refactor(summary): remove unnecessary Layer.unwrap from defaultLayer

fix(account): coalesce concurrent console token refreshes (#20503 )
test(app): migrate more e2e suites to isolated backend (#20505 )
2026-04-01 14:21:29 -04:00 · 2026-04-01 13:16:35 -04:00 · 2026-04-01 13:15:42 -04:00 · 2026-04-01 16:50:21 +00:00 · 2026-04-01 16:47:33 +00:00 · 2026-04-01 16:14:37 +00:00
33 changed files with 1445 additions and 919 deletions
@@ -11,6 +11,7 @@ adamdotdevin
 -agusbasari29 AI PR slop
 ariane-emory
 -atharvau AI review spamming literally every PR
+-borealbytes
 -danieljoshuanazareth
 -danieljoshuanazareth
 edemaine
@@ -612,7 +612,7 @@
  },
  "catalog": {
    "@cloudflare/workers-types": "4.20251008.0",
-    "@effect/platform-node": "4.0.0-beta.42",
+    "@effect/platform-node": "4.0.0-beta.43",
    "@hono/zod-validator": "0.4.2",
    "@kobalte/core": "0.13.11",
    "@octokit/rest": "22.0.0",
@@ -636,7 +636,7 @@
    "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.42",
+    "effect": "4.0.0-beta.43",
    "fuzzysort": "3.1.0",
    "hono": "4.10.7",
    "hono-openapi": "1.1.2",
@@ -995,9 +995,9 @@

    "@effect/language-service": ["@effect/language-service@0.79.0", "", { "bin": { "effect-language-service": "cli.js" } }, "sha512-DEmIOsg1GjjP6s9HXH1oJrW+gDmzkhVv9WOZl6to5eNyyCrjz1S2PDqQ7aYrW/HuifhfwI5Bik1pK4pj7Z+lrg=="],

-    "@effect/platform-node": ["@effect/platform-node@4.0.0-beta.42", "", { "dependencies": { "@effect/platform-node-shared": "^4.0.0-beta.42", "mime": "^4.1.0", "undici": "^7.24.0" }, "peerDependencies": { "effect": "^4.0.0-beta.42", "ioredis": "^5.7.0" } }, "sha512-kbdRML2FBa4q8U8rZQcnmLKZ5zN/z1bAA7t5D1/UsBHZqJgnfRgu1CP6kaEfb1Nie6YyaWshxTktZQryjvW/Yg=="],
+    "@effect/platform-node": ["@effect/platform-node@4.0.0-beta.43", "", { "dependencies": { "@effect/platform-node-shared": "^4.0.0-beta.43", "mime": "^4.1.0", "undici": "^7.24.0" }, "peerDependencies": { "effect": "^4.0.0-beta.43", "ioredis": "^5.7.0" } }, "sha512-Uq6E1rjaIpjHauzjwoB2HzAg3battYt2Boy8XO50GoHiWCXKE6WapYZ0/AnaBx5v5qg2sOfqpuiLsUf9ZgxOkA=="],

-    "@effect/platform-node-shared": ["@effect/platform-node-shared@4.0.0-beta.42", "", { "dependencies": { "@types/ws": "^8.18.1", "ws": "^8.19.0" }, "peerDependencies": { "effect": "^4.0.0-beta.42" } }, "sha512-PC+lxLsrwob3+nBChAPrQq32olCeyApgXBvs1NrRsoArLViNT76T/68CttuCAksCZj5e1bZ1ZibLPel3vUmx2g=="],
+    "@effect/platform-node-shared": ["@effect/platform-node-shared@4.0.0-beta.43", "", { "dependencies": { "@types/ws": "^8.18.1", "ws": "^8.19.0" }, "peerDependencies": { "effect": "^4.0.0-beta.43" } }, "sha512-A9q0GEb61pYcQ06Dr6gXj1nKlDI3KHsar1sk3qb1ZY+kVSR64tBAylI8zGon23KY+NPtTUj/sEIToB7jc3Qt5w=="],

    "@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=="],

@@ -2771,7 +2771,7 @@

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

-    "effect": ["effect@4.0.0-beta.42", "", { "dependencies": { "@standard-schema/spec": "^1.1.0", "fast-check": "^4.5.3", "find-my-way-ts": "^0.1.6", "ini": "^6.0.0", "kubernetes-types": "^1.30.0", "msgpackr": "^1.11.8", "multipasta": "^0.2.7", "toml": "^3.0.0", "uuid": "^13.0.0", "yaml": "^2.8.2" } }, "sha512-c1UrRP+tLzyHb4Fepl8XBDJlLQLkrcMXrRBba441GQRxMbeQ/aIOSFcBwSda1iMJ5l9F0lYc3Bhe33/whrmavQ=="],
+    "effect": ["effect@4.0.0-beta.43", "", { "dependencies": { "@standard-schema/spec": "^1.1.0", "fast-check": "^4.5.3", "find-my-way-ts": "^0.1.6", "ini": "^6.0.0", "kubernetes-types": "^1.30.0", "msgpackr": "^1.11.8", "multipasta": "^0.2.7", "toml": "^3.0.0", "uuid": "^13.0.0", "yaml": "^2.8.2" } }, "sha512-AJYyDimIwJOn87uUz/JzmgDc5GfjxJbXvEbTvNzMa+M3Uer344bLo/O5mMRkqc1vBleA+Ygs4+dbE3QsqOkKTQ=="],

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

@@ -1,8 +1,8 @@
 {
  "nodeModules": {
-    "x86_64-linux": "sha256-C7y5FMI1pGEgMw/vcPoBhK9tw5uGg1bk0gPXPUUVhgU=",
-    "aarch64-linux": "sha256-cUlQ9jp4WIaJkd4GRoHMWc+REG/OnnGCmsQUNmvg4is=",
-    "aarch64-darwin": "sha256-3GXmqG7yihJ91wS/jlW19qxGI62b1bFJnpGB4LcMlpY=",
-    "x86_64-darwin": "sha256-cUF0TfYg2nXnU80kWFpr9kNHlu9txiatIgrHTltgx4g="
+    "x86_64-linux": "sha256-bjfe8/aD0hvUQQEfaNdmKV/Y3dzpf8oz1OUJdgf61WI=",
+    "aarch64-linux": "sha256-iU9v+ekSCB/qTUG+pOOpSMhPh+0hWnWU5jzDNllEkxU=",
+    "aarch64-darwin": "sha256-SgNydQLeAjbX0J49f2VKcgKg2Y30pK826R2qQJBMWE4=",
+    "x86_64-darwin": "sha256-/rzwNuI9x55qi0UcU7QvPUTupErmkt62T09g1omXkQk="
  }
 }
@@ -25,7 +25,7 @@
      "packages/slack"
    ],
    "catalog": {
-      "@effect/platform-node": "4.0.0-beta.42",
+      "@effect/platform-node": "4.0.0-beta.43",
      "@types/bun": "1.3.11",
      "@octokit/rest": "22.0.0",
      "@hono/zod-validator": "0.4.2",
@@ -45,7 +45,7 @@
      "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.42",
+      "effect": "4.0.0-beta.43",
      "ai": "6.0.138",
      "hono": "4.10.7",
      "hono-openapi": "1.1.2",
@@ -312,10 +312,11 @@ export async function openSettings(page: Page) {
  return dialog
 }

-export async function seedProjects(page: Page, input: { directory: string; extra?: string[] }) {
+export async function seedProjects(page: Page, input: { directory: string; extra?: string[]; serverUrl?: string }) {
  await page.addInitScript(
    (args: { directory: string; serverUrl: string; extra: string[] }) => {
      const key = "opencode.global.dat:server"
+      const defaultKey = "opencode.settings.dat:defaultServerUrl"
      const raw = localStorage.getItem(key)
      const parsed = (() => {
        if (!raw) return undefined
@@ -331,6 +332,7 @@ export async function seedProjects(page: Page, input: { directory: string; extra
      const lastProject = store.lastProject && typeof store.lastProject === "object" ? store.lastProject : {}
      const projects = store.projects && typeof store.projects === "object" ? store.projects : {}
      const nextProjects = { ...(projects as Record<string, unknown>) }
+      const nextList = list.includes(args.serverUrl) ? list : [args.serverUrl, ...list]

      const add = (origin: string, directory: string) => {
        const current = nextProjects[origin]
@@ -356,17 +358,18 @@ export async function seedProjects(page: Page, input: { directory: string; extra
      localStorage.setItem(
        key,
        JSON.stringify({
-          list,
+          list: nextList,
          projects: nextProjects,
          lastProject,
        }),
      )
+      localStorage.setItem(defaultKey, args.serverUrl)
    },
-    { directory: input.directory, serverUrl, extra: input.extra ?? [] },
+    { directory: input.directory, serverUrl: input.serverUrl ?? serverUrl, extra: input.extra ?? [] },
  )
 }

-export async function createTestProject() {
+export async function createTestProject(input?: { serverUrl?: string }) {
  const root = await fs.mkdtemp(path.join(os.tmpdir(), "opencode-e2e-project-"))
  const id = `e2e-${path.basename(root)}`

@@ -381,7 +384,7 @@ export async function createTestProject() {
    stdio: "ignore",
  })

-  return resolveDirectory(root)
+  return resolveDirectory(root, input?.serverUrl)
 }

 export async function cleanupTestProject(directory: string) {
@@ -430,22 +433,22 @@ export async function waitSlug(page: Page, skip: string[] = []) {
  return next
 }

-export async function resolveSlug(slug: string) {
+export async function resolveSlug(slug: string, input?: { serverUrl?: string }) {
  const directory = base64Decode(slug)
  if (!directory) throw new Error(`Failed to decode workspace slug: ${slug}`)
-  const resolved = await resolveDirectory(directory)
+  const resolved = await resolveDirectory(directory, input?.serverUrl)
  return { directory: resolved, slug: base64Encode(resolved), raw: slug }
 }

-export async function waitDir(page: Page, directory: string) {
-  const target = await resolveDirectory(directory)
+export async function waitDir(page: Page, directory: string, input?: { serverUrl?: string }) {
+  const target = await resolveDirectory(directory, input?.serverUrl)
  await expect
    .poll(
      async () => {
        await assertHealthy(page, "waitDir")
        const slug = slugFromUrl(page.url())
        if (!slug) return ""
-        return resolveSlug(slug)
+        return resolveSlug(slug, input)
          .then((item) => item.directory)
          .catch(() => "")
      },
@@ -455,15 +458,15 @@ export async function waitDir(page: Page, directory: string) {
  return { directory: target, slug: base64Encode(target) }
 }

-export async function waitSession(page: Page, input: { directory: string; sessionID?: string }) {
-  const target = await resolveDirectory(input.directory)
+export async function waitSession(page: Page, input: { directory: string; sessionID?: string; serverUrl?: string }) {
+  const target = await resolveDirectory(input.directory, input.serverUrl)
  await expect
    .poll(
      async () => {
        await assertHealthy(page, "waitSession")
        const slug = slugFromUrl(page.url())
        if (!slug) return false
-        const resolved = await resolveSlug(slug).catch(() => undefined)
+        const resolved = await resolveSlug(slug, { serverUrl: input.serverUrl }).catch(() => undefined)
        if (!resolved || resolved.directory !== target) return false
        const current = sessionIDFromUrl(page.url())
        if (input.sessionID && current !== input.sessionID) return false
@@ -473,7 +476,7 @@ export async function waitSession(page: Page, input: { directory: string; sessio
        if (input.sessionID && (!state || state.sessionID !== input.sessionID)) return false
        if (!input.sessionID && state?.sessionID) return false
        if (state?.dir) {
-          const dir = await resolveDirectory(state.dir).catch(() => state.dir ?? "")
+          const dir = await resolveDirectory(state.dir, input.serverUrl).catch(() => state.dir ?? "")
          if (dir !== target) return false
        }

@@ -489,9 +492,9 @@ export async function waitSession(page: Page, input: { directory: string; sessio
  return { directory: target, slug: base64Encode(target) }
 }

