "1.14.46",
  "$schema": "https://json.schemastore.org/package.json",
  "private": true,
  "type": "module",
@@ -20,10 +20,12 @@
    "@ai-sdk/anthropic": "3.0.64",
    "@ai-sdk/openai": "3.0.48",
    "@ai-sdk/openai-compatible": "2.0.37",
    "@hono/zod-validator": "catalog:",
    "@opencode-ai/console-core": "workspace:*",
    "@opencode-ai/console-resource": "workspace:*",
    "@openauthjs/openauth": "0.0.0-20250322224806",
    "ai": "catalog:",
    "hono": "catalog:",
    "zod": "catalog:"
  }
 }
@@ -1,6 +1,6 @@
 {
  "name": "@opencode-ai/console-mail",
-  "version": "1.14.48",
+  "version": "1.14.46",
  "dependencies": {
    "@jsx-email/all": "2.2.3",
    "@jsx-email/cli": "1.4.3",
@@ -1,6 +1,6 @@
 {
  "$schema": "https://json.schemastore.org/package.json",
-  "version": "1.14.48",
+  "version": "1.14.46",
  "name": "@opencode-ai/core",
  "type": "module",
  "license": "MIT",
@@ -0,0 +1,370 @@
 import { Effect, Option, Schema, SchemaAST } from "effect"
 import z from "zod"
 /**
 * Annotation key for providing a hand-crafted Zod schema that the walker
 * should use instead of re-deriving from the AST.  Attach it via
 * `Schema.String.annotate({ [ZodOverride]: z.string().startsWith("per") })`.
 */
 export const ZodOverride: unique symbol = Symbol.for("effect-zod/override")
 // AST nodes are immutable and frequently shared across schemas (e.g. a single
 // Schema.Class embedded in multiple parents). Memoizing by node identity
 // avoids rebuilding equivalent Zod subtrees and keeps derived children stable
 // by reference across callers.
 const walkCache = new WeakMap<SchemaAST.AST, z.ZodTypeAny>()
 // Shared empty ParseOptions for the rare callers that need one — avoids
 // allocating a fresh object per parse inside refinements and transforms.
 const EMPTY_PARSE_OPTIONS = {} as SchemaAST.ParseOptions
 export function zod<S extends Schema.Top>(schema: S): z.ZodType<Schema.Schema.Type<S>> {
  return walk(schema.ast) as z.ZodType<Schema.Schema.Type<S>>
 }
 /**
 * Derive a Zod value from an Effect Schema (or a Schema-backed export with a
 * `.zod` static) and narrow the result to `z.ZodObject<any>` so `.shape`,
 * `.omit`, `.extend`, and friends are accessible.
 *
 * The `zod()` walker returns `z.ZodType<T>` because not every AST node decodes
 * to an object; this helper keeps the "I started from a `Schema.Struct`" cast
 * in one place instead of sprinkling `as unknown as z.ZodObject<any>` across
 * call sites.
 *
 * The return is intentionally loose — carrying Schema field types through the
 * mapped `.omit()` / `.extend()` surface triggers brand-intersection
 * explosions for branded primitives (`string & Brand<"SessionID">` extends
 * `object` via the brand and gets walked into the prototype by `DeepPartial`,
 * `updateSchema`, etc.), and zod's inference through `z.ZodType<T | undefined>`
 * wrappers also can't reconstruct `T` cleanly. Consumers that care about the
 * post-`.omit()` shape should cast `c.req.valid(...)` to the expected type.
 */
 export function zodObject<S extends Schema.Top>(schema: S): z.ZodObject<any> {
  const derived: z.ZodTypeAny = "zod" in schema && isZodType(schema.zod) ? schema.zod : walk(schema.ast)
  return derived as unknown as z.ZodObject<any>
 }
 function isZodType(value: unknown): value is z.ZodTypeAny {
  return typeof value === "object" && value !== null && "_zod" in value
 }
 /**
 * Emit a JSON Schema for a tool/route parameter schema — derives the zod form
 * via the walker so Effect Schema inputs flow through the same zod-openapi
 * pipeline the LLM/SDK layer already depends on.  `io: "input"` mirrors what
 * `session/prompt.ts` has always passed to `ai`'s `jsonSchema()` helper.
 */
 export function toJsonSchema<S extends Schema.Top>(schema: S) {
  return z.toJSONSchema(zod(schema), { io: "input" })
 }
 function walk(ast: SchemaAST.AST): z.ZodTypeAny {
  const cached = walkCache.get(ast)
  if (cached) return cached
  const result = walkUncached(ast)
  walkCache.set(ast, result)
  return result
 }
 function walkUncached(ast: SchemaAST.AST): z.ZodTypeAny {
  const override = (ast.annotations as any)?.[ZodOverride] as z.ZodTypeAny | undefined
  // `description` annotations layer on top of an override so callers can
  // reuse a shared override schema (e.g. `SessionID`) and still add a
  // per-field description on the outer wrapper.
  const base = override ?? bodyWithChecks(ast)
  const desc = SchemaAST.resolveDescription(ast)
  const ref = SchemaAST.resolveIdentifier(ast)
  const described = desc ? base.describe(desc) : base
  return ref ? described.meta({ ref }) : described
 }
 function bodyWithChecks(ast: SchemaAST.AST): z.ZodTypeAny {
  // Schema.Class wraps its fields in a Declaration AST plus an encoding that
  // constructs the class instance. For the Zod derivation we want the plain
  // field shape (the decoded/consumer view), not the class instance — so
  // Declarations fall through to body(), not encoded(). User-level
  // Schema.decodeTo / Schema.transform attach encoding to non-Declaration
  // nodes, where we do apply the transform.
  //
  // Schema.withDecodingDefault also attaches encoding, but we want `.default(v)`
  // on the inner Zod rather than a transform wrapper — so optional ASTs whose
  // encoding resolves a default from Option.none() route through body()/opt().
  const hasEncoding = ast.encoding?.length && (ast._tag !== "Declaration" || ast.typeParameters.length === 0)
  const hasTransform = hasEncoding && !(SchemaAST.isOptional(ast) && extractDefault(ast) !== undefined)
  const base = hasTransform ? encoded(ast) : body(ast)
  return ast.checks?.length ? applyChecks(base, ast.checks, ast) : base
 }
 // Walk the encoded side and apply each link's decode to produce the decoded
 // shape. A node `Target` produced by `from.decodeTo(Target)` carries
 // `Target.encoding = [Link(from, transformation)]`. Chained decodeTo calls
 // nest the encoding via `Link.to` so walking it recursively threads all
 // prior transforms — typical encoding.length is 1.
 function encoded(ast: SchemaAST.AST): z.ZodTypeAny {
  const encoding = ast.encoding!
  return encoding.reduce<z.ZodTypeAny>(
    (acc, link) => acc.transform((v) => decode(link.transformation, v)),
    walk(encoding[0].to),
  )
 }
 // Transformations built via pure `SchemaGetter.transform(fn)` (the common
 // decodeTo case) resolve synchronously, so running with no services is safe.
 // Effectful / middleware-based transforms will surface as Effect defects.
 function decode(transformation: SchemaAST.Link["transformation"], value: unknown): unknown {
  const exit = Effect.runSyncExit(
    (transformation.decode as any).run(Option.some(value), EMPTY_PARSE_OPTIONS) as Effect.Effect<
      Option.Option<unknown>
    >,
  )
  if (exit._tag === "Failure") throw new Error(`effect-zod: transform failed: ${String(exit.cause)}`)
  return Option.getOrElse(exit.value, () => value)
 }
 // Flatten FilterGroups and any nested variants into a linear list of Filters.
 // Well-known filters (Schema.isInt, isGreaterThan, isPattern, …) are
 // translated into native Zod methods so their JSON Schema output includes
 // the corresponding constraint (type: integer, exclusiveMinimum, pattern, …).
 // Anything else falls back to a single .superRefine layer — runtime-only,
 // emits no JSON Schema constraint.
 function applyChecks(out: z.ZodTypeAny, checks: SchemaAST.Checks, ast: SchemaAST.AST): z.ZodTypeAny {
  const filters: SchemaAST.Filter<unknown>[] = []
  const collect = (c: SchemaAST.Check<unknown>) => {
    if (c._tag === "FilterGroup") c.checks.forEach(collect)
    else filters.push(c)
  }
  checks.forEach(collect)
  const unhandled: SchemaAST.Filter<unknown>[] = []
  const translated = filters.reduce<z.ZodTypeAny>((acc, filter) => {
    const next = translateFilter(acc, filter)
    if (next) return next
    unhandled.push(filter)
    return acc
  }, out)
  if (unhandled.length === 0) return translated
  return translated.superRefine((value, ctx) => {
    for (const filter of unhandled) {
      const issue = filter.run(value, ast, EMPTY_PARSE_OPTIONS)
      if (!issue) continue
      const message = issueMessage(issue) ?? (filter.annotations as any)?.message ?? "Validation failed"
      ctx.addIssue({ code: "custom", message })
    }
  })
 }
 // Translate a well-known Effect Schema filter into a native Zod method call on
 // `out`. Dispatch is keyed on `filter.annotations.meta._tag`, which every
 // built-in check factory (isInt, isGreaterThan, isPattern, …) attaches at
 // construction time. Returns `undefined` for unrecognised filters so the
 // caller can fall back to the generic .superRefine path.
 function translateFilter(out: z.ZodTypeAny, filter: SchemaAST.Filter<unknown>): z.ZodTypeAny | undefined {
  const meta = (filter.annotations as { meta?: Record<string, unknown> } | undefined)?.meta
  if (!meta || typeof meta._tag !== "string") return undefined
  switch (meta._tag) {
    case "isInt":
      return call(out, "int")
    case "isFinite":
      return call(out, "finite")
    case "isGreaterThan":
      return call(out, "gt", meta.exclusiveMinimum)
    case "isGreaterThanOrEqualTo":
      return call(out, "gte", meta.minimum)
    case "isLessThan":
      return call(out, "lt", meta.exclusiveMaximum)
    case "isLessThanOrEqualTo":
      return call(out, "lte", meta.maximum)
    case "isBetween": {
      const lo = meta.exclusiveMinimum ? call(out, "gt", meta.minimum) : call(out, "gte", meta.minimum)
      if (!lo) return undefined
      return meta.exclusiveMaximum ? call(lo, "lt", meta.maximum) : call(lo, "lte", meta.maximum)
    }
    case "isMultipleOf":
      return call(out, "multipleOf", meta.divisor)
    case "isMinLength":
      return call(out, "min", meta.minLength)
    case "isMaxLength":
      return call(out, "max", meta.maxLength)
    case "isLengthBetween": {
      const lo = call(out, "min", meta.minimum)
      if (!lo) return undefined
      return call(lo, "max", meta.maximum)
    }
    case "isPattern":
      return call(out, "regex", meta.regExp)
    case "isStartsWith":
      return call(out, "startsWith", meta.startsWith)
    case "isEndsWith":
      return call(out, "endsWith", meta.endsWith)
    case "isIncludes":
      return call(out, "includes", meta.includes)
    case "isUUID":
      return call(out, "uuid")
    case "isULID":
      return call(out, "ulid")
    case "isBase64":
      return call(out, "base64")
    case "isBase64Url":
      return call(out, "base64url")
  }
  return undefined
 }
 // Invoke a named Zod method on `target` if it exists, otherwise return
 // undefined so the caller can fall back. Using this helper instead of a
 // typed cast keeps `translateFilter` free of per-case narrowing noise.
 function call(target: z.ZodTypeAny, method: string, ...args: unknown[]): z.ZodTypeAny | undefined {
  const fn = (target as unknown as Record<string, ((...a: unknown[]) => z.ZodTypeAny) | undefined>)[method]
  return typeof fn === "function" ? fn.apply(target, args) : undefined
 }
 function issueMessage(issue: any): string | undefined {
  if (typeof issue?.annotations?.message === "string") return issue.annotations.message
  if (typeof issue?.message === "string") return issue.message
  return undefined
 }
 function body(ast: SchemaAST.AST): z.ZodTypeAny {
  if (SchemaAST.isOptional(ast)) return opt(ast)
  switch (ast._tag) {
    case "String":
      return z.string()
    case "Number":
      return z.number()
    case "Boolean":
      return z.boolean()
    case "Null":
      return z.null()
    case "Undefined":
      return z.undefined()
    case "Any":
    case "Unknown":
      return z.unknown()
    case "Never":
      return z.never()
    case "Literal":
      return z.literal(ast.literal)
    case "Union":
      return union(ast)
    case "Objects":
      return object(ast)
    case "Arrays":
      return array(ast)
    case "Declaration":
      return decl(ast)
    default:
      return fail(ast)
  }
 }
 function opt(ast: SchemaAST.AST): z.ZodTypeAny {
  if (ast._tag !== "Union") return fail(ast)
  const items = ast.types.filter((item) => item._tag !== "Undefined")
  const inner =
    items.length === 1
      ? walk(items[0])
      : items.length > 1
        ? z.union(items.map(walk) as [z.ZodTypeAny, z.ZodTypeAny, ...Array<z.ZodTypeAny>])
        : z.undefined()
  // Schema.withDecodingDefault attaches an encoding `Link` whose transformation
  // decode Getter resolves `Option.none()` to `Option.some(default)`.  Invoke
  // it to extract the default and emit `.default(...)` instead of `.optional()`.
  const fallback = extractDefault(ast)
  if (fallback !== undefined) return inner.default(fallback.value)
  return inner.optional()
 }
 type DecodeLink = {
  readonly transformation: {
    readonly decode: {
      readonly run: (
        input: Option.Option<unknown>,
        options: SchemaAST.ParseOptions,
      ) => Effect.Effect<Option.Option<unknown>, unknown>
    }
  }
 }
 function extractDefault(ast: SchemaAST.AST): { value: unknown } | undefined {
  const encoding = (ast as { encoding?: ReadonlyArray<DecodeLink> }).encoding
  if (!encoding?.length) return undefined
  // Walk the chain of encoding Links in order; the first Getter that produces
  // a value from Option.none wins.  withDecodingDefault always puts its
  // defaulting Link adjacent to the optional Union.
  for (const link of encoding) {
    const probe = Effect.runSyncExit(link.transformation.decode.run(Option.none(), {}))
    if (probe._tag !== "Success") continue
    if (Option.isSome(probe.value)) return { value: probe.value.value }
  }
  return undefined
 }
 function union(ast: SchemaAST.Union): z.ZodTypeAny {
  // When every member is a string literal, emit z.enum() so that
  // JSON Schema produces { "enum": [...] } instead of { "anyOf": [{ "const": ... }] }.
  if (ast.types.length >= 2 && ast.types.every((t) => t._tag === "Literal" && typeof t.literal === "string")) {
    return z.enum(ast.types.map((t) => (t as SchemaAST.Literal).literal as string) as [string, ...string[]])
  }
  const items = ast.types.map(walk)
  if (items.length === 1) return items[0]
  if (items.length < 2) return fail(ast)
  const discriminator = ast.annotations?.discriminator
  if (typeof discriminator === "string") {
    return z.discriminatedUnion(discriminator, items as [z.ZodObject<any>, z.ZodObject<any>, ...z.ZodObject<any>[]])
  }
  return z.union(items as [z.ZodTypeAny, z.ZodTypeAny, ...Array<z.ZodTypeAny>])
 }
 function object(ast: SchemaAST.Objects): z.ZodTypeAny {
  // Pure record: { [k: string]: V }
  if (ast.propertySignatures.length === 0 && ast.indexSignatures.length === 1) {
    const sig = ast.indexSignatures[0]
    if (sig.parameter._tag !== "String") return fail(ast)
    return z.record(z.string(), walk(sig.type))
  }
  // Pure object with known fields and no index signatures.
  if (ast.indexSignatures.length === 0) {
    return z.object(Object.fromEntries(ast.propertySignatures.map((sig) => [String(sig.name), walk(sig.type)])))
  }
  // Struct with a catchall (StructWithRest): known fields + index signature.
  // Only supports a single string-keyed index signature; multi-signature or
  // symbol/number keys fall through to fail.
  if (ast.indexSignatures.length !== 1) return fail(ast)
  const sig = ast.indexSignatures[0]
  if (sig.parameter._tag !== "String") return fail(ast)
  return z
    .object(Object.fromEntries(ast.propertySignatures.map((p) => [String(p.name), walk(p.type)])))
    .catchall(walk(sig.type))
 }
 function array(ast: SchemaAST.Arrays): z.ZodTypeAny {
  // Pure variadic arrays: { elements: [], rest: [item] }
  if (ast.elements.length === 0) {
    if (ast.rest.length !== 1) return fail(ast)
    return z.array(walk(ast.rest[0]))
  }
  // Fixed-length tuples: { elements: [a, b, ...], rest: [] }
  // Tuples with a variadic tail (...rest) are not yet supported.
  if (ast.rest.length > 0) return fail(ast)
  const items = ast.elements.map(walk)
  return z.tuple(items as [z.ZodTypeAny, ...Array<z.ZodTypeAny>])
 }
 function decl(ast: SchemaAST.Declaration): z.ZodTypeAny {
  if (ast.typeParameters.length !== 1) return fail(ast)
  return walk(ast.typeParameters[0])
 }
 function fail(ast: SchemaAST.AST): never {
  const ref = SchemaAST.resolveIdentifier(ast)
  throw new Error(`unsupported effect schema: ${ref ?? ast._tag}`)
 }
@@ -1,4 +1,5 @@
 import { Config } from "effect"
 import { InstallationChannel } from "../installation/version"
 function truthy(key: string) {
  const value = process.env[key]?.toLowerCase()
@@ -10,6 +11,13 @@ function falsy(key: string) {
  return value === "false" || value === "0"
 }
 // Channels where new experiments default to ON (unstable / internal users).