-export async function waitSessionSaved(directory: string, sessionID: string, timeout = 30_000) {
-  const sdk = createSdk(directory)
-  const target = await resolveDirectory(directory)
+export async function waitSessionSaved(directory: string, sessionID: string, timeout = 30_000, serverUrl?: string) {
+  const sdk = createSdk(directory, serverUrl)
+  const target = await resolveDirectory(directory, serverUrl)

  await expect
    .poll(
@@ -501,7 +504,7 @@ export async function waitSessionSaved(directory: string, sessionID: string, tim
          .then((x) => x.data)
          .catch(() => undefined)
        if (!data?.directory) return ""
-        return resolveDirectory(data.directory).catch(() => data.directory)
+        return resolveDirectory(data.directory, serverUrl).catch(() => data.directory)
      },
      { timeout },
    )
@@ -666,8 +669,9 @@ export async function cleanupSession(input: {
  sessionID: string
  directory?: string
  sdk?: ReturnType<typeof createSdk>
+  serverUrl?: string
 }) {
-  const sdk = input.sdk ?? (input.directory ? createSdk(input.directory) : undefined)
+  const sdk = input.sdk ?? (input.directory ? createSdk(input.directory, input.serverUrl) : undefined)
  if (!sdk) throw new Error("cleanupSession requires sdk or directory")
  await waitSessionIdle(sdk, input.sessionID, 5_000).catch(() => undefined)
  const current = await status(sdk, input.sessionID).catch(() => undefined)
@@ -1019,3 +1023,13 @@ export async function openWorkspaceMenu(page: Page, workspaceSlug: string) {
  await expect(menu).toBeVisible()
  return menu
 }
+
+export async function assistantText(sdk: ReturnType<typeof createSdk>, sessionID: string) {
+  const messages = await sdk.session.messages({ sessionID, limit: 50 }).then((r) => r.data ?? [])
+  return messages
+    .filter((m) => m.info.role === "assistant")
+    .flatMap((m) => m.parts)
+    .filter((p) => p.type === "text")
+    .map((p) => p.text)
+    .join("\n")
+}
@@ -0,0 +1,125 @@
+import { spawn } from "node:child_process"
+import fs from "node:fs/promises"
+import net from "node:net"
+import os from "node:os"
+import path from "node:path"
+import { fileURLToPath } from "node:url"
+
+type Handle = {
+  url: string
+  stop: () => Promise<void>
+}
+
+function freePort() {
+  return new Promise<number>((resolve, reject) => {
+    const server = net.createServer()
+    server.once("error", reject)
+    server.listen(0, () => {
+      const address = server.address()
+      if (!address || typeof address === "string") {
+        server.close(() => reject(new Error("Failed to acquire a free port")))
+        return
+      }
+      server.close((err) => {
+        if (err) reject(err)
+        else resolve(address.port)
+      })
+    })
+  })
+}
+
+async function waitForHealth(url: string, probe = "/global/health") {
+  const end = Date.now() + 120_000
+  let last = ""
+  while (Date.now() < end) {
+    try {
+      const res = await fetch(`${url}${probe}`)
+      if (res.ok) return
+      last = `status ${res.status}`
+    } catch (err) {
+      last = err instanceof Error ? err.message : String(err)
+    }
+    await new Promise((resolve) => setTimeout(resolve, 250))
+  }
+  throw new Error(`Timed out waiting for backend health at ${url}${probe}${last ? ` (${last})` : ""}`)
+}
+
+const LOG_CAP = 100
+
+function cap(input: string[]) {
+  if (input.length > LOG_CAP) input.splice(0, input.length - LOG_CAP)
+}
+
+function tail(input: string[]) {
+  return input.slice(-40).join("")
+}
+
+export async function startBackend(label: string): Promise<Handle> {
+  const port = await freePort()
+  const sandbox = await fs.mkdtemp(path.join(os.tmpdir(), `opencode-e2e-${label}-`))
+  const appDir = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..")
+  const repoDir = path.resolve(appDir, "../..")
+  const opencodeDir = path.join(repoDir, "packages", "opencode")
+  const env = {
+    ...process.env,
+    OPENCODE_DISABLE_SHARE: process.env.OPENCODE_DISABLE_SHARE ?? "true",
+    OPENCODE_DISABLE_LSP_DOWNLOAD: "true",
+    OPENCODE_DISABLE_DEFAULT_PLUGINS: "true",
+    OPENCODE_EXPERIMENTAL_DISABLE_FILEWATCHER: "true",
+    OPENCODE_TEST_HOME: path.join(sandbox, "home"),
+    XDG_DATA_HOME: path.join(sandbox, "share"),
+    XDG_CACHE_HOME: path.join(sandbox, "cache"),
+    XDG_CONFIG_HOME: path.join(sandbox, "config"),
+    XDG_STATE_HOME: path.join(sandbox, "state"),
+    OPENCODE_CLIENT: "app",
+    OPENCODE_STRICT_CONFIG_DEPS: "true",
+  } satisfies Record<string, string | undefined>
+  const out: string[] = []
+  const err: string[] = []
+  const proc = spawn(
+    "bun",
+    ["run", "--conditions=browser", "./src/index.ts", "serve", "--port", String(port), "--hostname", "127.0.0.1"],
+    {
+      cwd: opencodeDir,
+      env,
+      stdio: ["ignore", "pipe", "pipe"],
+    },
+  )
+  proc.stdout?.on("data", (chunk) => {
+    out.push(String(chunk))
+    cap(out)
+  })
+  proc.stderr?.on("data", (chunk) => {
+    err.push(String(chunk))
+    cap(err)
+  })
+
+  const url = `http://127.0.0.1:${port}`
+  try {
+    await waitForHealth(url)
+  } catch (error) {
+    proc.kill("SIGTERM")
+    await fs.rm(sandbox, { recursive: true, force: true }).catch(() => undefined)
+    throw new Error(
+      [
+        `Failed to start isolated e2e backend for ${label}`,
+        error instanceof Error ? error.message : String(error),
+        tail(out),
+        tail(err),
+      ]
+        .filter(Boolean)
+        .join("\n"),
+    )
+  }
+
+  return {
+    url,
+    async stop() {
+      if (proc.exitCode === null) {
+        proc.kill("SIGTERM")
+        await new Promise((resolve) => proc.once("exit", () => resolve(undefined))).catch(() => undefined)
+      }
+      await fs.rm(sandbox, { recursive: true, force: true }).catch(() => undefined)
+    },
+  }
+}
@@ -3,6 +3,7 @@ import { ManagedRuntime } from "effect"
 import type { E2EWindow } from "../src/testing/terminal"
 import type { Item, Reply, Usage } from "../../opencode/test/lib/llm-server"
 import { TestLLMServer } from "../../opencode/test/lib/llm-server"
+import { startBackend } from "./backend"
 import {
  healthPhase,
  cleanupSession,
@@ -19,6 +20,20 @@ import { createSdk, dirSlug, getWorktree, sessionPath } from "./utils"
 type LLMFixture = {
  url: string
  push: (...input: (Item | Reply)[]) => Promise<void>
+  pushMatch: (
+    match: (hit: { url: URL; body: Record<string, unknown> }) => boolean,
+    ...input: (Item | Reply)[]
+  ) => Promise<void>
+  textMatch: (
+    match: (hit: { url: URL; body: Record<string, unknown> }) => boolean,
+    value: string,
+    opts?: { usage?: Usage },
+  ) => Promise<void>
+  toolMatch: (
+    match: (hit: { url: URL; body: Record<string, unknown> }) => boolean,
+    name: string,
+    input: unknown,
+  ) => Promise<void>
  text: (value: string, opts?: { usage?: Usage }) => Promise<void>
  tool: (name: string, input: unknown) => Promise<void>
  toolHang: (name: string, input: unknown) => Promise<void>
@@ -46,32 +61,54 @@ const seedModel = (() => {
  }
 })()

+type ProjectHandle = {
+  directory: string
+  slug: string
+  gotoSession: (sessionID?: string) => Promise<void>
+  trackSession: (sessionID: string, directory?: string) => void
+  trackDirectory: (directory: string) => void
+  sdk: ReturnType<typeof createSdk>
+}
+
+type ProjectOptions = {
+  extra?: string[]
+  model?: { providerID: string; modelID: string }
+  setup?: (directory: string) => Promise<void>
+  beforeGoto?: (project: { directory: string; sdk: ReturnType<typeof createSdk> }) => Promise<void>
+}
+
 type TestFixtures = {
  llm: LLMFixture
  sdk: ReturnType<typeof createSdk>
  gotoSession: (sessionID?: string) => Promise<void>
-  withProject: <T>(
-    callback: (project: {
-      directory: string
-      slug: string
-      gotoSession: (sessionID?: string) => Promise<void>
-      trackSession: (sessionID: string, directory?: string) => void
-      trackDirectory: (directory: string) => void
-    }) => Promise<T>,
-    options?: {
-      extra?: string[]
-      model?: { providerID: string; modelID: string }
-      setup?: (directory: string) => Promise<void>
-    },
-  ) => Promise<T>
+  withProject: <T>(callback: (project: ProjectHandle) => Promise<T>, options?: ProjectOptions) => Promise<T>
+  withBackendProject: <T>(callback: (project: ProjectHandle) => Promise<T>, options?: ProjectOptions) => Promise<T>
 }

 type WorkerFixtures = {
+  backend: {
+    url: string
+    sdk: (directory?: string) => ReturnType<typeof createSdk>
+  }
  directory: string
  slug: string
 }

 export const test = base.extend<TestFixtures, WorkerFixtures>({
+  backend: [
+    async ({}, use, workerInfo) => {
+      const handle = await startBackend(`w${workerInfo.workerIndex}`)
+      try {
+        await use({
+          url: handle.url,
+          sdk: (directory?: string) => createSdk(directory, handle.url),
+        })
+      } finally {
+        await handle.stop()
+      }
+    },
+    { scope: "worker" },
+  ],
  llm: async ({}, use) => {
    const rt = ManagedRuntime.make(TestLLMServer.layer)
    try {
@@ -79,6 +116,9 @@ export const test = base.extend<TestFixtures, WorkerFixtures>({
      await use({
        url: svc.url,
        push: (...input) => rt.runPromise(svc.push(...input)),
+        pushMatch: (match, ...input) => rt.runPromise(svc.pushMatch(match, ...input)),
+        textMatch: (match, value, opts) => rt.runPromise(svc.textMatch(match, value, opts)),
+        toolMatch: (match, name, input) => rt.runPromise(svc.toolMatch(match, name, input)),
        text: (value, opts) => rt.runPromise(svc.text(value, opts)),
        tool: (name, input) => rt.runPromise(svc.tool(name, input)),
        toolHang: (name, input) => rt.runPromise(svc.toolHang(name, input)),
@@ -146,51 +186,74 @@ export const test = base.extend<TestFixtures, WorkerFixtures>({
    await use(gotoSession)
  },
  withProject: async ({ page }, use) => {
-    await use(async (callback, options) => {
-      const root = await createTestProject()
-      const sessions = new Map<string, string>()
-      const dirs = new Set<string>()
-      await options?.setup?.(root)
-      await seedStorage(page, { directory: root, extra: options?.extra, model: options?.model })
-
-      const gotoSession = async (sessionID?: string) => {
-        await page.goto(sessionPath(root, sessionID))
-        await waitSession(page, { directory: root, sessionID })
-        const current = sessionIDFromUrl(page.url())
-        if (current) trackSession(current)
-      }
-
-      const trackSession = (sessionID: string, directory?: string) => {
-        sessions.set(sessionID, directory ?? root)
-      }
-
-      const trackDirectory = (directory: string) => {
-        if (directory !== root) dirs.add(directory)
-      }
-
-      try {
-        await gotoSession()
-        const slug = await waitSlug(page)
-        return await callback({ directory: root, slug, gotoSession, trackSession, trackDirectory })
-      } finally {
-        setHealthPhase(page, "cleanup")
-        await Promise.allSettled(
-          Array.from(sessions, ([sessionID, directory]) => cleanupSession({ sessionID, directory })),
-        )
-        await Promise.allSettled(Array.from(dirs, (directory) => cleanupTestProject(directory)))
-        await cleanupTestProject(root)
-        setHealthPhase(page, "test")
-      }
-    })
+    await use((callback, options) => runProject(page, callback, options))
+  },
+  withBackendProject: async ({ page, backend }, use) => {
+    await use((callback, options) =>
+      runProject(page, callback, { ...options, serverUrl: backend.url, sdk: backend.sdk }),
+    )
  },
 })

+async function runProject<T>(
+  page: Page,
+  callback: (project: ProjectHandle) => Promise<T>,
+  options?: ProjectOptions & {
+    serverUrl?: string
+    sdk?: (directory?: string) => ReturnType<typeof createSdk>
+  },
+) {
+  const url = options?.serverUrl
+  const root = await createTestProject(url ? { serverUrl: url } : undefined)
+  const sdk = options?.sdk?.(root) ?? createSdk(root, url)
+  const sessions = new Map<string, string>()
+  const dirs = new Set<string>()
+  await options?.setup?.(root)
+  await seedStorage(page, {
+    directory: root,
+    extra: options?.extra,
+    model: options?.model,
+    serverUrl: url,
+  })
+
+  const gotoSession = async (sessionID?: string) => {
+    await page.goto(sessionPath(root, sessionID))
+    await waitSession(page, { directory: root, sessionID, serverUrl: url })
+    const current = sessionIDFromUrl(page.url())
+    if (current) trackSession(current)
+  }
+
+  const trackSession = (sessionID: string, directory?: string) => {
+    sessions.set(sessionID, directory ?? root)
+  }
+
+  const trackDirectory = (directory: string) => {
+    if (directory !== root) dirs.add(directory)
+  }
+
+  try {
+    await options?.beforeGoto?.({ directory: root, sdk })
+    await gotoSession()
+    const slug = await waitSlug(page)
+    return await callback({ directory: root, slug, gotoSession, trackSession, trackDirectory, sdk })
+  } finally {
+    setHealthPhase(page, "cleanup")
+    await Promise.allSettled(
+      Array.from(sessions, ([sessionID, directory]) => cleanupSession({ sessionID, directory, serverUrl: url })),
+    )
+    await Promise.allSettled(Array.from(dirs, (directory) => cleanupTestProject(directory)))
+    await cleanupTestProject(root)
+    setHealthPhase(page, "test")
+  }
+}
+
 async function seedStorage(
  page: Page,
  input: {
    directory: string
    extra?: string[]
    model?: { providerID: string; modelID: string }
+    serverUrl?: string
  },
 ) {
  await seedProjects(page, input)
@@ -0,0 +1,46 @@
+import { createSdk } from "../utils"
+
+export const openaiModel = { providerID: "openai", modelID: "gpt-5.3-chat-latest" }
+
+type Hit = { body: Record<string, unknown> }
+
+export function bodyText(hit: Hit) {
+  return JSON.stringify(hit.body)
+}
+
+export function titleMatch(hit: Hit) {
+  return bodyText(hit).includes("Generate a title for this conversation")
+}
+
+export function promptMatch(token: string) {
+  return (hit: Hit) => bodyText(hit).includes(token)
+}
+
+export async function withMockOpenAI<T>(input: { serverUrl: string; llmUrl: string; fn: () => Promise<T> }) {
+  const sdk = createSdk(undefined, input.serverUrl)
+  const prev = await sdk.global.config.get().then((res) => res.data ?? {})
+
+  try {
+    await sdk.global.config.update({
+      config: {
+        ...prev,
+        model: `${openaiModel.providerID}/${openaiModel.modelID}`,
+        enabled_providers: ["openai"],
+        provider: {
+          ...prev.provider,
+          openai: {
+            ...prev.provider?.openai,
+            options: {
+              ...prev.provider?.openai?.options,
+              apiKey: "test-key",
+              baseURL: input.llmUrl,
+            },
+          },
+        },
+      },
+    })
+    return await input.fn()
+  } finally {
+    await sdk.global.config.update({ config: prev })
+  }
+}
@@ -1,47 +1,52 @@
 import { test, expect } from "../fixtures"
 import { promptSelector } from "../selectors"
-import { cleanupSession, sessionIDFromUrl, withSession } from "../actions"
+import { assistantText, sessionIDFromUrl, withSession } from "../actions"
+import { openaiModel, promptMatch, titleMatch, withMockOpenAI } from "./mock"

 const text = (value: string | null) => (value ?? "").replace(/\u200B/g, "").trim()