 // Stable channels (`prod`, `latest`) stay opt-in.
 const UNSTABLE_CHANNELS = new Set(["dev", "beta", "local"])
 function unstableDefault(key: string) {
  return truthy(key) || (!falsy(key) && UNSTABLE_CHANNELS.has(InstallationChannel))
 }
 function number(key: string) {
  const value = process.env[key]
  if (!value) return undefined
@@ -48,6 +56,9 @@ export const Flag = {
  OPENCODE_DISABLE_CLAUDE_CODE_PROMPT: OPENCODE_DISABLE_CLAUDE_CODE || truthy("OPENCODE_DISABLE_CLAUDE_CODE_PROMPT"),
  OPENCODE_DISABLE_CLAUDE_CODE_SKILLS,
  OPENCODE_DISABLE_EXTERNAL_SKILLS: truthy("OPENCODE_DISABLE_EXTERNAL_SKILLS"),
  // Default-on for dev/beta/local; opt-in for stable. Set
  // OPENCODE_EXPERIMENTAL_CUSTOMIZE_SKILL=false to force off, =true to force on.
  OPENCODE_EXPERIMENTAL_CUSTOMIZE_SKILL: unstableDefault("OPENCODE_EXPERIMENTAL_CUSTOMIZE_SKILL"),
  OPENCODE_FAKE_VCS: process.env["OPENCODE_FAKE_VCS"],
  OPENCODE_SERVER_PASSWORD: process.env["OPENCODE_SERVER_PASSWORD"],
  OPENCODE_SERVER_USERNAME: process.env["OPENCODE_SERVER_USERNAME"],
@@ -85,7 +96,6 @@ 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.
@@ -1,4 +1,5 @@
 import { Option, Schema, SchemaGetter } from "effect"
 import { zod, ZodOverride } from "./effect-zod"
 /**
 * Integer greater than zero.
@@ -20,6 +21,7 @@ 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() }),
  )
 /**
@@ -1,8 +1,8 @@
-import { Schema } from "effect"
+import z from "zod"
 export abstract class NamedError extends Error {
-  abstract schema(): Schema.Top
+  abstract schema(): z.core.$ZodType
-  abstract toObject(): { name: string; data: unknown }
+  abstract toObject(): { name: string; data: any }
  static hasName(error: unknown, name: string): boolean {
    return (
@@ -10,42 +10,30 @@ export abstract class NamedError extends Error {
    )
  }
-  static create<Name extends string, Fields extends Schema.Struct.Fields>(
+  static create<Name extends string, Data extends z.core.$ZodType>(name: Name, data: Data) {
-    name: Name,
+    const schema = z
-    fields: Fields,
+      .object({
-  ): ReturnType<typeof NamedError.createSchemaClass<Name, Schema.Struct<Fields>>>
+        name: z.literal(name),
  static create<Name extends string, DataSchema extends Schema.Top>(
    name: Name,
    data: DataSchema,
  ): ReturnType<typeof NamedError.createSchemaClass<Name, DataSchema>>
  static create<Name extends string>(name: Name, data: Schema.Top | Schema.Struct.Fields) {
    return NamedError.createSchemaClass(name, Schema.isSchema(data) ? data : Schema.Struct(data))
  }
  private static createSchemaClass<Name extends string, DataSchema extends Schema.Top>(name: Name, data: DataSchema) {
    const schema = Schema.Struct({
      name: Schema.Literal(name),
        data,
-    }).annotate({ identifier: name })
+      })
-    type Data = Schema.Schema.Type<DataSchema>
+      .meta({
-
+        ref: name,
      })
    const result = class extends NamedError {
      public static readonly Schema = schema
      public static readonly EffectSchema = schema
      public static readonly tag = name
-      public override readonly name = name
+      public override readonly name = name as Name
      constructor(
-        public readonly data: Data,
+        public readonly data: z.input<Data>,
        options?: ErrorOptions,
      ) {
        super(name, options)
        this.name = name
      }
-      static isInstance(input: unknown): input is InstanceType<typeof result> {
+      static isInstance(input: any): input is InstanceType<typeof result> {
-        return NamedError.hasName(input, name)
+        return typeof input === "object" && "name" in input && input.name === name
      }
      schema() {
@@ -63,7 +51,10 @@ export abstract class NamedError extends Error {
    return result
  }
-  public static readonly Unknown = NamedError.create("UnknownError", {
+  public static readonly Unknown = NamedError.create(
-    message: Schema.String,
+    "UnknownError",
-  })
+    z.object({
      message: z.string(),
    }),
  )
 }
@@ -1,7 +1,7 @@
 {
  "name": "@opencode-ai/desktop",
  "private": true,
-  "version": "1.14.48",
+  "version": "1.14.46",
  "type": "module",
  "license": "MIT",
  "homepage": "https://opencode.ai",
@@ -291,19 +291,25 @@ const main = Effect.gen(function* () {
      if (mainWindow) sendSqliteMigrationProgress(mainWindow, progress)
    })
    ensureLoopbackNoProxy()
    useEnvProxy()
    logger.log("spawning sidecar", { url })
    const { listener, health } = yield* Effect.promise(() =>
-      spawnLocalServer(hostname, port, password, {
+      spawnLocalServer(
        hostname,
        port,
        password,
        () => {
          ensureLoopbackNoProxy()
          useEnvProxy()
        },
        {
          needsMigration,
          userDataPath: app.getPath("userData"),
          onSqliteProgress: (progress) => initEmitter.emit("sqlite", progress),
          onStdout: (message) => logger.log("sidecar stdout", { message }),
          onStderr: (message) => logger.warn("sidecar stderr", { message }),
          onExit: (code) => logger.warn("sidecar exited", { code }),
-      }),
+        },
      ),
    )
    server = listener
    yield* Deferred.succeed(serverReady, {
@@ -70,8 +70,10 @@ export async function spawnLocalServer(
  hostname: string,
  port: number,
  password: string,
  configureEnv: () => void,
  options: SpawnLocalServerOptions,
 ) {
  configureEnv?.()
  const sidecar = join(dirname(fileURLToPath(import.meta.url)), "sidecar.js")
  const child = utilityProcess.fork(sidecar, [], {
    cwd: process.cwd(),
@@ -1,6 +1,6 @@
 {
  "name": "@opencode-ai/enterprise",
-  "version": "1.14.48",
+  "version": "1.14.46",
  "private": true,
  "type": "module",
  "license": "MIT",
@@ -15,6 +15,7 @@ import { Binary } from "@opencode-ai/core/util/binary"
 import { NamedError } from "@opencode-ai/core/util/error"
 import { DateTime } from "luxon"
 import { createStore } from "solid-js/store"
 import z from "zod"
 import NotFound from "../[...404]"
 import { Tabs } from "@opencode-ai/ui/tabs"
 import { MessageNav } from "@opencode-ai/ui/message-nav"
@@ -32,28 +33,13 @@ const ClientOnlyWorkerPoolProvider = clientOnly(() =>
  })),
 )
-class SessionDataMissingError extends NamedError {
+const SessionDataMissingError = NamedError.create(
-  public override readonly name = "SessionDataMissingError"
+  "SessionDataMissingError",
-
+  z.object({
-  constructor(
+    sessionID: z.string(),
-    public readonly data: { sessionID: string; message?: string },
+    message: z.string().optional(),
-    options?: ErrorOptions,
+  }),
-  ) {
+)
    super("SessionDataMissingError", options)
  }
  static isInstance(input: unknown): input is SessionDataMissingError {
    return NamedError.hasName(input, "SessionDataMissingError")
  }
  schema(): never {
    throw new Error("SessionDataMissingError does not expose a schema")
  }
  toObject() {
    return { name: this.name, data: this.data }
  }
 }
 const getData = query(async (shareID) => {
  "use server"
@@ -1,7 +1,7 @@
 id = "opencode"
 name = "OpenCode"
 description = "The open source coding agent."
-version = "1.14.48"
+version = "1.14.46"
 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/v1.14.48/opencode-darwin-arm64.zip"
+archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.46/opencode-darwin-arm64.zip"
 cmd = "./opencode"
 args = ["acp"]
 [agent_servers.opencode.targets.darwin-x86_64]
-archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.48/opencode-darwin-x64.zip"
+archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.46/opencode-darwin-x64.zip"
 cmd = "./opencode"
 args = ["acp"]
 [agent_servers.opencode.targets.linux-aarch64]
-archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.48/opencode-linux-arm64.tar.gz"
+archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.46/opencode-linux-arm64.tar.gz"
 cmd = "./opencode"
 args = ["acp"]
 [agent_servers.opencode.targets.linux-x86_64]
-archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.48/opencode-linux-x64.tar.gz"
+archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.46/opencode-linux-x64.tar.gz"
 cmd = "./opencode"
 args = ["acp"]
 [agent_servers.opencode.targets.windows-x86_64]
-archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.48/opencode-windows-x64.zip"
+archive = "https://github.com/anomalyco/opencode/releases/download/v1.14.46/opencode-windows-x64.zip"
 cmd = "./opencode.exe"
 args = ["acp"]
@@ -1,6 +1,6 @@
 {
  "name": "@opencode-ai/function",
-  "version": "1.14.48",
+  "version": "1.14.46",
  "$schema": "https://json.schemastore.org/package.json",
  "private": true,
  "type": "module",
@@ -70,15 +70,19 @@ 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
+By default, requests match on canonicalized method, URL, headers, and JSON
-request executed at runtime is served by the Nth recorded interaction, and
+body (object keys sorted). Two dispatch strategies are available:
 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
+- **`match`** (default) — find the first recorded interaction whose request
-it silently returns the first recorded response for repeated identical
+  matches the incoming request. Same request twice returns the same response.
-requests, masking state changes that retry/polling/cache-hit tests need to
+- **`sequential`** — return interactions in the order they were recorded,
-observe. If you reorder requests in a test, re-record the cassette.
+  validating each one matches as the cursor advances. Use for ordered flows
  where the same URL is hit multiple times with meaningful state changes
  (pagination, retries, polling).
 ```ts
 HttpRecorder.cassetteLayer("flow/poll-until-done", { dispatch: "sequential" })
 ```
 Supply your own matcher via `match: (incoming, recorded) => boolean` for
 custom equivalence (e.g. ignoring a timestamp field in the body).
@@ -190,6 +194,7 @@ type RecordReplayOptions = {
  directory?: string // default: <cwd>/test/fixtures/recordings
  metadata?: Record<string, unknown> // merged into cassette.metadata
  redactor?: Redactor // default: Redactor.defaults()
  dispatch?: "match" | "sequential" // default: "match"
  match?: (incoming, recorded) => boolean // custom matcher
 }
 ```
@@ -206,4 +211,4 @@ type RecordReplayOptions = {
 | `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.      |
+| `matching.ts`  | Request matcher, canonicalization, dispatch strategies, mismatch diagnostics.    |
@@ -1,6 +1,6 @@
 {
  "$schema": "https://json.schemastore.org/package.json",
-  "version": "1.14.48",
+  "version": "1.14.46",
  "name": "@opencode-ai/http-recorder",
  "type": "module",
  "license": "MIT",
@@ -11,7 +11,7 @@ import {
  UrlParams,
 } from "effect/unstable/http"
 import * as CassetteService from "./cassette"
-import { defaultMatcher, selectSequential, type RequestMatcher } from "./matching"
+import { defaultMatcher, selectMatch, selectSequential, type RequestMatcher } from "./matching"
 import { appendOrFail, makeReplayState, resolveAutoMode } from "./recorder"
 import { defaults, type Redactor } from "./redactor"
 import { redactUrl } from "./redaction"
@@ -24,6 +24,7 @@ export interface RecordReplayOptions {
  readonly directory?: string
  readonly metadata?: CassetteMetadata
  readonly redactor?: Redactor
  readonly dispatch?: "match" | "sequential"
  readonly match?: RequestMatcher
 }
@@ -70,6 +71,7 @@ export const recordingLayer = (
      const match = options.match ?? defaultMatcher
      const requested = options.mode ?? "auto"
      const mode = requested === "auto" ? yield* resolveAutoMode(cassetteService, name) : requested
      const sequential = options.dispatch === "sequential"
      const replay = yield* makeReplayState(cassetteService, name, httpInteractions)
      const snapshotRequest = (request: HttpClientRequest.HttpClientRequest) =>
@@ -117,12 +119,14 @@ export const recordingLayer = (
              transportError(request, `Fixture "${name}" not found. Run locally to record it (CI=true forces replay).`),
            ),
          )
-          const result = selectSequential(interactions, incoming, match, yield* replay.cursor)
+          const result = sequential
            ? selectSequential(interactions, incoming, match, yield* replay.cursor)
            : selectMatch(interactions, incoming, match)
          if (!result.interaction)
            return yield* Effect.fail(
              transportError(request, `Fixture "${name}" does not match the current request: ${result.detail}.`),
            )
-          yield* replay.advance
+          if (sequential) yield* replay.advance
          return HttpClientResponse.fromWeb(
            request,
            new Response(decodeResponseBody(result.interaction.response), result.interaction.response),
@@ -92,6 +92,24 @@ export const requestDiff = (expected: RequestSnapshot, received: RequestSnapshot
  return lines
 }
 export const mismatchDetail = (interactions: ReadonlyArray<HttpInteraction>, incoming: RequestSnapshot): string => {
  if (interactions.length === 0) return "cassette has no recorded HTTP interactions"
  const ranked = interactions
    .map((interaction, index) => ({ index, lines: requestDiff(interaction.request, incoming) }))
    .toSorted((a, b) => a.lines.length - b.lines.length || a.index - b.index)
  const best = ranked[0]
  return ["no recorded interaction matched", `closest interaction: #${best.index + 1}`, ...best.lines].join("\n")
 }
 export const selectMatch = (
  interactions: ReadonlyArray<HttpInteraction>,
  incoming: RequestSnapshot,
  match: RequestMatcher,
 ): { readonly interaction: HttpInteraction | undefined; readonly detail: string } => {
  const interaction = interactions.find((candidate) => match(incoming, candidate.request))
  return { interaction, detail: interaction ? "" : mismatchDetail(interactions, incoming) }
 }
 export const selectSequential = (
  interactions: ReadonlyArray<HttpInteraction>,
  incoming: RequestSnapshot,
@@ -230,10 +230,19 @@ describe("http-recorder", () => {
    )
  })
-  test("replay returns recorded responses in order for identical requests", async () => {
+  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 () => {
    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"}')
@@ -241,8 +250,21 @@ describe("http-recorder", () => {
    )
  })
-  test("replay reports cursor exhaustion when more requests are made than recorded", async () => {
+  test("default matcher returns the first match for identical requests", async () => {
-    await run(
+    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" },
      Effect.gen(function* () {
        yield* post("https://example.test/echo", { step: 1 })
        yield* post("https://example.test/echo", { step: 2 })
@@ -252,8 +274,10 @@ describe("http-recorder", () => {
    )
  })
-  test("replay validates each recorded request in order", async () => {
+  test("sequential dispatch still validates each recorded request", async () => {
-    await run(
+    await runWith(
      "record-replay/multi-step",
      { dispatch: "sequential" },
      Effect.gen(function* () {
        yield* post("https://example.test/echo", { step: 1 })
        const exit = yield* Effect.exit(post("https://example.test/echo", { step: 3 }))
@@ -307,13 +331,14 @@ describe("http-recorder", () => {
    }
  })
-  test("mismatch diagnostics show redacted request differences against the expected interaction", async () => {
+  test("mismatch diagnostics show closest redacted request differences", 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:")
@@ -8,10 +8,6 @@
 - 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.
@@ -170,12 +166,12 @@ 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 = ToolCallPart.make({ id: "call_1", name: "lookup", input: { query: "weather" } })
+const call = LLM.toolCall({ id: "call_1", name: "lookup", input: { query: "weather" } })
-const result = Message.tool({ id: "call_1", name: "lookup", result: { forecast: "sunny" } })
+const result = LLM.toolMessage({ id: "call_1", name: "lookup", result: { forecast: "sunny" } })
 const followUp = LLM.request({
  model,
-  messages: [Message.user("Weather?"), Message.assistant([call]), result],
+  messages: [LLM.user("Weather?"), LLM.assistant([call]), result],
 })
 ```
@@ -293,6 +289,6 @@ 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 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.