 // Regression test for Issue #12453: the synchronous POST /message endpoint holds
 // the connection open while the agent works, causing "Failed to fetch" over
 // VPN/Tailscale. The fix switches to POST /prompt_async which returns immediately.
-test("prompt succeeds when sync message endpoint is unreachable", async ({ page, sdk, gotoSession }) => {
+test("prompt succeeds when sync message endpoint is unreachable", async ({
+  page,
+  llm,
+  backend,
+  withBackendProject,
+}) => {
  test.setTimeout(120_000)

  // Simulate Tailscale/VPN killing the long-lived sync connection
  await page.route("**/session/*/message", (route) => route.abort("connectionfailed"))

-  await gotoSession()
+  await withMockOpenAI({
+    serverUrl: backend.url,
+    llmUrl: llm.url,
+    fn: async () => {
+      const token = `E2E_ASYNC_${Date.now()}`
+      await llm.textMatch(titleMatch, "E2E Title")
+      await llm.textMatch(promptMatch(token), token)

-  const token = `E2E_ASYNC_${Date.now()}`
-  await page.locator(promptSelector).click()
-  await page.keyboard.type(`Reply with exactly: ${token}`)
-  await page.keyboard.press("Enter")
+      await withBackendProject(
+        async (project) => {
+          await page.locator(promptSelector).click()
+          await page.keyboard.type(`Reply with exactly: ${token}`)
+          await page.keyboard.press("Enter")

-  await expect(page).toHaveURL(/\/session\/[^/?#]+/, { timeout: 30_000 })
-  const sessionID = sessionIDFromUrl(page.url())!
+          await expect(page).toHaveURL(/\/session\/[^/?#]+/, { timeout: 30_000 })
+          const sessionID = sessionIDFromUrl(page.url())!
+          project.trackSession(sessionID)

-  try {
-    // Agent response arrives via SSE despite sync endpoint being dead
-    await expect
-      .poll(
-        async () => {
-          const messages = await sdk.session.messages({ sessionID, limit: 50 }).then((r) => r.data ?? [])
-          return messages
-            .filter((m) => m.info.role === "assistant")
-            .flatMap((m) => m.parts)
-            .filter((p) => p.type === "text")
-            .map((p) => p.text)
-            .join("\n")
+          await expect.poll(() => llm.calls()).toBeGreaterThanOrEqual(1)
+
+          await expect.poll(() => assistantText(project.sdk, sessionID), { timeout: 90_000 }).toContain(token)
+        },
+        {
+          model: openaiModel,
        },
-        { timeout: 90_000 },
      )
-      .toContain(token)
-  } finally {
-    await cleanupSession({ sdk, sessionID })
-  }
+    },
+  })
 })

 test("failed prompt send restores the composer input", async ({ page, sdk, gotoSession }) => {
@@ -1,8 +1,9 @@
 import type { ToolPart } from "@opencode-ai/sdk/v2/client"
 import type { Page } from "@playwright/test"
 import { test, expect } from "../fixtures"
-import { withSession } from "../actions"
+import { assistantText, sessionIDFromUrl } from "../actions"
 import { promptSelector } from "../selectors"
+import { openaiModel, promptMatch, titleMatch, withMockOpenAI } from "./mock"

 const text = (value: string | null) => (value ?? "").replace(/\u200B/g, "").trim()

@@ -43,20 +44,13 @@ async function wait(page: Page, value: string) {
  await expect.poll(async () => text(await page.locator(promptSelector).textContent())).toBe(value)
 }

-async function reply(sdk: Parameters<typeof withSession>[0], sessionID: string, token: string) {
+async function reply(
+  sdk: { session: { messages: Parameters<typeof assistantText>[0]["session"] } },
+  sessionID: string,
+  token: string,
+) {
  await expect
-    .poll(
-      async () => {
-        const messages = await sdk.session.messages({ sessionID, limit: 50 }).then((r) => r.data ?? [])
-        return messages
-          .filter((item) => item.info.role === "assistant")
-          .flatMap((item) => item.parts)
-          .filter((item) => item.type === "text")
-          .map((item) => item.text)
-          .join("\n")
-      },
-      { timeout: 90_000 },
-    )
+    .poll(() => assistantText(sdk as Parameters<typeof assistantText>[0], sessionID), { timeout: 90_000 })
    .toContain(token)
 }

@@ -79,106 +73,145 @@ async function shell(sdk: Parameters<typeof withSession>[0], sessionID: string,
    .toContain(token)
 }

-test("prompt history restores unsent draft with arrow navigation", async ({ page, sdk, gotoSession }) => {
+test("prompt history restores unsent draft with arrow navigation", async ({
+  page,
+  llm,
+  backend,
+  withBackendProject,
+}) => {
  test.setTimeout(120_000)

-  await withSession(sdk, `e2e prompt history ${Date.now()}`, async (session) => {
-    await gotoSession(session.id)
+  await withMockOpenAI({
+    serverUrl: backend.url,
+    llmUrl: llm.url,
+    fn: async () => {
+      const firstToken = `E2E_HISTORY_ONE_${Date.now()}`
+      const secondToken = `E2E_HISTORY_TWO_${Date.now()}`
+      const first = `Reply with exactly: ${firstToken}`
+      const second = `Reply with exactly: ${secondToken}`
+      const draft = `draft ${Date.now()}`

-    const prompt = page.locator(promptSelector)
-    const firstToken = `E2E_HISTORY_ONE_${Date.now()}`
-    const secondToken = `E2E_HISTORY_TWO_${Date.now()}`
-    const first = `Reply with exactly: ${firstToken}`
-    const second = `Reply with exactly: ${secondToken}`
-    const draft = `draft ${Date.now()}`
+      await llm.textMatch(titleMatch, "E2E Title")
+      await llm.textMatch(promptMatch(firstToken), firstToken)
+      await llm.textMatch(promptMatch(secondToken), secondToken)

-    await prompt.click()
-    await page.keyboard.type(first)
-    await page.keyboard.press("Enter")
-    await wait(page, "")
-    await reply(sdk, session.id, firstToken)
+      await withBackendProject(
+        async (project) => {
+          const prompt = page.locator(promptSelector)

-    await prompt.click()
-    await page.keyboard.type(second)
-    await page.keyboard.press("Enter")
-    await wait(page, "")
-    await reply(sdk, session.id, secondToken)
+          await prompt.click()
+          await page.keyboard.type(first)
+          await page.keyboard.press("Enter")
+          await wait(page, "")

-    await prompt.click()
-    await page.keyboard.type(draft)
-    await wait(page, draft)
+          await expect(page).toHaveURL(/\/session\/[^/?#]+/, { timeout: 30_000 })
+          const sessionID = sessionIDFromUrl(page.url())!
+          project.trackSession(sessionID)
+          await reply(project.sdk, sessionID, firstToken)

-    // Clear the draft before navigating history (ArrowUp only works when prompt is empty)
-    await prompt.fill("")
-    await wait(page, "")
+          await prompt.click()
+          await page.keyboard.type(second)
+          await page.keyboard.press("Enter")
+          await wait(page, "")
+          await reply(project.sdk, sessionID, secondToken)

-    await page.keyboard.press("ArrowUp")
-    await wait(page, second)
+          await prompt.click()
+          await page.keyboard.type(draft)
+          await wait(page, draft)

-    await page.keyboard.press("ArrowUp")
-    await wait(page, first)
+          await prompt.fill("")
+          await wait(page, "")

-    await page.keyboard.press("ArrowDown")
-    await wait(page, second)
+          await page.keyboard.press("ArrowUp")
+          await wait(page, second)

-    await page.keyboard.press("ArrowDown")
-    await wait(page, "")
+          await page.keyboard.press("ArrowUp")
+          await wait(page, first)
+
+          await page.keyboard.press("ArrowDown")
+          await wait(page, second)
+
+          await page.keyboard.press("ArrowDown")
+          await wait(page, "")
+        },
+        {
+          model: openaiModel,
+        },
+      )
+    },
  })
 })

-test("shell history stays separate from normal prompt history", async ({ page, sdk, gotoSession }) => {
+test("shell history stays separate from normal prompt history", async ({ page, llm, backend, withBackendProject }) => {
  test.setTimeout(120_000)

-  await withSession(sdk, `e2e shell history ${Date.now()}`, async (session) => {
-    await gotoSession(session.id)
+  await withMockOpenAI({
+    serverUrl: backend.url,
+    llmUrl: llm.url,
+    fn: async () => {
+      const firstToken = `E2E_SHELL_ONE_${Date.now()}`
+      const secondToken = `E2E_SHELL_TWO_${Date.now()}`
+      const normalToken = `E2E_NORMAL_${Date.now()}`
+      const first = `echo ${firstToken}`
+      const second = `echo ${secondToken}`
+      const normal = `Reply with exactly: ${normalToken}`

-    const prompt = page.locator(promptSelector)
-    const firstToken = `E2E_SHELL_ONE_${Date.now()}`
-    const secondToken = `E2E_SHELL_TWO_${Date.now()}`
-    const normalToken = `E2E_NORMAL_${Date.now()}`
-    const first = `echo ${firstToken}`
-    const second = `echo ${secondToken}`
-    const normal = `Reply with exactly: ${normalToken}`
+      await llm.textMatch(titleMatch, "E2E Title")
+      await llm.textMatch(promptMatch(normalToken), normalToken)

-    await prompt.click()
-    await page.keyboard.type("!")
-    await page.keyboard.type(first)
-    await page.keyboard.press("Enter")
-    await wait(page, "")
-    await shell(sdk, session.id, first, firstToken)
+      await withBackendProject(
+        async (project) => {
+          const prompt = page.locator(promptSelector)

-    await prompt.click()
-    await page.keyboard.type("!")
-    await page.keyboard.type(second)
-    await page.keyboard.press("Enter")
-    await wait(page, "")
-    await shell(sdk, session.id, second, secondToken)
+          await prompt.click()
+          await page.keyboard.type("!")
+          await page.keyboard.type(first)
+          await page.keyboard.press("Enter")
+          await wait(page, "")

-    await prompt.click()
-    await page.keyboard.type("!")
-    await page.keyboard.press("ArrowUp")
-    await wait(page, second)
+          await expect(page).toHaveURL(/\/session\/[^/?#]+/, { timeout: 30_000 })
+          const sessionID = sessionIDFromUrl(page.url())!
+          project.trackSession(sessionID)
+          await shell(project.sdk, sessionID, first, firstToken)

-    await page.keyboard.press("ArrowUp")
-    await wait(page, first)
+          await prompt.click()
+          await page.keyboard.type("!")
+          await page.keyboard.type(second)
+          await page.keyboard.press("Enter")
+          await wait(page, "")
+          await shell(project.sdk, sessionID, second, secondToken)

-    await page.keyboard.press("ArrowDown")
-    await wait(page, second)
+          await prompt.click()
+          await page.keyboard.type("!")
+          await page.keyboard.press("ArrowUp")
+          await wait(page, second)

-    await page.keyboard.press("ArrowDown")
-    await wait(page, "")
+          await page.keyboard.press("ArrowUp")
+          await wait(page, first)

-    await page.keyboard.press("Escape")
-    await wait(page, "")
+          await page.keyboard.press("ArrowDown")
+          await wait(page, second)

-    await prompt.click()
-    await page.keyboard.type(normal)
-    await page.keyboard.press("Enter")
-    await wait(page, "")
-    await reply(sdk, session.id, normalToken)
+          await page.keyboard.press("ArrowDown")
+          await wait(page, "")

-    await prompt.click()
-    await page.keyboard.press("ArrowUp")
-    await wait(page, normal)
+          await page.keyboard.press("Escape")
+          await wait(page, "")
+
+          await prompt.click()
+          await page.keyboard.type(normal)
+          await page.keyboard.press("Enter")
+          await wait(page, "")
+          await reply(project.sdk, sessionID, normalToken)
+
+          await prompt.click()
+          await page.keyboard.press("ArrowUp")
+          await wait(page, normal)
+        },
+        {
+          model: openaiModel,
+        },
+      )
+    },
  })
 })
@@ -2,7 +2,6 @@ import type { ToolPart } from "@opencode-ai/sdk/v2/client"
 import { test, expect } from "../fixtures"
 import { sessionIDFromUrl } from "../actions"
 import { promptSelector } from "../selectors"
-import { createSdk } from "../utils"

 const isBash = (part: unknown): part is ToolPart => {
  if (!part || typeof part !== "object") return false
@@ -11,13 +10,12 @@ const isBash = (part: unknown): part is ToolPart => {
  return "state" in part
 }

-test("shell mode runs a command in the project directory", async ({ page, withProject }) => {
+test("shell mode runs a command in the project directory", async ({ page, withBackendProject }) => {
  test.setTimeout(120_000)

-  await withProject(async ({ directory, gotoSession, trackSession }) => {
-    const sdk = createSdk(directory)
+  await withBackendProject(async ({ directory, gotoSession, trackSession, sdk }) => {
    const prompt = page.locator(promptSelector)
-    const cmd = process.platform === "win32" ? "dir" : "ls"
+    const cmd = process.platform === "win32" ? "dir" : "command ls"

    await gotoSession()
    await prompt.click()
@@ -22,43 +22,45 @@ async function seed(sdk: Parameters<typeof withSession>[0], sessionID: string) {
    .toBeGreaterThan(0)
 }

-test("/share and /unshare update session share state", async ({ page, sdk, gotoSession }) => {
+test("/share and /unshare update session share state", async ({ page, withBackendProject }) => {
  test.skip(shareDisabled, "Share is disabled in this environment (OPENCODE_DISABLE_SHARE).")