+**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.
 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.
@@ -1,129 +0,0 @@
 # @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
@@ -1,6 +1,6 @@
 {
  "$schema": "https://json.schemastore.org/package.json",
-  "version": "1.14.48",
+  "version": "1.14.46",
  "name": "@opencode-ai/llm",
  "type": "module",
  "license": "MIT",
@@ -1,111 +0,0 @@
 // 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 })
 }
@@ -44,8 +44,32 @@ 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
@@ -71,10 +95,10 @@ export const request = (input: RequestInput) => {
  return new LLMRequest({
    ...rest,
    system: SystemPart.content(requestSystem),
-    messages: [...(messages?.map(Message.make) ?? []), ...(prompt === undefined ? [] : [Message.user(prompt)])],
+    messages: [...(messages?.map(message) ?? []), ...(prompt === undefined ? [] : [user(prompt)])],
-    tools: tools?.map(ToolDefinition.make) ?? [],
+    tools: tools?.map(toolDefinition) ?? [],
-    toolChoice: requestToolChoice ? ToolChoice.make(requestToolChoice) : undefined,
+    toolChoice: requestToolChoice ? toolChoice(requestToolChoice) : undefined,
-    generation: requestGeneration === undefined ? undefined : GenerationOptions.make(requestGeneration),
+    generation: requestGeneration === undefined ? undefined : generation(requestGeneration),
    providerOptions: requestProviderOptions,
    http: requestHttp === undefined ? undefined : HttpOptions.make(requestHttp),
  })
@@ -16,8 +16,6 @@ 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"
@@ -27,10 +25,7 @@ export const PATH = "/messages"
 // =============================================================================
 // Request Body Schema
 // =============================================================================
-const AnthropicCacheControl = Schema.Struct({
+const AnthropicCacheControl = Schema.Struct({ type: Schema.tag("ephemeral") })
  type: Schema.tag("ephemeral"),
  ttl: Schema.optional(Schema.Literals(["5m", "1h"])),
 })
 const AnthropicTextBlock = Schema.Struct({
  type: Schema.tag("text"),
@@ -191,7 +186,6 @@ 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
@@ -199,24 +193,8 @@ const invalid = ProviderShared.invalidRequest
 // =============================================================================
 // Request Lowering
 // =============================================================================
-// Anthropic accepts at most 4 explicit cache_control breakpoints per request,
+const cacheControl = (cache: CacheHint | undefined) =>
-// across `tools`, `system`, and `messages`. Beyond the cap the API returns a
+  cache?.type === "ephemeral" ? { type: "ephemeral" as const } : undefined
 // 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 })
@@ -226,11 +204,10 @@ const signatureFromMetadata = (metadata: ProviderMetadata | undefined): string |
  return typeof anthropic.signature === "string" ? anthropic.signature : undefined
 }
-const lowerTool = (breakpoints: Cache.Breakpoints, tool: ToolDefinition): AnthropicTool => ({
+const lowerTool = (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"]>) =>
@@ -272,10 +249,7 @@ 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* (
+const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (request: LLMRequest) {
  request: LLMRequest,
  breakpoints: Cache.Breakpoints,
 ) {
  const messages: AnthropicMessage[] = []
  for (const message of request.messages) {
@@ -284,7 +258,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
      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(breakpoints, part.cache) })
+        content.push({ type: "text", text: part.text, cache_control: cacheControl(part.cache) })
      }
      messages.push({ role: "user", content })
      continue
@@ -294,7 +268,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
      const content: AnthropicAssistantBlock[] = []
      for (const part of message.content) {
        if (part.type === "text") {
-          content.push({ type: "text", text: part.text, cache_control: cacheControl(breakpoints, part.cache) })
+          content.push({ type: "text", text: part.text, cache_control: cacheControl(part.cache) })
          continue
        }
        if (part.type === "reasoning") {
@@ -330,7 +304,6 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
        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 })
@@ -357,33 +330,18 @@ 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 →
+  return {
-  // messages. Tools live highest in the cache hierarchy, so when callers
+    model: request.model.id,
-  // over-mark we keep their tool hints and shed the message-tail ones first.
+    system:
  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),
+            cache_control: cacheControl(part.cache),
-        }))
+          })),
-  const messages = yield* lowerMessages(request, breakpoints)
+    messages: yield* lowerMessages(request),
-  if (breakpoints.dropped > 0) {
+    tools: request.tools.length === 0 || request.toolChoice?.type === "none" ? undefined : request.tools.map(lowerTool),
    yield* Effect.logWarning(
      `Anthropic Messages: dropped ${breakpoints.dropped} cache breakpoint(s); the API allows at most ${ANTHROPIC_BREAKPOINT_CAP} per request.`,
    )
  }
  return {
    model: request.model.id,
    system,
    messages,
    tools,
    tool_choice: toolChoice,
    stream: true as const,
    max_tokens: generation?.maxTokens ?? request.model.limits.output ?? 4096,
@@ -502,45 +460,37 @@ const onContentBlockStart = (state: ParserState, event: AnthropicEvent): StepRes
  if (!block) return [state, NO_EVENTS]
  if ((block.type === "tool_use" || block.type === "server_tool_use") && event.index !== undefined) {
    const events: LLMEvent[] = []
    const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
    return [
      {
        ...state,
        lifecycle,
        tools: ToolStream.start(state.tools, event.index, {
          id: block.id ?? String(event.index),
          name: block.name ?? "",
          providerExecuted: block.type === "server_tool_use",
        }),
      },
-      [...events, LLMEvent.toolInputStart({ id: block.id ?? String(event.index), name: block.name ?? "" })],
+      NO_EVENTS,
    ]
  }
  if (block.type === "text" && block.text) {
-    const events: LLMEvent[] = []
+    return [state, [LLMEvent.textDelta({ id: `text-${event.index ?? 0}`, text: block.text })]]
    return [
      { ...state, lifecycle: Lifecycle.textDelta(state.lifecycle, events, `text-${event.index ?? 0}`, block.text) },
      events,
    ]
  }
  if (block.type === "thinking" && block.thinking) {
    const events: LLMEvent[] = []
    return [
-      {
+      state,
-        ...state,
+      [
-        lifecycle: Lifecycle.reasoningDelta(state.lifecycle, events, `reasoning-${event.index ?? 0}`, block.thinking),
+        LLMEvent.reasoningDelta({
-      },
+          id: `reasoning-${event.index ?? 0}`,
-      events,
+          text: block.thinking,
        }),
      ],
    ]
  }
  const result = serverToolResultEvent(block)
-  if (!result) return [state, NO_EVENTS]
+  return [state, result ? [result] : NO_EVENTS]
  const events: LLMEvent[] = []
  return [{ ...state, lifecycle: Lifecycle.stepStart(state.lifecycle, events) }, [...events, result]]
 }
 const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(function* (
@@ -550,37 +500,25 @@ const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(f
  const delta = event.delta
  if (delta?.type === "text_delta" && delta.text) {
-    const events: LLMEvent[] = []
+    return [state, [LLMEvent.textDelta({ id: `text-${event.index ?? 0}`, text: delta.text })]] satisfies StepResult
    return [
      { ...state, lifecycle: Lifecycle.textDelta(state.lifecycle, events, `text-${event.index ?? 0}`, delta.text) },
      events,
    ] satisfies StepResult
  }
  if (delta?.type === "thinking_delta" && delta.thinking) {
    const events: LLMEvent[] = []
    return [
-      {
+      state,
-        ...state,
+      [LLMEvent.reasoningDelta({ id: `reasoning-${event.index ?? 0}`, text: delta.thinking })],
        lifecycle: Lifecycle.reasoningDelta(state.lifecycle, events, `reasoning-${event.index ?? 0}`, delta.thinking),
      },
      events,
    ] satisfies StepResult
  }
  if (delta?.type === "signature_delta" && delta.signature) {
    const events: LLMEvent[] = []
    return [
-      {
+      state,
-        ...state,
+      [
-        lifecycle: Lifecycle.reasoningEnd(
+        LLMEvent.reasoningEnd({
-          state.lifecycle,
+          id: `reasoning-${event.index ?? 0}`,
-          events,
+          providerMetadata: anthropicMetadata({ signature: delta.signature }),
-          `reasoning-${event.index ?? 0}`,
+        }),
-          anthropicMetadata({ signature: delta.signature }),
+      ],
        ),
      },
      events,
    ] satisfies StepResult
  }
@@ -594,10 +532,7 @@ const onContentBlockDelta = Effect.fn("AnthropicMessages.onContentBlockDelta")(f
      "Anthropic Messages tool argument delta is missing its tool call",
    )
    if (ToolStream.isError(result)) return yield* result
-    const events: LLMEvent[] = []
+    return [{ ...state, tools: result.tools }, result.event ? [result.event] : NO_EVENTS] satisfies StepResult
    const lifecycle = result.events.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
    events.push(...result.events)
    return [{ ...state, lifecycle, tools: result.tools }, events] satisfies StepResult
  }
  return [state, NO_EVENTS] satisfies StepResult
@@ -609,30 +544,23 @@ const onContentBlockStop = Effect.fn("AnthropicMessages.onContentBlockStop")(fun
 ) {
  if (event.index === undefined) return [state, NO_EVENTS] satisfies StepResult
  const result = yield* ToolStream.finish(ADAPTER, state.tools, event.index)
-  const events: LLMEvent[] = []
+  return [{ ...state, tools: result.tools }, result.event ? [result.event] : NO_EVENTS] satisfies StepResult
  const resultEvents = result.events ?? []
  const lifecycle = resultEvents.length
    ? Lifecycle.stepStart(state.lifecycle, events)
    : Lifecycle.reasoningEnd(
        Lifecycle.textEnd(state.lifecycle, events, `text-${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))
-  const events: LLMEvent[] = []
+  return [
-  const lifecycle = Lifecycle.finish(state.lifecycle, events, {
+    { ...state, usage },
    [
      LLMEvent.requestFinish({
        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 => [
@@ -666,7 +594,7 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: Protocol.jsonEvent(AnthropicEvent),
-    initial: () => ({ tools: ToolStream.empty<number>(), lifecycle: Lifecycle.initial() }),
+    initial: () => ({ tools: ToolStream.empty<number>() }),
    step,
  },
 })
@@ -17,7 +17,6 @@ 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"
@@ -109,7 +108,7 @@ type BedrockMessage = Schema.Schema.Type<typeof BedrockMessage>
 const BedrockSystemBlock = Schema.Union([BedrockTextBlock, BedrockCache.CachePointBlock])
 type BedrockSystemBlock = Schema.Schema.Type<typeof BedrockSystemBlock>
-const BedrockToolSpec = Schema.Struct({
+const BedrockTool = Schema.Struct({
  toolSpec: Schema.Struct({
    name: Schema.String,
    description: Schema.String,
@@ -118,9 +117,6 @@ const BedrockToolSpec = 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([
@@ -218,7 +214,7 @@ type BedrockEvent = Schema.Schema.Type<typeof BedrockEvent>
 // =============================================================================
 // Request Lowering
 // =============================================================================
-const lowerToolSpec = (tool: ToolDefinition): BedrockToolSpec => ({
+const lowerTool = (tool: ToolDefinition): BedrockTool => ({
  toolSpec: {
    name: tool.name,
    description: tool.description,
@@ -226,22 +222,11 @@ const lowerToolSpec = (tool: ToolDefinition): BedrockToolSpec => ({
  },
 })
 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(breakpoints, cache)
+  const cachePoint = BedrockCache.block(cache)
  return cachePoint ? [{ text }, cachePoint] : [{ text }]
 }
@@ -272,10 +257,7 @@ const lowerToolResult = (part: ToolResultPart): BedrockToolResultBlock => ({
  },
 })
-const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
+const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (request: LLMRequest) {
  request: LLMRequest,
  breakpoints: BedrockCache.Breakpoints,
 ) {
  const messages: BedrockMessage[] = []
  for (const message of request.messages) {
@@ -285,7 +267,7 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
        if (!ProviderShared.supportsContent(part, ["text", "media"]))
          return yield* ProviderShared.unsupportedContent("Bedrock Converse", "user", ["text", "media"])
        if (part.type === "text") {
-          content.push(...textWithCache(breakpoints, part.text, part.cache))
+          content.push(...textWithCache(part.text, part.cache))
          continue
        }
        if (part.type === "media") {
@@ -307,7 +289,7 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
            "tool-call",
          ])
        if (part.type === "text") {
-          content.push(...textWithCache(breakpoints, part.text, part.cache))
+          content.push(...textWithCache(part.text, part.cache))
          continue
        }
        if (part.type === "reasoning") {
@@ -327,13 +309,11 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
      continue
    }
-    const content: BedrockUserBlock[] = []
+    const content: BedrockToolResultBlock[] = []
    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 })
  }
@@ -343,32 +323,16 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
 // System prompts share the cache-point convention: emit the text block, then
 // optionally a positional `cachePoint` marker.
-const lowerSystem = (
+const lowerSystem = (system: ReadonlyArray<LLMRequest["system"][number]>): BedrockSystemBlock[] =>
-  breakpoints: BedrockCache.Breakpoints,
+  system.flatMap((part) => textWithCache(part.text, part.cache))
  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 ${BedrockCache.BEDROCK_BREAKPOINT_CAP} per request.`,
    )
  }
  return {
    modelId: request.model.id,
-    messages,
+    messages: yield* lowerMessages(request),
-    system,
+    system: request.system.length === 0 ? undefined : lowerSystem(request.system),
    inferenceConfig:
      generation?.maxTokens === undefined &&
      generation?.temperature === undefined &&
@@ -381,7 +345,10 @@ const fromRequest = Effect.fn("BedrockConverse.fromRequest")(function* (request:
            topP: generation?.topP,
            stopSequences: generation?.stop,
          },
-    toolConfig,
+    toolConfig:
      request.tools.length > 0 && request.toolChoice?.type !== "none"
        ? { tools: request.tools.map(lowerTool), toolChoice }
        : undefined,
  }
 })
@@ -421,64 +388,45 @@ interface ParserState {
  // `metadata` (carries usage). Hold the terminal event in state so `onHalt`
  // can emit exactly one finish after both chunks have had a chance to arrive.
  readonly pendingFinish: { readonly reason: FinishReason; readonly usage?: Usage } | undefined
  readonly hasToolCalls: boolean
  readonly lifecycle: Lifecycle.State
 }
 const step = (state: ParserState, event: BedrockEvent) =>
  Effect.gen(function* () {
    if (event.contentBlockStart?.start?.toolUse) {
      const index = event.contentBlockStart.contentBlockIndex
      const events: LLMEvent[] = []
      const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
      return [
        {
          ...state,
          lifecycle,
          tools: ToolStream.start(state.tools, index, {
            id: event.contentBlockStart.start.toolUse.toolUseId,
            name: event.contentBlockStart.start.toolUse.name,
          }),
        },
        [],
      ] as const
    }
    if (event.contentBlockDelta?.delta?.text) {
      return [
        state,
        [
-          ...events,
+          LLMEvent.textDelta({
-          LLMEvent.toolInputStart({
+            id: `text-${event.contentBlockDelta.contentBlockIndex}`,
-            id: event.contentBlockStart.start.toolUse.toolUseId,
+            text: event.contentBlockDelta.delta.text,
            name: event.contentBlockStart.start.toolUse.name,
          }),
        ],
      ] as const
    }
    if (event.contentBlockDelta?.delta?.text) {
      const events: LLMEvent[] = []
      return [
        {
          ...state,
          lifecycle: Lifecycle.textDelta(
            state.lifecycle,
            events,
            `text-${event.contentBlockDelta.contentBlockIndex}`,
            event.contentBlockDelta.delta.text,
          ),
        },
        events,
      ] as const
    }
    if (event.contentBlockDelta?.delta?.reasoningContent?.text) {
      const events: LLMEvent[] = []
      return [
-        {
+        state,
-          ...state,
+        [
-          lifecycle: Lifecycle.reasoningDelta(
+          LLMEvent.reasoningDelta({
-            state.lifecycle,
+            id: `reasoning-${event.contentBlockDelta.contentBlockIndex}`,
-            events,
+            text: event.contentBlockDelta.delta.reasoningContent.text,
-            `reasoning-${event.contentBlockDelta.contentBlockIndex}`,
+          }),
-            event.contentBlockDelta.delta.reasoningContent.text,
+        ],
          ),
        },
        events,
      ] as const
    }
@@ -492,33 +440,12 @@ const step = (state: ParserState, event: BedrockEvent) =>
        "Bedrock Converse tool delta is missing its tool call",
      )
      if (ToolStream.isError(result)) return yield* result
-      const events: LLMEvent[] = []
+      return [{ ...state, tools: result.tools }, result.event ? [result.event] : []] as const
      const lifecycle = result.events.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
      events.push(...result.events)
      return [{ ...state, lifecycle, tools: result.tools }, events] as const
    }
    if (event.contentBlockStop) {
      const result = yield* ToolStream.finish(ADAPTER, state.tools, event.contentBlockStop.contentBlockIndex)
-      const events: LLMEvent[] = []
+      return [{ ...state, tools: result.tools }, result.event ? [result.event] : []] as const
      const resultEvents = result.events ?? []
      const lifecycle = resultEvents.length
        ? Lifecycle.stepStart(state.lifecycle, events)
        : Lifecycle.reasoningEnd(
            Lifecycle.textEnd(state.lifecycle, events, `text-${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) {
@@ -558,15 +485,7 @@ const framing = BedrockEventStream.framing(ADAPTER)
 const onHalt = (state: ParserState): ReadonlyArray<LLMEvent> =>
  state.pendingFinish
-    ? (() => {
+    ? [LLMEvent.requestFinish({ 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
      })()
    : []
 // =============================================================================
@@ -584,12 +503,7 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: BedrockEvent,
-    initial: () => ({
+    initial: () => ({ tools: ToolStream.empty<number>(), pendingFinish: undefined }),
      tools: ToolStream.empty<number>(),
      pendingFinish: undefined,
      hasToolCalls: false,
      lifecycle: Lifecycle.initial(),
    }),
    step,
    onHalt,
  },
@@ -16,7 +16,6 @@ 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"
@@ -135,9 +134,10 @@ interface ParserState {
  readonly hasToolCalls: boolean
  readonly nextToolCallId: number
  readonly usage?: Usage
  readonly lifecycle: Lifecycle.State
 }
 const invalid = ProviderShared.invalidRequest
 const mediaData = ProviderShared.mediaBytes
 // =============================================================================
@@ -285,16 +285,16 @@ const fromRequest = Effect.fn("Gemini.fromRequest")(function* (request: LLMReque
 // `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.