-  await withSession(sdk, `e2e slash share ${Date.now()}`, async (session) => {
-    const prompt = page.locator(promptSelector)
+  await withBackendProject(async (project) => {
+    await withSession(project.sdk, `e2e slash share ${Date.now()}`, async (session) => {
+      const prompt = page.locator(promptSelector)

-    await seed(sdk, session.id)
-    await gotoSession(session.id)
+      await seed(project.sdk, session.id)
+      await project.gotoSession(session.id)

-    await prompt.click()
-    await page.keyboard.type("/share")
-    await expect(page.locator('[data-slash-id="session.share"]').first()).toBeVisible()
-    await page.keyboard.press("Enter")
+      await prompt.click()
+      await page.keyboard.type("/share")
+      await expect(page.locator('[data-slash-id="session.share"]').first()).toBeVisible()
+      await page.keyboard.press("Enter")

-    await expect
-      .poll(
-        async () => {
-          const data = await sdk.session.get({ sessionID: session.id }).then((r) => r.data)
-          return data?.share?.url || undefined
-        },
-        { timeout: 30_000 },
-      )
-      .not.toBeUndefined()
+      await expect
+        .poll(
+          async () => {
+            const data = await project.sdk.session.get({ sessionID: session.id }).then((r) => r.data)
+            return data?.share?.url || undefined
+          },
+          { timeout: 30_000 },
+        )
+        .not.toBeUndefined()

-    await prompt.click()
-    await page.keyboard.type("/unshare")
-    await expect(page.locator('[data-slash-id="session.unshare"]').first()).toBeVisible()
-    await page.keyboard.press("Enter")
+      await prompt.click()
+      await page.keyboard.type("/unshare")
+      await expect(page.locator('[data-slash-id="session.unshare"]').first()).toBeVisible()
+      await page.keyboard.press("Enter")

-    await expect
-      .poll(
-        async () => {
-          const data = await sdk.session.get({ sessionID: session.id }).then((r) => r.data)
-          return data?.share?.url || undefined
-        },
-        { timeout: 30_000 },
-      )
-      .toBeUndefined()
+      await expect
+        .poll(
+          async () => {
+            const data = await project.sdk.session.get({ sessionID: session.id }).then((r) => r.data)
+            return data?.share?.url || undefined
+          },
+          { timeout: 30_000 },
+        )
+        .toBeUndefined()
+    })
  })
 })
@@ -1,44 +1,9 @@
-import fs from "node:fs/promises"
-import path from "node:path"
 import { test, expect } from "../fixtures"
 import { promptSelector } from "../selectors"
-import { sessionIDFromUrl } from "../actions"
-import { createSdk } from "../utils"
+import { assistantText, sessionIDFromUrl } from "../actions"
+import { openaiModel, promptMatch, titleMatch, withMockOpenAI } from "./mock"

-async function config(dir: string, url: string) {
-  await fs.writeFile(
-    path.join(dir, "opencode.json"),
-    JSON.stringify({
-      $schema: "https://opencode.ai/config.json",
-      enabled_providers: ["e2e-llm"],
-      provider: {
-        "e2e-llm": {
-          name: "E2E LLM",
-          npm: "@ai-sdk/openai-compatible",
-          env: [],
-          models: {
-            "test-model": {
-              name: "Test Model",
-              tool_call: true,
-              limit: { context: 128000, output: 32000 },
-            },
-          },
-          options: {
-            apiKey: "test-key",
-            baseURL: url,
-          },
-        },
-      },
-      agent: {
-        build: {
-          model: "e2e-llm/test-model",
-        },
-      },
-    }),
-  )
-}
-
-test("can send a prompt and receive a reply", async ({ page, llm, withProject }) => {
+test("can send a prompt and receive a reply", async ({ page, llm, backend, withBackendProject }) => {
  test.setTimeout(120_000)

  const pageErrors: string[] = []
@@ -48,48 +13,41 @@ test("can send a prompt and receive a reply", async ({ page, llm, withProject })
  page.on("pageerror", onPageError)

  try {
-    await withProject(
-      async (project) => {
-        const sdk = createSdk(project.directory)
+    await withMockOpenAI({
+      serverUrl: backend.url,
+      llmUrl: llm.url,
+      fn: async () => {
        const token = `E2E_OK_${Date.now()}`

-        await llm.text(token)
-        await project.gotoSession()
+        await llm.textMatch(titleMatch, "E2E Title")
+        await llm.textMatch(promptMatch(token), token)

-        const prompt = page.locator(promptSelector)
-        await prompt.click()
-        await page.keyboard.type(`Reply with exactly: ${token}`)
-        await page.keyboard.press("Enter")
+        await withBackendProject(
+          async (project) => {
+            const prompt = page.locator(promptSelector)
+            await prompt.click()
+            await page.keyboard.type(`Reply with exactly: ${token}`)
+            await page.keyboard.press("Enter")

-        await expect(page).toHaveURL(/\/session\/[^/?#]+/, { timeout: 30_000 })
+            await expect(page).toHaveURL(/\/session\/[^/?#]+/, { timeout: 30_000 })

-        const sessionID = (() => {
-          const id = sessionIDFromUrl(page.url())
-          if (!id) throw new Error(`Failed to parse session id from url: ${page.url()}`)
-          return id
-        })()
-        project.trackSession(sessionID)
+            const sessionID = (() => {
+              const id = sessionIDFromUrl(page.url())
+              if (!id) throw new Error(`Failed to parse session id from url: ${page.url()}`)
+              return id
+            })()
+            project.trackSession(sessionID)

-        await expect
-          .poll(
-            async () => {
-              const messages = await sdk.session.messages({ sessionID, limit: 50 }).then((r) => r.data ?? [])
-              return messages
-                .filter((m) => m.info.role === "assistant")
-                .flatMap((m) => m.parts)
-                .filter((p) => p.type === "text")
-                .map((p) => p.text)
-                .join("\n")
-            },
-            { timeout: 30_000 },
-          )
-          .toContain(token)
+            await expect.poll(() => llm.calls()).toBeGreaterThanOrEqual(1)
+
+            await expect.poll(() => assistantText(project.sdk, sessionID), { timeout: 30_000 }).toContain(token)
+          },
+          {
+            model: openaiModel,
+          },
+        )
      },
-      {
-        model: { providerID: "e2e-llm", modelID: "test-model" },
-        setup: (dir) => config(dir, llm.url),
-      },
-    )
+    })
  } finally {
    page.off("pageerror", onPageError)
  }
@@ -1,7 +1,7 @@
 import { seedSessionTask, withSession } from "../actions"
 import { test, expect } from "../fixtures"

-test("task tool child-session link does not trigger stale show errors", async ({ page, sdk, gotoSession }) => {
+test("task tool child-session link does not trigger stale show errors", async ({ page, withBackendProject }) => {
  test.setTimeout(120_000)

  const errs: string[] = []
@@ -10,28 +10,32 @@ test("task tool child-session link does not trigger stale show errors", async ({
  }
  page.on("pageerror", onError)

-  await withSession(sdk, `e2e child nav ${Date.now()}`, async (session) => {
-    const child = await seedSessionTask(sdk, {
-      sessionID: session.id,
-      description: "Open child session",
-      prompt: "Search the repository for AssistantParts and then reply with exactly CHILD_OK.",
+  await withBackendProject(async ({ gotoSession, trackSession, sdk }) => {
+    await withSession(sdk, `e2e child nav ${Date.now()}`, async (session) => {
+      trackSession(session.id)
+      const child = await seedSessionTask(sdk, {
+        sessionID: session.id,
+        description: "Open child session",
+        prompt: "Search the repository for AssistantParts and then reply with exactly CHILD_OK.",
+      })
+      trackSession(child.sessionID)
+
+      try {
+        await gotoSession(session.id)
+
+        const link = page
+          .locator("a.subagent-link")
+          .filter({ hasText: /open child session/i })
+          .first()
+        await expect(link).toBeVisible({ timeout: 30_000 })
+        await link.click()
+
+        await expect(page).toHaveURL(new RegExp(`/session/${child.sessionID}(?:[/?#]|$)`), { timeout: 30_000 })
+        await page.waitForTimeout(1000)
+        expect(errs).toEqual([])
+      } finally {
+        page.off("pageerror", onError)
+      }
    })
-
-    try {
-      await gotoSession(session.id)
-
-      const link = page
-        .locator("a.subagent-link")
-        .filter({ hasText: /open child session/i })
-        .first()
-      await expect(link).toBeVisible({ timeout: 30_000 })
-      await link.click()
-
-      await expect(page).toHaveURL(new RegExp(`/session/${child.sessionID}(?:[/?#]|$)`), { timeout: 30_000 })
-      await page.waitForTimeout(1000)
-      expect(errs).toEqual([])
-    } finally {
-      page.off("pageerror", onError)
-    }
  })
 })
@@ -256,235 +256,50 @@ async function withMockPermission<T>(
  }
 }

-test("default dock shows prompt input", async ({ page, sdk, gotoSession }) => {
-  await withDockSession(sdk, "e2e composer dock default", async (session) => {
-    await gotoSession(session.id)
+test("default dock shows prompt input", async ({ page, withBackendProject }) => {
+  await withBackendProject(async (project) => {
+    await withDockSession(project.sdk, "e2e composer dock default", async (session) => {
+      await project.gotoSession(session.id)

-    await expect(page.locator(sessionComposerDockSelector)).toBeVisible()
-    await expect(page.locator(promptSelector)).toBeVisible()
-    await expect(page.locator(questionDockSelector)).toHaveCount(0)
-    await expect(page.locator(permissionDockSelector)).toHaveCount(0)
+      await expect(page.locator(sessionComposerDockSelector)).toBeVisible()
+      await expect(page.locator(promptSelector)).toBeVisible()
+      await expect(page.locator(questionDockSelector)).toHaveCount(0)
+      await expect(page.locator(permissionDockSelector)).toHaveCount(0)

-    await page.locator(promptSelector).click()
-    await expect(page.locator(promptSelector)).toBeFocused()
-  })
-})
-
-test("auto-accept toggle works before first submit", async ({ page, gotoSession }) => {
-  await gotoSession()
-
-  const button = page.locator('[data-action="prompt-permissions"]').first()
-  await expect(button).toBeVisible()
-  await expect(button).toHaveAttribute("aria-pressed", "false")
-
-  await setAutoAccept(page, true)
-  await setAutoAccept(page, false)
-})
-
-test("blocked question flow unblocks after submit", async ({ page, sdk, gotoSession }) => {
-  await withDockSession(sdk, "e2e composer dock question", async (session) => {
-    await withDockSeed(sdk, session.id, async () => {
-      await gotoSession(session.id)
-
-      await seedSessionQuestion(sdk, {
-        sessionID: session.id,
-        questions: [
-          {
-            header: "Need input",
-            question: "Pick one option",
-            options: [
-              { label: "Continue", description: "Continue now" },
-              { label: "Stop", description: "Stop here" },
-            ],
-          },
-        ],
-      })
-
-      const dock = page.locator(questionDockSelector)
-      await expectQuestionBlocked(page)
-
-      await dock.locator('[data-slot="question-option"]').first().click()
-      await dock.getByRole("button", { name: /submit/i }).click()
-
-      await expectQuestionOpen(page)
+      await page.locator(promptSelector).click()
+      await expect(page.locator(promptSelector)).toBeFocused()
    })
  })
 })

-test("blocked question flow supports keyboard shortcuts", async ({ page, sdk, gotoSession }) => {
-  await withDockSession(sdk, "e2e composer dock question keyboard", async (session) => {
-    await withDockSeed(sdk, session.id, async () => {
-      await gotoSession(session.id)
+test("auto-accept toggle works before first submit", async ({ page, withBackendProject }) => {
+  await withBackendProject(async ({ gotoSession }) => {
+    await gotoSession()

-      await seedSessionQuestion(sdk, {
-        sessionID: session.id,
-        questions: [
-          {
-            header: "Need input",
-            question: "Pick one option",
-            options: [
-              { label: "Continue", description: "Continue now" },
-              { label: "Stop", description: "Stop here" },
-            ],
-          },
-        ],
-      })
+    const button = page.locator('[data-action="prompt-permissions"]').first()
+    await expect(button).toBeVisible()
+    await expect(button).toHaveAttribute("aria-pressed", "false")

-      const dock = page.locator(questionDockSelector)
-      const first = dock.locator('[data-slot="question-option"]').first()
-      const second = dock.locator('[data-slot="question-option"]').nth(1)
-
-      await expectQuestionBlocked(page)
-      await expect(first).toBeFocused()
-
-      await page.keyboard.press("ArrowDown")
-      await expect(second).toBeFocused()
-
-      await page.keyboard.press("Space")
-      await page.keyboard.press(`${modKey}+Enter`)
-      await expectQuestionOpen(page)
-    })
-  })
-})
-
-test("blocked question flow supports escape dismiss", async ({ page, sdk, gotoSession }) => {
-  await withDockSession(sdk, "e2e composer dock question escape", async (session) => {
-    await withDockSeed(sdk, session.id, async () => {
-      await gotoSession(session.id)
-
-      await seedSessionQuestion(sdk, {
-        sessionID: session.id,
-        questions: [
-          {
-            header: "Need input",
-            question: "Pick one option",
-            options: [
-              { label: "Continue", description: "Continue now" },
-              { label: "Stop", description: "Stop here" },
-            ],
-          },
-        ],
-      })
-
-      const dock = page.locator(questionDockSelector)
-      const first = dock.locator('[data-slot="question-option"]').first()
-
-      await expectQuestionBlocked(page)
-      await expect(first).toBeFocused()
-
-      await page.keyboard.press("Escape")
-      await expectQuestionOpen(page)
-    })
-  })
-})
-
-test("blocked permission flow supports allow once", async ({ page, sdk, gotoSession }) => {
-  await withDockSession(sdk, "e2e composer dock permission once", async (session) => {
-    await gotoSession(session.id)
+    await setAutoAccept(page, true)
    await setAutoAccept(page, false)
-    await withMockPermission(
-      page,
-      {
-        id: "per_e2e_once",
-        sessionID: session.id,
-        permission: "bash",
-        patterns: ["/tmp/opencode-e2e-perm-once"],
-        metadata: { description: "Need permission for command" },
-      },
-      undefined,
-      async (state) => {
-        await page.goto(page.url())
-        await expectPermissionBlocked(page)
-
-        await clearPermissionDock(page, /allow once/i)
-        await state.resolved()
-        await page.goto(page.url())
-        await expectPermissionOpen(page)
-      },
-    )
  })
 })