 // Output is left undefined when the visible component is missing, so we
 // don't fabricate an inclusive number from a partial breakdown.
 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
+    usage.candidatesTokenCount !== undefined
      ? usage.candidatesTokenCount + (usage.thoughtsTokenCount ?? 0)
      : undefined
  return new Usage({
    inputTokens: usage.promptTokenCount,
    outputTokens,
@@ -324,14 +324,7 @@ const mapFinishReason = (finishReason: string | undefined, hasToolCalls: boolean
 const finish = (state: ParserState): ReadonlyArray<LLMEvent> =>
  state.finishReason || state.usage
-    ? (() => {
+    ? [LLMEvent.requestFinish({ 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) => {
@@ -348,21 +341,21 @@ 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) {
-      lifecycle = part.thought
+      events.push(
-        ? Lifecycle.reasoningDelta(lifecycle, events, "reasoning-0", part.text)
+        part.thought
-        : Lifecycle.textDelta(lifecycle, events, "text-0", part.text)
+          ? LLMEvent.reasoningDelta({ id: "reasoning-0", text: part.text })
          : LLMEvent.textDelta({ id: "text-0", text: part.text }),
      )
      continue
    }
    if ("functionCall" in part) {
      const input = part.functionCall.args
      const id = `tool_${nextToolCallId++}`
      lifecycle = Lifecycle.stepStart(lifecycle, events)
      events.push(LLMEvent.toolCall({ id, name: part.functionCall.name, input }))
      hasToolCalls = true
    }
@@ -372,7 +365,6 @@ const step = (state: ParserState, event: GeminiEvent) => {
    {
      ...nextState,
      hasToolCalls,
      lifecycle,
      nextToolCallId,
      finishReason: candidate.finishReason ?? nextState.finishReason,
    },
@@ -396,7 +388,7 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: Protocol.jsonEvent(GeminiEvent),
-    initial: () => ({ hasToolCalls: false, nextToolCallId: 0, lifecycle: Lifecycle.initial() }),
+    initial: () => ({ hasToolCalls: false, nextToolCallId: 0 }),
    step,
    onHalt: finish,
  },
@@ -16,7 +16,6 @@ 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"
@@ -148,7 +147,6 @@ interface ParserState {
  readonly toolCallEvents: ReadonlyArray<LLMEvent>
  readonly usage?: Usage
  readonly finishReason?: FinishReason
  readonly lifecycle: Lifecycle.State
 }
 const invalid = ProviderShared.invalidRequest
@@ -323,9 +321,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
    const toolDeltas = delta?.tool_calls ?? []
    let tools = state.tools
-    let lifecycle = state.lifecycle
+    if (delta?.content) events.push(LLMEvent.textDelta({ id: "text-0", text: delta.content }))
    if (delta?.content) lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.content)
    for (const tool of toolDeltas) {
      const result = ToolStream.appendOrStart(
@@ -337,8 +333,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
      )
      if (ToolStream.isError(result)) return yield* result
      tools = result.tools
-      if (result.events.length) lifecycle = Lifecycle.stepStart(lifecycle, events)
+      if (result.event) events.push(result.event)
      events.push(...result.events)
    }
    // Finalize accumulated tool inputs eagerly when finish_reason arrives so
@@ -354,20 +349,15 @@ 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
-  const lifecycle = state.toolCallEvents.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
+  return [...state.toolCallEvents, ...(reason ? [LLMEvent.requestFinish({ reason, usage: state.usage })] : [])]
  events.push(...state.toolCallEvents)
  if (reason) Lifecycle.finish(lifecycle, events, { reason, usage: state.usage })
  return events
 }
 // =============================================================================
@@ -387,7 +377,7 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: Protocol.jsonEvent(OpenAIChatEvent),
-    initial: () => ({ tools: ToolStream.empty<number>(), toolCallEvents: [], lifecycle: Lifecycle.initial() }),
+    initial: () => ({ tools: ToolStream.empty<number>(), toolCallEvents: [] }),
    step,
    onHalt: finishEvents,
  },
@@ -17,7 +17,6 @@ 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"
@@ -166,7 +165,6 @@ 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
@@ -387,32 +385,23 @@ 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, [LLMEvent.textDelta({ id: event.item_id ?? "text-0", text: event.delta })]]
  return [
    { ...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,
+        providerMetadata: openaiMetadata({ itemId: item.id }),
      }),
    },
-    [...events, LLMEvent.toolInputStart({ id: item.call_id ?? item.id, name: item.name ?? "", providerMetadata })],
+    NO_EVENTS,
  ]
 }
@@ -429,10 +418,10 @@ const onFunctionCallArgumentsDelta = Effect.fn("OpenAIResponses.onFunctionCallAr
    "OpenAI Responses tool argument delta is missing its tool call",
  )
  if (ToolStream.isError(result)) return yield* result
-  const events: LLMEvent[] = []
+  return [
-  const lifecycle = result.events.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
+    { hasFunctionCall: state.hasFunctionCall, tools: result.tools },
-  events.push(...result.events)
+    result.event ? [result.event] : NO_EVENTS,
-  return [{ ...state, lifecycle, tools: result.tools }, events] satisfies StepResult
+  ] satisfies StepResult
 })
 const onOutputItemDone = Effect.fn("OpenAIResponses.onOutputItemDone")(function* (
@@ -451,34 +440,21 @@ 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 },
-        ...state,
+      result.event ? [result.event] : NO_EVENTS,
        lifecycle,
        hasFunctionCall: resultEvents.some(LLMEvent.is.toolCall) ? true : state.hasFunctionCall,
        tools: result.tools,
      },
      events,
    ] satisfies StepResult
  }
-  if (isHostedToolItem(item)) {
+  if (isHostedToolItem(item)) return [state, hostedToolEvents(item)] satisfies StepResult
    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 => {
+const onResponseFinish = (state: ParserState, event: OpenAIResponsesEvent): StepResult => [
-  const events: LLMEvent[] = []
+  state,
-  const lifecycle = Lifecycle.finish(state.lifecycle, events, {
+  [
    LLMEvent.requestFinish({
      reason: mapFinishReason(event, state.hasFunctionCall),
      usage: mapUsage(event.response?.usage),
      providerMetadata:
@@ -488,9 +464,9 @@ const onResponseFinish = (state: ParserState, event: OpenAIResponsesEvent): Step
              serviceTier: event.response.service_tier,
            })
          : undefined,
-  })
+    }),
-  return [{ ...state, lifecycle }, events]
+  ],
-}
+]
 const onResponseFailed = (state: ParserState, event: OpenAIResponsesEvent): StepResult => [
  state,
@@ -530,7 +506,7 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: Protocol.jsonEvent(OpenAIResponsesEvent),
-    initial: () => ({ hasFunctionCall: false, tools: ToolStream.empty<string>(), lifecycle: Lifecycle.initial() }),
+    initial: () => ({ hasFunctionCall: false, tools: ToolStream.empty<string>() }),
    step,
    terminal: (event) => TERMINAL_TYPES.has(event.type),
  },
@@ -43,12 +43,10 @@ export interface ToolAccumulator {
 * 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`
+ * Under the `LLM.Usage` contract, `inputTokens` and `outputTokens` are
- * are the non-cached input and visible output only. The provider-supplied
+ * inclusive totals, so the computed fallback already covers cache reads /
- * `total` is the source of truth when present; the computed fallback
+ * writes and reasoning — used mainly for Anthropic-style providers that
- * under-counts cache and reasoning by design and exists mainly so
+ * don't surface a top-level total.