-test("blocked permission flow supports reject", async ({ page, sdk, gotoSession }) => {
-  await withDockSession(sdk, "e2e composer dock permission reject", async (session) => {
-    await gotoSession(session.id)
-    await setAutoAccept(page, false)
-    await withMockPermission(
-      page,
-      {
-        id: "per_e2e_reject",
-        sessionID: session.id,
-        permission: "bash",
-        patterns: ["/tmp/opencode-e2e-perm-reject"],
-      },
-      undefined,
-      async (state) => {
-        await page.goto(page.url())
-        await expectPermissionBlocked(page)
+test("blocked question flow unblocks after submit", async ({ page, withBackendProject }) => {
+  await withBackendProject(async (project) => {
+    await withDockSession(project.sdk, "e2e composer dock question", async (session) => {
+      await withDockSeed(project.sdk, session.id, async () => {
+        await project.gotoSession(session.id)

-        await clearPermissionDock(page, /deny/i)
-        await state.resolved()
-        await page.goto(page.url())
-        await expectPermissionOpen(page)
-      },
-    )
-  })
-})
-
-test("blocked permission flow supports allow always", async ({ page, sdk, gotoSession }) => {
-  await withDockSession(sdk, "e2e composer dock permission always", async (session) => {
-    await gotoSession(session.id)
-    await setAutoAccept(page, false)
-    await withMockPermission(
-      page,
-      {
-        id: "per_e2e_always",
-        sessionID: session.id,
-        permission: "bash",
-        patterns: ["/tmp/opencode-e2e-perm-always"],
-        metadata: { description: "Need permission for command" },
-      },
-      undefined,
-      async (state) => {
-        await page.goto(page.url())
-        await expectPermissionBlocked(page)
-
-        await clearPermissionDock(page, /allow always/i)
-        await state.resolved()
-        await page.goto(page.url())
-        await expectPermissionOpen(page)
-      },
-    )
-  })
-})
-
-test("child session question request blocks parent dock and unblocks after submit", async ({
-  page,
-  sdk,
-  gotoSession,
-}) => {
-  await withDockSession(sdk, "e2e composer dock child question parent", async (session) => {
-    await gotoSession(session.id)
-
-    const child = await sdk.session
-      .create({
-        title: "e2e composer dock child question",
-        parentID: session.id,
-      })
-      .then((r) => r.data)
-    if (!child?.id) throw new Error("Child session create did not return an id")
-
-    try {
-      await withDockSeed(sdk, child.id, async () => {
-        await seedSessionQuestion(sdk, {
-          sessionID: child.id,
+        await seedSessionQuestion(project.sdk, {
+          sessionID: session.id,
          questions: [
            {
-              header: "Child input",
-              question: "Pick one child option",
+              header: "Need input",
+              question: "Pick one option",
              options: [
-                { label: "Continue", description: "Continue child" },
-                { label: "Stop", description: "Stop child" },
+                { label: "Continue", description: "Continue now" },
+                { label: "Stop", description: "Stop here" },
              ],
            },
          ],
@@ -498,40 +313,96 @@ test("child session question request blocks parent dock and unblocks after submi

        await expectQuestionOpen(page)
      })
-    } finally {
-      await cleanupSession({ sdk, sessionID: child.id })
-    }
+    })
  })
 })

-test("child session permission request blocks parent dock and supports allow once", async ({
-  page,
-  sdk,
-  gotoSession,
-}) => {
-  await withDockSession(sdk, "e2e composer dock child permission parent", async (session) => {
-    await gotoSession(session.id)
-    await setAutoAccept(page, false)
+test("blocked question flow supports keyboard shortcuts", async ({ page, withBackendProject }) => {
+  await withBackendProject(async (project) => {
+    await withDockSession(project.sdk, "e2e composer dock question keyboard", async (session) => {
+      await withDockSeed(project.sdk, session.id, async () => {
+        await project.gotoSession(session.id)

-    const child = await sdk.session
-      .create({
-        title: "e2e composer dock child permission",
-        parentID: session.id,
+        await seedSessionQuestion(project.sdk, {
+          sessionID: session.id,
+          questions: [
+            {
+              header: "Need input",
+              question: "Pick one option",
+              options: [
+                { label: "Continue", description: "Continue now" },
+                { label: "Stop", description: "Stop here" },
+              ],
+            },
+          ],
+        })
+
+        const dock = page.locator(questionDockSelector)
+        const first = dock.locator('[data-slot="question-option"]').first()
+        const second = dock.locator('[data-slot="question-option"]').nth(1)
+
+        await expectQuestionBlocked(page)
+        await expect(first).toBeFocused()
+
+        await page.keyboard.press("ArrowDown")
+        await expect(second).toBeFocused()
+
+        await page.keyboard.press("Space")
+        await page.keyboard.press(`${modKey}+Enter`)
+        await expectQuestionOpen(page)
      })
-      .then((r) => r.data)
-    if (!child?.id) throw new Error("Child session create did not return an id")
+    })
+  })
+})

-    try {
+test("blocked question flow supports escape dismiss", async ({ page, withBackendProject }) => {
+  await withBackendProject(async (project) => {
+    await withDockSession(project.sdk, "e2e composer dock question escape", async (session) => {
+      await withDockSeed(project.sdk, session.id, async () => {
+        await project.gotoSession(session.id)
+
+        await seedSessionQuestion(project.sdk, {
+          sessionID: session.id,
+          questions: [
+            {
+              header: "Need input",
+              question: "Pick one option",
+              options: [
+                { label: "Continue", description: "Continue now" },
+                { label: "Stop", description: "Stop here" },
+              ],
+            },
+          ],
+        })
+
+        const dock = page.locator(questionDockSelector)
+        const first = dock.locator('[data-slot="question-option"]').first()
+
+        await expectQuestionBlocked(page)
+        await expect(first).toBeFocused()
+
+        await page.keyboard.press("Escape")
+        await expectQuestionOpen(page)
+      })
+    })
+  })
+})
+
+test("blocked permission flow supports allow once", async ({ page, withBackendProject }) => {
+  await withBackendProject(async (project) => {
+    await withDockSession(project.sdk, "e2e composer dock permission once", async (session) => {
+      await project.gotoSession(session.id)
+      await setAutoAccept(page, false)
      await withMockPermission(
        page,
        {
-          id: "per_e2e_child",
-          sessionID: child.id,
+          id: "per_e2e_once",
+          sessionID: session.id,
          permission: "bash",
-          patterns: ["/tmp/opencode-e2e-perm-child"],
-          metadata: { description: "Need child permission" },
+          patterns: ["/tmp/opencode-e2e-perm-once"],
+          metadata: { description: "Need permission for command" },
        },
-        { child },
+        undefined,
        async (state) => {
          await page.goto(page.url())
          await expectPermissionBlocked(page)
@@ -539,67 +410,218 @@ test("child session permission request blocks parent dock and supports allow onc
          await clearPermissionDock(page, /allow once/i)
          await state.resolved()
          await page.goto(page.url())
-
          await expectPermissionOpen(page)
        },
      )
-    } finally {
-      await cleanupSession({ sdk, sessionID: child.id })
-    }
-  })
-})
-
-test("todo dock transitions and collapse behavior", async ({ page, sdk, gotoSession }) => {
-  await withDockSession(sdk, "e2e composer dock todo", async (session) => {
-    const dock = await todoDock(page, session.id)
-    await gotoSession(session.id)
-    await expect(page.locator(sessionComposerDockSelector)).toBeVisible()
-
-    try {
-      await dock.open([
-        { content: "first task", status: "pending", priority: "high" },
-        { content: "second task", status: "in_progress", priority: "medium" },
-      ])
-      await dock.expectOpen(["pending", "in_progress"])
-
-      await dock.collapse()
-      await dock.expectCollapsed(["pending", "in_progress"])
-
-      await dock.expand()
-      await dock.expectOpen(["pending", "in_progress"])
-
-      await dock.finish([
-        { content: "first task", status: "completed", priority: "high" },
-        { content: "second task", status: "cancelled", priority: "medium" },
-      ])
-      await dock.expectClosed()
-    } finally {
-      await dock.clear()
-    }
-  })
-})
-
-test("keyboard focus stays off prompt while blocked", async ({ page, sdk, gotoSession }) => {
-  await withDockSession(sdk, "e2e composer dock keyboard", async (session) => {
-    await withDockSeed(sdk, session.id, async () => {
-      await gotoSession(session.id)
-
-      await seedSessionQuestion(sdk, {
-        sessionID: session.id,
-        questions: [
-          {
-            header: "Need input",
-            question: "Pick one option",
-            options: [{ label: "Continue", description: "Continue now" }],
-          },
-        ],
-      })
-
-      await expectQuestionBlocked(page)
-
-      await page.locator("main").click({ position: { x: 5, y: 5 } })
-      await page.keyboard.type("abc")
-      await expect(page.locator(promptSelector)).toHaveCount(0)
+    })
+  })
+})
+
+test("blocked permission flow supports reject", async ({ page, withBackendProject }) => {
+  await withBackendProject(async (project) => {
+    await withDockSession(project.sdk, "e2e composer dock permission reject", async (session) => {
+      await project.gotoSession(session.id)
+      await setAutoAccept(page, false)
+      await withMockPermission(
+        page,
+        {
+          id: "per_e2e_reject",
+          sessionID: session.id,
+          permission: "bash",
+          patterns: ["/tmp/opencode-e2e-perm-reject"],
+        },
+        undefined,
+        async (state) => {
+          await page.goto(page.url())
+          await expectPermissionBlocked(page)
+
+          await clearPermissionDock(page, /deny/i)
+          await state.resolved()
+          await page.goto(page.url())
+          await expectPermissionOpen(page)
+        },
+      )
+    })
+  })
+})
+
+test("blocked permission flow supports allow always", async ({ page, withBackendProject }) => {
+  await withBackendProject(async (project) => {
+    await withDockSession(project.sdk, "e2e composer dock permission always", async (session) => {
+      await project.gotoSession(session.id)
+      await setAutoAccept(page, false)
+      await withMockPermission(
+        page,
+        {
+          id: "per_e2e_always",
+          sessionID: session.id,
+          permission: "bash",
+          patterns: ["/tmp/opencode-e2e-perm-always"],
+          metadata: { description: "Need permission for command" },
+        },
+        undefined,
+        async (state) => {
+          await page.goto(page.url())
+          await expectPermissionBlocked(page)
+
+          await clearPermissionDock(page, /allow always/i)
+          await state.resolved()
+          await page.goto(page.url())
+          await expectPermissionOpen(page)
+        },
+      )
+    })
+  })
+})
+
+test("child session question request blocks parent dock and unblocks after submit", async ({
+  page,
+  withBackendProject,
+}) => {
+  await withBackendProject(async (project) => {
+    await withDockSession(project.sdk, "e2e composer dock child question parent", async (session) => {
+      await project.gotoSession(session.id)
+
+      const child = await project.sdk.session
+        .create({
+          title: "e2e composer dock child question",
+          parentID: session.id,
+        })
+        .then((r) => r.data)
+      if (!child?.id) throw new Error("Child session create did not return an id")
+
+      try {
+        await withDockSeed(project.sdk, child.id, async () => {
+          await seedSessionQuestion(project.sdk, {
+            sessionID: child.id,
+            questions: [
+              {
+                header: "Child input",
+                question: "Pick one child option",
+                options: [
+                  { label: "Continue", description: "Continue child" },
+                  { label: "Stop", description: "Stop child" },
+                ],
+              },
+            ],
+          })
+
+          const dock = page.locator(questionDockSelector)
+          await expectQuestionBlocked(page)
+
+          await dock.locator('[data-slot="question-option"]').first().click()
+          await dock.getByRole("button", { name: /submit/i }).click()
+
+          await expectQuestionOpen(page)
+        })
+      } finally {
+        await cleanupSession({ sdk: project.sdk, sessionID: child.id })
+      }
+    })
+  })
+})
+
+test("child session permission request blocks parent dock and supports allow once", async ({
+  page,
+  withBackendProject,
+}) => {
+  await withBackendProject(async (project) => {
+    await withDockSession(project.sdk, "e2e composer dock child permission parent", async (session) => {
+      await project.gotoSession(session.id)
+      await setAutoAccept(page, false)
+
+      const child = await project.sdk.session
+        .create({
+          title: "e2e composer dock child permission",
+          parentID: session.id,
+        })
+        .then((r) => r.data)
+      if (!child?.id) throw new Error("Child session create did not return an id")
+
+      try {
+        await withMockPermission(
+          page,
+          {
+            id: "per_e2e_child",
+            sessionID: child.id,
+            permission: "bash",
+            patterns: ["/tmp/opencode-e2e-perm-child"],
+            metadata: { description: "Need child permission" },
+          },
+          { child },
+          async (state) => {
+            await page.goto(page.url())
+            await expectPermissionBlocked(page)
+
+            await clearPermissionDock(page, /allow once/i)
+            await state.resolved()
+            await page.goto(page.url())
+
+            await expectPermissionOpen(page)
+          },
+        )
+      } finally {
+        await cleanupSession({ sdk: project.sdk, sessionID: child.id })
+      }
+    })
+  })
+})
+
+test("todo dock transitions and collapse behavior", async ({ page, withBackendProject }) => {
+  await withBackendProject(async (project) => {
+    await withDockSession(project.sdk, "e2e composer dock todo", async (session) => {
+      const dock = await todoDock(page, session.id)
+      await project.gotoSession(session.id)
+      await expect(page.locator(sessionComposerDockSelector)).toBeVisible()
+
+      try {
+        await dock.open([
+          { content: "first task", status: "pending", priority: "high" },
+          { content: "second task", status: "in_progress", priority: "medium" },
+        ])
+        await dock.expectOpen(["pending", "in_progress"])
+
+        await dock.collapse()
+        await dock.expectCollapsed(["pending", "in_progress"])
+
+        await dock.expand()
+        await dock.expectOpen(["pending", "in_progress"])
+
+        await dock.finish([
+          { content: "first task", status: "completed", priority: "high" },
+          { content: "second task", status: "cancelled", priority: "medium" },
+        ])
+        await dock.expectClosed()
+      } finally {
+        await dock.clear()
+      }
+    })
+  })
+})
+
+test("keyboard focus stays off prompt while blocked", async ({ page, withBackendProject }) => {
+  await withBackendProject(async (project) => {
+    await withDockSession(project.sdk, "e2e composer dock keyboard", async (session) => {
+      await withDockSeed(project.sdk, session.id, async () => {
+        await project.gotoSession(session.id)
+
+        await seedSessionQuestion(project.sdk, {
+          sessionID: session.id,
+          questions: [
+            {
+              header: "Need input",
+              question: "Pick one option",
+              options: [{ label: "Continue", description: "Continue now" }],
+            },
+          ],
+        })
+
+        await expectQuestionBlocked(page)
+
+        await page.locator("main").click({ position: { x: 5, y: 5 } })
+        await page.keyboard.type("abc")
+        await expect(page.locator(promptSelector)).toHaveCount(0)
+      })
    })
  })
 })
@@ -49,13 +49,13 @@ async function seedConversation(input: {
  return { prompt, userMessageID }
 }

-test("slash undo sets revert and restores prior prompt", async ({ page, withProject }) => {
+test("slash undo sets revert and restores prior prompt", async ({ page, withBackendProject }) => {
  test.setTimeout(120_000)

  const token = `undo_${Date.now()}`

-  await withProject(async (project) => {
-    const sdk = createSdk(project.directory)
+  await withBackendProject(async (project) => {
+    const sdk = project.sdk

    await withSession(sdk, `e2e undo ${Date.now()}`, async (session) => {
      await project.gotoSession(session.id)
@@ -81,13 +81,13 @@ test("slash undo sets revert and restores prior prompt", async ({ page, withProj
  })
 })

-test("slash redo clears revert and restores latest state", async ({ page, withProject }) => {
+test("slash redo clears revert and restores latest state", async ({ page, withBackendProject }) => {
  test.setTimeout(120_000)

  const token = `redo_${Date.now()}`

-  await withProject(async (project) => {
-    const sdk = createSdk(project.directory)
+  await withBackendProject(async (project) => {
+    const sdk = project.sdk

    await withSession(sdk, `e2e redo ${Date.now()}`, async (session) => {
      await project.gotoSession(session.id)
@@ -128,14 +128,14 @@ test("slash redo clears revert and restores latest state", async ({ page, withPr
  })
 })

-test("slash undo/redo traverses multi-step revert stack", async ({ page, withProject }) => {
+test("slash undo/redo traverses multi-step revert stack", async ({ page, withBackendProject }) => {
  test.setTimeout(120_000)

  const firstToken = `undo_redo_first_${Date.now()}`
  const secondToken = `undo_redo_second_${Date.now()}`

-  await withProject(async (project) => {
-    const sdk = createSdk(project.directory)
+  await withBackendProject(async (project) => {
+    const sdk = project.sdk

    await withSession(sdk, `e2e undo redo stack ${Date.now()}`, async (session) => {
      await project.gotoSession(session.id)
@@ -31,144 +31,152 @@ async function seedMessage(sdk: Sdk, sessionID: string) {
    .toBeGreaterThan(0)
 }

-test("session can be renamed via header menu", async ({ page, sdk, gotoSession }) => {
+test("session can be renamed via header menu", async ({ page, withBackendProject }) => {
  const stamp = Date.now()
  const originalTitle = `e2e rename test ${stamp}`
  const renamedTitle = `e2e renamed ${stamp}`

-  await withSession(sdk, originalTitle, async (session) => {
-    await seedMessage(sdk, session.id)
-    await gotoSession(session.id)
-    await expect(page.getByRole("heading", { level: 1 }).first()).toHaveText(originalTitle)
+  await withBackendProject(async (project) => {
+    await withSession(project.sdk, originalTitle, async (session) => {
+      await seedMessage(project.sdk, session.id)
+      await project.gotoSession(session.id)
+      await expect(page.getByRole("heading", { level: 1 }).first()).toHaveText(originalTitle)

-    const menu = await openSessionMoreMenu(page, session.id)
-    await clickMenuItem(menu, /rename/i)
+      const menu = await openSessionMoreMenu(page, session.id)
+      await clickMenuItem(menu, /rename/i)

-    const input = page.locator(".scroll-view__viewport").locator(inlineInputSelector).first()
-    await expect(input).toBeVisible()
-    await expect(input).toBeFocused()
-    await input.fill(renamedTitle)
-    await expect(input).toHaveValue(renamedTitle)
-    await input.press("Enter")
+      const input = page.locator(".scroll-view__viewport").locator(inlineInputSelector).first()
+      await expect(input).toBeVisible()
+      await expect(input).toBeFocused()
+      await input.fill(renamedTitle)
+      await expect(input).toHaveValue(renamedTitle)
+      await input.press("Enter")

-    await expect
-      .poll(
-        async () => {
-          const data = await sdk.session.get({ sessionID: session.id }).then((r) => r.data)
-          return data?.title
-        },
-        { timeout: 30_000 },
-      )
-      .toBe(renamedTitle)
+      await expect
+        .poll(
+          async () => {
+            const data = await project.sdk.session.get({ sessionID: session.id }).then((r) => r.data)
+            return data?.title
+          },
+          { timeout: 30_000 },
+        )
+        .toBe(renamedTitle)

-    await expect(page.getByRole("heading", { level: 1 }).first()).toHaveText(renamedTitle)
+      await expect(page.getByRole("heading", { level: 1 }).first()).toHaveText(renamedTitle)
+    })
  })
 })

-test("session can be archived via header menu", async ({ page, sdk, gotoSession }) => {
+test("session can be archived via header menu", async ({ page, withBackendProject }) => {
  const stamp = Date.now()
  const title = `e2e archive test ${stamp}`

-  await withSession(sdk, title, async (session) => {
-    await seedMessage(sdk, session.id)
-    await gotoSession(session.id)
-    const menu = await openSessionMoreMenu(page, session.id)
-    await clickMenuItem(menu, /archive/i)
+  await withBackendProject(async (project) => {
+    await withSession(project.sdk, title, async (session) => {
+      await seedMessage(project.sdk, session.id)
+      await project.gotoSession(session.id)
+      const menu = await openSessionMoreMenu(page, session.id)
+      await clickMenuItem(menu, /archive/i)

-    await expect
-      .poll(
-        async () => {
-          const data = await sdk.session.get({ sessionID: session.id }).then((r) => r.data)
-          return data?.time?.archived
-        },
-        { timeout: 30_000 },
-      )
-      .not.toBeUndefined()
+      await expect
+        .poll(
+          async () => {
+            const data = await project.sdk.session.get({ sessionID: session.id }).then((r) => r.data)
+            return data?.time?.archived
+          },
+          { timeout: 30_000 },
+        )
+        .not.toBeUndefined()

-    await openSidebar(page)
-    await expect(page.locator(sessionItemSelector(session.id))).toHaveCount(0)
+      await openSidebar(page)
+      await expect(page.locator(sessionItemSelector(session.id))).toHaveCount(0)
+    })
  })
 })

-test("session can be deleted via header menu", async ({ page, sdk, gotoSession }) => {
+test("session can be deleted via header menu", async ({ page, withBackendProject }) => {
  const stamp = Date.now()
  const title = `e2e delete test ${stamp}`

-  await withSession(sdk, title, async (session) => {
-    await seedMessage(sdk, session.id)
-    await gotoSession(session.id)
-    const menu = await openSessionMoreMenu(page, session.id)
-    await clickMenuItem(menu, /delete/i)
-    await confirmDialog(page, /delete/i)
+  await withBackendProject(async (project) => {
+    await withSession(project.sdk, title, async (session) => {
+      await seedMessage(project.sdk, session.id)
+      await project.gotoSession(session.id)
+      const menu = await openSessionMoreMenu(page, session.id)
+      await clickMenuItem(menu, /delete/i)
+      await confirmDialog(page, /delete/i)

-    await expect
-      .poll(
-        async () => {
-          const data = await sdk.session
-            .get({ sessionID: session.id })
-            .then((r) => r.data)
-            .catch(() => undefined)
-          return data?.id
-        },
-        { timeout: 30_000 },
-      )
-      .toBeUndefined()
+      await expect
+        .poll(
+          async () => {
+            const data = await project.sdk.session
+              .get({ sessionID: session.id })
+              .then((r) => r.data)
+              .catch(() => undefined)
+            return data?.id
+          },
+          { timeout: 30_000 },
+        )
+        .toBeUndefined()

-    await openSidebar(page)
-    await expect(page.locator(sessionItemSelector(session.id))).toHaveCount(0)
+      await openSidebar(page)
+      await expect(page.locator(sessionItemSelector(session.id))).toHaveCount(0)
+    })
  })
 })

-test("session can be shared and unshared via header button", async ({ page, sdk, gotoSession }) => {
+test("session can be shared and unshared via header button", async ({ page, withBackendProject }) => {
  test.skip(shareDisabled, "Share is disabled in this environment (OPENCODE_DISABLE_SHARE).")

  const stamp = Date.now()
  const title = `e2e share test ${stamp}`

-  await withSession(sdk, title, async (session) => {
-    await seedMessage(sdk, session.id)
-    await gotoSession(session.id)
+  await withBackendProject(async (project) => {
+    await withSession(project.sdk, title, async (session) => {
+      await seedMessage(project.sdk, session.id)
+      await project.gotoSession(session.id)

-    const shared = await openSharePopover(page)
-    const publish = shared.popoverBody.getByRole("button", { name: "Publish" }).first()
-    await expect(publish).toBeVisible({ timeout: 30_000 })
-    await publish.click()
+      const shared = await openSharePopover(page)
+      const publish = shared.popoverBody.getByRole("button", { name: "Publish" }).first()
+      await expect(publish).toBeVisible({ timeout: 30_000 })
+      await publish.click()

-    await expect(shared.popoverBody.getByRole("button", { name: "Unpublish" }).first()).toBeVisible({
-      timeout: 30_000,
-    })
+      await expect(shared.popoverBody.getByRole("button", { name: "Unpublish" }).first()).toBeVisible({
+        timeout: 30_000,
+      })

-    await expect
-      .poll(
-        async () => {
-          const data = await sdk.session.get({ sessionID: session.id }).then((r) => r.data)
-          return data?.share?.url || undefined
-        },
-        { timeout: 30_000 },
-      )
-      .not.toBeUndefined()
+      await expect
+        .poll(
+          async () => {
+            const data = await project.sdk.session.get({ sessionID: session.id }).then((r) => r.data)
+            return data?.share?.url || undefined
+          },
+          { timeout: 30_000 },
+        )
+        .not.toBeUndefined()

-    const unpublish = shared.popoverBody.getByRole("button", { name: "Unpublish" }).first()
-    await expect(unpublish).toBeVisible({ timeout: 30_000 })
-    await unpublish.click()
+      const unpublish = shared.popoverBody.getByRole("button", { name: "Unpublish" }).first()
+      await expect(unpublish).toBeVisible({ timeout: 30_000 })
+      await unpublish.click()

-    await expect(shared.popoverBody.getByRole("button", { name: "Publish" }).first()).toBeVisible({
-      timeout: 30_000,
-    })
+      await expect(shared.popoverBody.getByRole("button", { name: "Publish" }).first()).toBeVisible({
+        timeout: 30_000,
+      })

-    await expect
-      .poll(
-        async () => {
-          const data = await sdk.session.get({ sessionID: session.id }).then((r) => r.data)
-          return data?.share?.url || undefined
-        },
-        { timeout: 30_000 },
-      )
-      .toBeUndefined()
+      await expect
+        .poll(
+          async () => {
+            const data = await project.sdk.session.get({ sessionID: session.id }).then((r) => r.data)
+            return data?.share?.url || undefined
+          },
+          { timeout: 30_000 },
+        )
+        .toBeUndefined()

-    const unshared = await openSharePopover(page)
-    await expect(unshared.popoverBody.getByRole("button", { name: "Publish" }).first()).toBeVisible({
-      timeout: 30_000,
+      const unshared = await openSharePopover(page)
+      await expect(unshared.popoverBody.getByRole("button", { name: "Publish" }).first()).toBeVisible({
+        timeout: 30_000,
+      })
    })
  })
 })
@@ -26,21 +26,21 @@ export const serverNamePattern = new RegExp(`(?:${serverNames.map(escape).join("
 export const modKey = process.platform === "darwin" ? "Meta" : "Control"
 export const terminalToggleKey = "Control+Backquote"

-export function createSdk(directory?: string) {
-  return createOpencodeClient({ baseUrl: serverUrl, directory, throwOnError: true })
+export function createSdk(directory?: string, baseUrl = serverUrl) {
+  return createOpencodeClient({ baseUrl, directory, throwOnError: true })
 }

-export async function resolveDirectory(directory: string) {
-  return createSdk(directory)
+export async function resolveDirectory(directory: string, baseUrl = serverUrl) {
+  return createSdk(directory, baseUrl)
    .path.get()
    .then((x) => x.data?.directory ?? directory)
 }