 * 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,
@@ -69,9 +67,13 @@ export const totalTokens = (
 *
 * 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.
+ * provider-native breakdown stays available on `Usage.providerMetadata`
 * for debugging.
 */
-export const subtractTokens = (total: number | undefined, subtrahend: number | undefined): number | undefined => {
+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)
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Kit Langton	c20d070b9a	docs(llm): fix stale references in protocols/shared.ts and gemini.ts - subtractTokens JSDoc said the raw payload lives on Usage.native, but that field was renamed to providerMetadata earlier in this PR. - totalTokens JSDoc still described the abandoned "additive" first-pass contract where inputTokens/outputTokens were non-cached / visible only. We landed on inclusive totals; the fallback already covers cache and reasoning. - Removed a duplicate inline comment in Gemini's mapUsage — the function-level comment already explains the visible/reasoning sum and the undefined-when-incomplete rule.	2026-05-10 22:08:12 -04:00
Kit Langton	d048bd6f4b	test(llm): re-record golden scenarios against live providers Verifies the new Usage mapper code against live provider responses for OpenAI Chat, OpenAI Responses, Anthropic, Gemini, DeepSeek, and TogetherAI — 16 fresh recordings, all assertions pass. No existing cassettes were modified; these populate test slots that were previously skipped in replay mode. Recorded via: set -a; source .env.recorded.local; set +a RECORD=true bun test test/provider/*.recorded.test.ts Redactor stripped all auth headers; no secrets in the cassettes.	2026-05-10 22:03:02 -04:00
Kit Langton	ab9b79ef88	refactor(llm): rename Usage.native to providerMetadata Aligns the escape-hatch field name with `LLMEvent.providerMetadata` used elsewhere in this package (and with AI SDK / pydantic-ai / LangChain conventions for the same idea). Two parallel escape hatches having different names was a wart. The raw payload is now wrapped under the provider key — `{ openai: ... }`, `{ anthropic: ... }`, `{ google: ... }`, `{ bedrock: ... }` — using the existing `ProviderMetadata = Record<string, Record<string, unknown>>` schema rather than a flat record. Same shape as `LLMEvent.providerMetadata`, so consumers downstream can read both with the same code. Anthropic's `mergeUsage` merges the per-provider sub-record across `message_start` and `message_delta` instead of spreading at the top level.	2026-05-10 21:42:09 -04:00
Kit Langton	d4ff331052	refactor(llm): inclusive total + non-overlapping breakdown for Usage Final shape after considering ecosystem conventions: inputTokens — inclusive total (matches AI SDK / OpenAI / LangChain) outputTokens — inclusive total (includes reasoning) nonCachedInputTokens — breakdown: fresh prompt cacheReadInputTokens — breakdown: cache hit cacheWriteInputTokens — breakdown: cache write reasoningTokens — subset of outputTokens Invariant: nonCached + cacheRead + cacheWrite = inputTokens reasoningTokens <= outputTokens Why this shape: - `inputTokens` keeps its AI-SDK / OpenAI semantics, so a reader from any major ecosystem sees the number they expect. - The non-overlapping breakdown fields are populated alongside the inclusive totals — consumers read whichever they need without subtracting. This eliminates the underflow bug class (opencode#26620) structurally without diverging on naming. - Aligns with the AI SDK v3 spec proposal (vercel/ai#9921), which adds exactly this kind of non-overlapping breakdown to address the active ecosystem bugs around cache token double-counting and underflow (pydantic-ai#4364, langfuse#12306/#11979, vercel/ai#8349, langchain#32818, langchainjs#10249). Mappers: - OpenAI Chat / Responses / Bedrock: provider reports inclusive totals natively; mapper derives `nonCachedInputTokens` via `ProviderShared.subtractTokens`. - Gemini: `promptTokenCount` is inclusive; `candidatesTokenCount` is exclusive of `thoughtsTokenCount`, so mapper sums those to produce the inclusive `outputTokens`. Only computes the total when the visible component is reported (avoids fabricating an inclusive number from a partial breakdown). - Anthropic: `input_tokens` is non-cached natively; mapper sums it with cache reads/writes to produce the inclusive `inputTokens`. `output_tokens` is inclusive (Anthropic doesn't break thinking out, so `reasoningTokens` stays undefined). Added a `visibleOutputTokens` getter (clamped `outputTokens - reasoningTokens`) as the one safe escape hatch for consumers wanting the non-reasoning view. Added `ProviderShared.sumTokens` to derive an inclusive total from a non-overlapping breakdown, returning `undefined` when every input is undefined (so we don't fabricate a 0).	2026-05-10 20:39:22 -04:00
Kit Langton	f5d199db62	feat(llm): add Usage.totalInputTokens / totalOutputTokens getters Match the `LLMResponse.text` / `reasoning` / `toolCalls` getter pattern in the same file — `usage.totalInputTokens` reads naturally and lives where the Usage data does. Both sums are monotonic under the additive contract, so callers no longer need to remember which fields are non-overlapping. Test fixtures that previously asserted with `usage: { ... }` plain literals are now wrapped with `new Usage({...})` to match the runtime shape the mappers actually produce (an instance, not a struct).	2026-05-10 19:29:41 -04:00
Kit Langton	0d4f8d126f	refactor(llm): drop Usage.totalInput / totalOutput helpers The additive contract delivers value at the mapper boundary — every field is non-overlapping and non-negative, so any caller summing arbitrary subsets is correct by construction. Two-line helpers that just sum three or two known fields add API surface without paying for themselves, and there are no in-tree consumers today. If v2 wants them at integration time, the right place is a getter on the `Schema.Class` (matching the `LLMResponse.text` / `reasoning` / `toolCalls` pattern in the same file), not a static namespace helper.	2026-05-10 19:15:46 -04:00
Kit Langton	478f3ae50c	refactor(llm): trim Usage helpers + Bedrock subtraction Review pass: - Drop `Pick<>` type aliases on `Usage.totalInput` / `Usage.totalOutput` — the helpers can take `Usage` directly since every field is optional. - Collapse Bedrock's nested `subtractTokens(subtractTokens(...))` into a single subtraction against the summed cache subtotals. - Drop arithmetic-walkthrough comments in test fixtures (the raw fixture values are right next to the expected outputs). - Generalize the comment on `mapUsage` in `openai-chat.ts` so the rationale outlives the PR reference.	2026-05-10 13:22:49 -04:00
Kit Langton	b9451175a6	refactor(llm): make LLM.Usage a fully-additive contract Defines a single invariant for `LLM.Usage`: every field is non-negative and every meaningful aggregate is a sum, never a difference. Total billable input = inputTokens + cacheReadInputTokens + cacheWriteInputTokens. Total billable output = outputTokens + reasoningTokens. Adding two non-negatives cannot underflow, so consumers can no longer reproduce the underflow-then-clamp bug class fixed by #26620. Each protocol mapper now enforces the contract at the provider boundary via `ProviderShared.subtractTokens`, which clamps with `Math.max(0, …)` for defense against provider bugs: - OpenAI Chat / Responses: pull `cached_tokens` out of `prompt_tokens` / `input_tokens`; pull `reasoning_tokens` out of `completion_tokens` / `output_tokens`. The provider's `total_tokens` is preserved verbatim. - Gemini: pull `cachedContentTokenCount` out of `promptTokenCount`. Gemini already split visible candidates from thoughts. - Bedrock: pull `cacheReadInputTokens` and `cacheWriteInputTokens` out of `inputTokens`, matching AWS prompt-caching docs. - Anthropic: already non-overlapping per the Messages API; pass through. Adds `Usage.totalInput` / `Usage.totalOutput` helpers for callers that want the merged view, and a regression test covering the clamp behavior. The reasoning underflow fixed in #26620 was the most visible symptom of a broader semantic inconsistency in this package: providers also disagreed on whether `inputTokens` includes cache reads (Anthropic excluded; OpenAI/Gemini/Bedrock included), which would silently double-subtract the moment v2 wired LLM.Usage into Session.getUsage. Normalizing now, pre-integration, closes both holes in one move.	2026-05-10 13:07:58 -04:00