-export async function getWorktree() {
-  const sdk = createSdk()
+export async function getWorktree(baseUrl = serverUrl) {
+  const sdk = createSdk(undefined, baseUrl)
  const result = await sdk.path.get()
  const data = result.data
-  if (!data?.worktree) throw new Error(`Failed to resolve a worktree from ${serverUrl}/path`)
+  if (!data?.worktree) throw new Error(`Failed to resolve a worktree from ${baseUrl}/path`)
  return data.worktree
 }

@@ -1,4 +1,4 @@
-import { Clock, Duration, Effect, Layer, Option, Schema, SchemaGetter, ServiceMap } from "effect"
+import { Cache, Clock, Duration, Effect, Layer, Option, Schema, SchemaGetter, ServiceMap } from "effect"
 import { FetchHttpClient, HttpClient, HttpClientRequest, HttpClientResponse } from "effect/unstable/http"

 import { makeRuntime } from "@/effect/run-service"
@@ -175,9 +175,8 @@ export namespace Account {
          mapAccountServiceError("HTTP request failed"),
        )

-      const resolveToken = Effect.fnUntraced(function* (row: AccountRow) {
+      const refreshToken = Effect.fnUntraced(function* (row: AccountRow) {
        const now = yield* Clock.currentTimeMillis
-        if (row.token_expiry && row.token_expiry > now) return row.access_token

        const response = yield* executeEffectOk(
          HttpClientRequest.post(`${row.url}/auth/device/token`).pipe(
@@ -208,6 +207,30 @@ export namespace Account {
        return parsed.access_token
      })

+      const refreshTokenCache = yield* Cache.make<AccountID, AccessToken, AccountError>({
+        capacity: Number.POSITIVE_INFINITY,
+        timeToLive: Duration.zero,
+        lookup: Effect.fnUntraced(function* (accountID) {
+          const maybeAccount = yield* repo.getRow(accountID)
+          if (Option.isNone(maybeAccount)) {
+            return yield* Effect.fail(new AccountServiceError({ message: "Account not found during token refresh" }))
+          }
+
+          const account = maybeAccount.value
+          const now = yield* Clock.currentTimeMillis
+          if (account.token_expiry && account.token_expiry > now) return account.access_token
+
+          return yield* refreshToken(account)
+        }),
+      })
+
+      const resolveToken = Effect.fnUntraced(function* (row: AccountRow) {
+        const now = yield* Clock.currentTimeMillis
+        if (row.token_expiry && row.token_expiry > now) return row.access_token
+
+        return yield* Cache.get(refreshTokenCache, row.id)
+      })
+
      const resolveAccess = Effect.fnUntraced(function* (accountID: AccountID) {
        const maybeAccount = yield* repo.getRow(accountID)
        if (Option.isNone(maybeAccount)) return Option.none()
@@ -85,7 +85,7 @@ export namespace Command {

        commands[Default.INIT] = {
          name: Default.INIT,
-          description: "create/update AGENTS.md",
+          description: "guided AGENTS.md setup",
          source: "command",
          get template() {
            return PROMPT_INITIALIZE.replace("${path}", ctx.worktree)
@@ -1,10 +1,66 @@
-Please analyze this codebase and create an AGENTS.md file containing:
-1. Build/lint/test commands - especially for running a single test
-2. Code style guidelines including imports, formatting, types, naming conventions, error handling, etc.
+Create or update `AGENTS.md` for this repository.

-The file you create will be given to agentic coding agents (such as yourself) that operate in this repository. Make it about 150 lines long.
-If there are Cursor rules (in .cursor/rules/ or .cursorrules) or Copilot rules (in .github/copilot-instructions.md), make sure to include them.
-
-If there's already an AGENTS.md, improve it if it's located in ${path}
+The goal is a compact instruction file that helps future OpenCode sessions avoid mistakes and ramp up quickly. Every line should answer: "Would an agent likely miss this without help?" If not, leave it out.

+User-provided focus or constraints (honor these):
 $ARGUMENTS
+
+## How to investigate
+
+Read the highest-value sources first:
+- `README*`, root manifests, workspace config, lockfiles
+- build, test, lint, formatter, typecheck, and codegen config
+- CI workflows and pre-commit / task runner config
+- existing instruction files (`AGENTS.md`, `CLAUDE.md`, `.cursor/rules/`, `.cursorrules`, `.github/copilot-instructions.md`)
+- repo-local OpenCode config such as `opencode.json`
+
+If architecture is still unclear after reading config and docs, inspect a small number of representative code files to find the real entrypoints, package boundaries, and execution flow. Prefer reading the files that explain how the system is wired together over random leaf files.
+
+Prefer executable sources of truth over prose. If docs conflict with config or scripts, trust the executable source and only keep what you can verify.
+
+## What to extract
+
+Look for the highest-signal facts for an agent working in this repo:
+- exact developer commands, especially non-obvious ones
+- how to run a single test, a single package, or a focused verification step
+- required command order when it matters, such as `lint -> typecheck -> test`
+- monorepo or multi-package boundaries, ownership of major directories, and the real app/library entrypoints
+- framework or toolchain quirks: generated code, migrations, codegen, build artifacts, special env loading, dev servers, infra deploy flow
+- repo-specific style or workflow conventions that differ from defaults
+- testing quirks: fixtures, integration test prerequisites, snapshot workflows, required services, flaky or expensive suites
+- important constraints from existing instruction files worth preserving
+
+Good `AGENTS.md` content is usually hard-earned context that took reading multiple files to infer.
+
+## Questions
+
+Only ask the user questions if the repo cannot answer something important. Use the `question` tool for one short batch at most.
+
+Good questions:
+- undocumented team conventions
+- branch / PR / release expectations
+- missing setup or test prerequisites that are known but not written down
+
+Do not ask about anything the repo already makes clear.
+
+## Writing rules
+
+Include only high-signal, repo-specific guidance such as:
+- exact commands and shortcuts the agent would otherwise guess wrong
+- architecture notes that are not obvious from filenames
+- conventions that differ from language or framework defaults
+- setup requirements, environment quirks, and operational gotchas
+- references to existing instruction sources that matter
+
+Exclude:
+- generic software advice
+- long tutorials or exhaustive file trees
+- obvious language conventions
+- speculative claims or anything you could not verify
+- content better stored in another file referenced via `opencode.json` `instructions`
+
+When in doubt, omit.
+
+Prefer short sections and bullets. If the repo is simple, keep the file simple. If the repo is large, summarize the few structural facts that actually change how an agent should work.
+
+If `AGENTS.md` already exists at `${path}`, improve it in place rather than rewriting blindly. Preserve verified useful guidance, delete fluff or stale claims, and reconcile it with the current codebase.
@@ -386,9 +386,17 @@ export const make = Effect.gen(function* () {
                if (code !== 0 && Predicate.isNotNull(code)) return yield* Effect.ignore(kill(killGroup))
                return yield* Effect.void
              }
-              return yield* kill((command, proc, signal) =>
-                Effect.catch(killGroup(command, proc, signal), () => killOne(command, proc, signal)),
-              ).pipe(Effect.andThen(Deferred.await(signal)), Effect.ignore)
+              const send = (s: NodeJS.Signals) =>
+                Effect.catch(killGroup(command, proc, s), () => killOne(command, proc, s))
+              const sig = command.options.killSignal ?? "SIGTERM"
+              const attempt = send(sig).pipe(Effect.andThen(Deferred.await(signal)), Effect.asVoid)
+              const escalated = command.options.forceKillAfter
+                ? Effect.timeoutOrElse(attempt, {
+                    duration: command.options.forceKillAfter,
+                    orElse: () => send("SIGKILL").pipe(Effect.andThen(Deferred.await(signal)), Effect.asVoid),
+                  })
+                : attempt
+              return yield* Effect.ignore(escalated)
            }),
          )

@@ -413,14 +421,17 @@ export const make = Effect.gen(function* () {
                ),
              )
            }),
-            kill: (opts?: ChildProcess.KillOptions) =>
-              timeout(
-                proc,
-                command,
-                opts,
-              )((command, proc, signal) =>
-                Effect.catch(killGroup(command, proc, signal), () => killOne(command, proc, signal)),
-              ).pipe(Effect.andThen(Deferred.await(signal)), Effect.asVoid),
+            kill: (opts?: ChildProcess.KillOptions) => {
+              const sig = opts?.killSignal ?? "SIGTERM"
+              const send = (s: NodeJS.Signals) =>
+                Effect.catch(killGroup(command, proc, s), () => killOne(command, proc, s))
+              const attempt = send(sig).pipe(Effect.andThen(Deferred.await(signal)), Effect.asVoid)
+              if (!opts?.forceKillAfter) return attempt
+              return Effect.timeoutOrElse(attempt, {
+                duration: opts.forceKillAfter,
+                orElse: () => send("SIGKILL").pipe(Effect.andThen(Deferred.await(signal)), Effect.asVoid),
+              })
+            },
          })
        }
        case "PipedCommand": {
@@ -24,9 +24,9 @@ export namespace InstanceState {
    return ((...args: any[]) => Instance.restore(ctx, () => fn(...args))) as F
  }

-  export const context = Effect.fnUntraced(function* () {
+  export const context = Effect.gen(function* () {
    return (yield* InstanceRef) ?? Instance.current
-  })()
+  })

  export const directory = Effect.map(context, (ctx) => ctx.directory)

@@ -37,9 +37,9 @@ export namespace InstanceState {
      const cache = yield* ScopedCache.make<string, A, E, R>({
        capacity: Number.POSITIVE_INFINITY,
        lookup: () =>
-          Effect.fnUntraced(function* () {
+          Effect.gen(function* () {
            return yield* init(yield* context)
-          })(),
+          }),
      })

      const off = registerDisposer((directory) => Effect.runPromise(ScopedCache.invalidate(cache, directory)))
@@ -5,7 +5,6 @@ import { AppFileSystem } from "@/filesystem"
 import { git } from "@/util/git"
 import { Effect, Layer, ServiceMap } from "effect"
 import { formatPatch, structuredPatch } from "diff"
-import fs from "fs"
 import fuzzysort from "fuzzysort"
 import ignore from "ignore"
 import path from "path"
@@ -359,49 +358,46 @@ export namespace File {
        const isGlobalHome = Instance.directory === Global.Path.home && Instance.project.id === "global"
        const next: Entry = { files: [], dirs: [] }

-        yield* Effect.promise(async () => {
-          if (isGlobalHome) {
-            const dirs = new Set<string>()
-            const protectedNames = Protected.names()
-            const ignoreNested = new Set(["node_modules", "dist", "build", "target", "vendor"])
-            const shouldIgnoreName = (name: string) => name.startsWith(".") || protectedNames.has(name)
-            const shouldIgnoreNested = (name: string) => name.startsWith(".") || ignoreNested.has(name)
-            const top = await fs.promises
-              .readdir(Instance.directory, { withFileTypes: true })
-              .catch(() => [] as fs.Dirent[])
+        if (isGlobalHome) {
+          const dirs = new Set<string>()
+          const protectedNames = Protected.names()
+          const ignoreNested = new Set(["node_modules", "dist", "build", "target", "vendor"])
+          const shouldIgnoreName = (name: string) => name.startsWith(".") || protectedNames.has(name)
+          const shouldIgnoreNested = (name: string) => name.startsWith(".") || ignoreNested.has(name)
+          const top = yield* appFs.readDirectoryEntries(Instance.directory).pipe(Effect.orElseSucceed(() => []))

-            for (const entry of top) {
-              if (!entry.isDirectory()) continue
-              if (shouldIgnoreName(entry.name)) continue
-              dirs.add(entry.name + "/")
+          for (const entry of top) {
+            if (entry.type !== "directory") continue
+            if (shouldIgnoreName(entry.name)) continue
+            dirs.add(entry.name + "/")

-              const base = path.join(Instance.directory, entry.name)
-              const children = await fs.promises.readdir(base, { withFileTypes: true }).catch(() => [] as fs.Dirent[])
-              for (const child of children) {
-                if (!child.isDirectory()) continue
-                if (shouldIgnoreNested(child.name)) continue
-                dirs.add(entry.name + "/" + child.name + "/")
-              }
-            }
-
-            next.dirs = Array.from(dirs).toSorted()
-          } else {
-            const seen = new Set<string>()
-            for await (const file of Ripgrep.files({ cwd: Instance.directory })) {
-              next.files.push(file)
-              let current = file
-              while (true) {
-                const dir = path.dirname(current)
-                if (dir === ".") break
-                if (dir === current) break
-                current = dir
-                if (seen.has(dir)) continue
-                seen.add(dir)
-                next.dirs.push(dir + "/")
-              }
+            const base = path.join(Instance.directory, entry.name)
+            const children = yield* appFs.readDirectoryEntries(base).pipe(Effect.orElseSucceed(() => []))
+            for (const child of children) {
+              if (child.type !== "directory") continue
+              if (shouldIgnoreNested(child.name)) continue
+              dirs.add(entry.name + "/" + child.name + "/")
            }
          }
-        })
+
+          next.dirs = Array.from(dirs).toSorted()
+        } else {
+          const files = yield* Effect.promise(() => Array.fromAsync(Ripgrep.files({ cwd: Instance.directory })))
+          const seen = new Set<string>()
+          for (const file of files) {
+            next.files.push(file)
+            let current = file
+            while (true) {
+              const dir = path.dirname(current)
+              if (dir === ".") break
+              if (dir === current) break
+              current = dir
+              if (seen.has(dir)) continue
+              seen.add(dir)
+              next.dirs.push(dir + "/")
+            }
+          }
+        }

        const s = yield* InstanceState.get(state)
        s.cache = next
@@ -636,30 +632,27 @@ export namespace File {
        yield* ensure()
        const { cache } = yield* InstanceState.get(state)

-        return yield* Effect.promise(async () => {
-          const query = input.query.trim()
-          const limit = input.limit ?? 100
-          const kind = input.type ?? (input.dirs === false ? "file" : "all")
-          log.info("search", { query, kind })
+        const query = input.query.trim()
+        const limit = input.limit ?? 100
+        const kind = input.type ?? (input.dirs === false ? "file" : "all")
+        log.info("search", { query, kind })

-          const result = cache
-          const preferHidden = query.startsWith(".") || query.includes("/.")
+        const preferHidden = query.startsWith(".") || query.includes("/.")

-          if (!query) {
-            if (kind === "file") return result.files.slice(0, limit)
-            return sortHiddenLast(result.dirs.toSorted(), preferHidden).slice(0, limit)
-          }
+        if (!query) {
+          if (kind === "file") return cache.files.slice(0, limit)
+          return sortHiddenLast(cache.dirs.toSorted(), preferHidden).slice(0, limit)
+        }

-          const items =
-            kind === "file" ? result.files : kind === "directory" ? result.dirs : [...result.files, ...result.dirs]
+        const items =
+          kind === "file" ? cache.files : kind === "directory" ? cache.dirs : [...cache.files, ...cache.dirs]

-          const searchLimit = kind === "directory" && !preferHidden ? limit * 20 : limit
-          const sorted = fuzzysort.go(query, items, { limit: searchLimit }).map((item) => item.target)
-          const output = kind === "directory" ? sortHiddenLast(sorted, preferHidden).slice(0, limit) : sorted
+        const searchLimit = kind === "directory" && !preferHidden ? limit * 20 : limit
+        const sorted = fuzzysort.go(query, items, { limit: searchLimit }).map((item) => item.target)
+        const output = kind === "directory" ? sortHiddenLast(sorted, preferHidden).slice(0, limit) : sorted

-          log.info("search", { query, kind, results: output.length })
-          return output
-        })
+        log.info("search", { query, kind, results: output.length })
+        return output
      })

      log.info("init")
@@ -111,26 +111,25 @@ export namespace ProviderAuth {

  export class Service extends ServiceMap.Service<Service, Interface>()("@opencode/ProviderAuth") {}

-  export const layer = Layer.effect(
+  export const layer: Layer.Layer<Service, never, Auth.Service | Plugin.Service> = Layer.effect(
    Service,
    Effect.gen(function* () {
      const auth = yield* Auth.Service
+      const plugin = yield* Plugin.Service
      const state = yield* InstanceState.make<State>(
-        Effect.fn("ProviderAuth.state")(() =>
-          Effect.promise(async () => {
-            const plugins = await Plugin.list()
-            return {
-              hooks: Record.fromEntries(
-                Arr.filterMap(plugins, (x) =>
-                  x.auth?.provider !== undefined
-                    ? Result.succeed([ProviderID.make(x.auth.provider), x.auth] as const)
-                    : Result.failVoid,
-                ),
+        Effect.fn("ProviderAuth.state")(function* () {
+          const plugins = yield* plugin.list()
+          return {
+            hooks: Record.fromEntries(
+              Arr.filterMap(plugins, (x) =>
+                x.auth?.provider !== undefined
+                  ? Result.succeed([ProviderID.make(x.auth.provider), x.auth] as const)
+                  : Result.failVoid,
              ),
-              pending: new Map<ProviderID, AuthOAuthResult>(),
-            }
-          }),
-        ),
+            ),
+            pending: new Map<ProviderID, AuthOAuthResult>(),
+          }
+        }),
      )

      const methods = Effect.fn("ProviderAuth.methods")(function* () {
@@ -230,7 +229,9 @@ export namespace ProviderAuth {
    }),
  )

-  export const defaultLayer = layer.pipe(Layer.provide(Auth.defaultLayer))
+  export const defaultLayer = Layer.suspend(() =>
+    layer.pipe(Layer.provide(Auth.defaultLayer), Layer.provide(Plugin.defaultLayer)),
+  )

  const { runPromise } = makeRuntime(Service, defaultLayer)

@@ -961,11 +961,12 @@ export namespace Provider {
    }
  }

-  const layer: Layer.Layer<Service, never, Config.Service | Auth.Service> = Layer.effect(
+  const layer: Layer.Layer<Service, never, Config.Service | Auth.Service | Plugin.Service> = Layer.effect(
    Service,
    Effect.gen(function* () {
      const config = yield* Config.Service
      const auth = yield* Auth.Service
+      const plugin = yield* Plugin.Service

      const state = yield* InstanceState.make<State>(() =>
        Effect.gen(function* () {
@@ -1128,7 +1129,7 @@ export namespace Provider {
            }
          }

-          const plugins = yield* Effect.promise(() => Plugin.list())
+          const plugins = yield* plugin.list()
          for (const plugin of plugins) {
            if (!plugin.auth) continue
            const providerID = ProviderID.make(plugin.auth.provider)
@@ -1541,7 +1542,13 @@ export namespace Provider {
    }),
  )

-  export const defaultLayer = layer.pipe(Layer.provide(Config.defaultLayer), Layer.provide(Auth.defaultLayer))
+  export const defaultLayer = Layer.suspend(() =>
+    layer.pipe(
+      Layer.provide(Config.defaultLayer),
+      Layer.provide(Auth.defaultLayer),
+      Layer.provide(Plugin.defaultLayer),
+    ),
+  )

  const { runPromise } = makeRuntime(Service, defaultLayer)

@@ -28,7 +28,9 @@ import { ReadTool } from "../tool/read"
 import { FileTime } from "../file/time"
 import { Flag } from "../flag/flag"
 import { ulid } from "ulid"
-import { spawn } from "child_process"
+import { ChildProcess, ChildProcessSpawner } from "effect/unstable/process"
+import * as CrossSpawnSpawner from "@/effect/cross-spawn-spawner"
+import * as Stream from "effect/Stream"
 import { Command } from "../command"
 import { pathToFileURL, fileURLToPath } from "url"
 import { ConfigMarkdown } from "../config/markdown"
@@ -96,6 +98,7 @@ export namespace SessionPrompt {
      const filetime = yield* FileTime.Service
      const registry = yield* ToolRegistry.Service
      const truncate = yield* Truncate.Service
+      const spawner = yield* ChildProcessSpawner.ChildProcessSpawner
      const scope = yield* Scope.Scope

      const state = yield* InstanceState.make(
@@ -809,22 +812,26 @@ NOTE: At any point in time through this workflow you should feel free to ask the
          fish: { args: ["-c", input.command] },
          zsh: {
            args: [
-              "-c",
              "-l",
+              "-c",
              `
+                __oc_cwd=$PWD
                [[ -f ~/.zshenv ]] && source ~/.zshenv >/dev/null 2>&1 || true
                [[ -f "\${ZDOTDIR:-$HOME}/.zshrc" ]] && source "\${ZDOTDIR:-$HOME}/.zshrc" >/dev/null 2>&1 || true
+                cd "$__oc_cwd"
                eval ${JSON.stringify(input.command)}
              `,
            ],
          },
          bash: {
            args: [
-              "-c",
              "-l",
+              "-c",
              `
+                __oc_cwd=$PWD
                shopt -s expand_aliases
                [[ -f ~/.bashrc ]] && source ~/.bashrc >/dev/null 2>&1 || true
+                cd "$__oc_cwd"
                eval ${JSON.stringify(input.command)}
              `,
            ],
@@ -832,7 +839,7 @@ NOTE: At any point in time through this workflow you should feel free to ask the
          cmd: { args: ["/c", input.command] },
          powershell: { args: ["-NoProfile", "-Command", input.command] },
          pwsh: { args: ["-NoProfile", "-Command", input.command] },
-          "": { args: ["-c", `${input.command}`] },
+          "": { args: ["-c", input.command] },
        }

        const args = (invocations[shellName] ?? invocations[""]).args
@@ -842,51 +849,20 @@ NOTE: At any point in time through this workflow you should feel free to ask the
          { cwd, sessionID: input.sessionID, callID: part.callID },
          { env: {} },
        )
-        const proc = yield* Effect.sync(() =>
-          spawn(sh, args, {
-            cwd,
-            detached: process.platform !== "win32",
-            windowsHide: process.platform === "win32",
-            stdio: ["ignore", "pipe", "pipe"],
-            env: {
-              ...process.env,
-              ...shellEnv.env,
-              TERM: "dumb",
-            },
-          }),
-        )
+
+        const cmd = ChildProcess.make(sh, args, {
+          cwd,
+          extendEnv: true,
+          env: { ...shellEnv.env, TERM: "dumb" },
+          stdin: "ignore",
+          forceKillAfter: "3 seconds",
+        })

        let output = ""
-        const write = () => {
-          if (part.state.status !== "running") return
-          part.state.metadata = { output, description: "" }
-          void Effect.runFork(sessions.updatePart(part))
-        }
-
-        proc.stdout?.on("data", (chunk) => {
-          output += chunk.toString()
-          write()
-        })
-        proc.stderr?.on("data", (chunk) => {
-          output += chunk.toString()
-          write()
-        })
-
        let aborted = false
-        let exited = false
-        let finished = false
-        const kill = Effect.promise(() => Shell.killTree(proc, { exited: () => exited }))
-
-        const abortHandler = () => {
-          if (aborted) return
-          aborted = true
-          void Effect.runFork(kill)
-        }

        const finish = Effect.uninterruptible(
          Effect.gen(function* () {
-            if (finished) return
-            finished = true
            if (aborted) {
              output += "\n\n" + ["<metadata>", "User aborted the command", "</metadata>"].join("\n")
            }
@@ -908,20 +884,26 @@ NOTE: At any point in time through this workflow you should feel free to ask the
          }),
        )

-        const exit = yield* Effect.promise(() => {
-          signal.addEventListener("abort", abortHandler, { once: true })
-          if (signal.aborted) abortHandler()
-          return new Promise<void>((resolve) => {
-            const close = () => {
-              exited = true
-              proc.off("close", close)
-              resolve()
-            }
-            proc.once("close", close)
-          })
+        const exit = yield* Effect.gen(function* () {
+          const handle = yield* spawner.spawn(cmd)
+          yield* Stream.runForEach(Stream.decodeText(handle.all), (chunk) =>
+            Effect.sync(() => {
+              output += chunk
+              if (part.state.status === "running") {
+                part.state.metadata = { output, description: "" }
+                void Effect.runFork(sessions.updatePart(part))
+              }
+            }),
+          )
+          yield* handle.exitCode
        }).pipe(
-          Effect.onInterrupt(() => Effect.sync(abortHandler)),
-          Effect.ensuring(Effect.sync(() => signal.removeEventListener("abort", abortHandler))),
+          Effect.scoped,
+          Effect.onInterrupt(() =>
+            Effect.sync(() => {
+              aborted = true
+            }),
+          ),
+          Effect.orDie,
          Effect.ensuring(finish),
          Effect.exit,
        )
@@ -1735,6 +1717,7 @@ NOTE: At any point in time through this workflow you should feel free to ask the
        Layer.provide(Session.defaultLayer),
        Layer.provide(Agent.defaultLayer),
        Layer.provide(Bus.layer),
+        Layer.provide(CrossSpawnSpawner.defaultLayer),
      ),
    ),
  )
@@ -150,15 +150,11 @@ export namespace SessionSummary {
    }),
  )

-  export const defaultLayer = Layer.unwrap(
-    Effect.sync(() =>
-      layer.pipe(
-        Layer.provide(Session.defaultLayer),
-        Layer.provide(Snapshot.defaultLayer),
-        Layer.provide(Storage.defaultLayer),
-        Layer.provide(Bus.layer),
-      ),
-    ),
+  export const defaultLayer = layer.pipe(
+    Layer.provide(Session.defaultLayer),
+    Layer.provide(Snapshot.defaultLayer),
+    Layer.provide(Storage.defaultLayer),
+    Layer.provide(Bus.layer),
  )

  const { runPromise } = makeRuntime(Service, defaultLayer)
@@ -148,6 +148,70 @@ it.live("token refresh persists the new token", () =>
  }),
 )

+it.live("concurrent config and token requests coalesce token refresh", () =>
+  Effect.gen(function* () {
+    const id = AccountID.make("user-1")
+
+    yield* AccountRepo.use((r) =>
+      r.persistAccount({
+        id,
+        email: "user@example.com",
+        url: "https://one.example.com",
+        accessToken: AccessToken.make("at_old"),
+        refreshToken: RefreshToken.make("rt_old"),
+        expiry: Date.now() - 1_000,
+        orgID: Option.some(OrgID.make("org-9")),
+      }),
+    )
+
+    let refreshCalls = 0
+    const client = HttpClient.make((req) =>
+      Effect.promise(async () => {
+        if (req.url === "https://one.example.com/auth/device/token") {
+          refreshCalls += 1
+
+          if (refreshCalls === 1) {
+            await new Promise((resolve) => setTimeout(resolve, 25))
+            return json(req, {
+              access_token: "at_new",
+              refresh_token: "rt_new",
+              expires_in: 60,
+            })
+          }
+
+          return json(
+            req,
+            {
+              error: "invalid_grant",
+              error_description: "refresh token already used",
+            },
+            400,
+          )
+        }
+
+        if (req.url === "https://one.example.com/api/config") {
+          return json(req, { config: { theme: "light", seats: 5 } })
+        }
+
+        return json(req, {}, 404)
+      }),
+    )
+
+    const [cfg, token] = yield* Account.Service.use((s) =>
+      Effect.all([s.config(id, OrgID.make("org-9")), s.token(id)], { concurrency: 2 }),
+    ).pipe(Effect.provide(live(client)))
+
+    expect(Option.getOrThrow(cfg)).toEqual({ theme: "light", seats: 5 })
+    expect(String(Option.getOrThrow(token))).toBe("at_new")
+    expect(refreshCalls).toBe(1)
+
+    const row = yield* AccountRepo.use((r) => r.getRow(id))
+    const value = Option.getOrThrow(row)
+    expect(value.access_token).toBe(AccessToken.make("at_new"))
+    expect(value.refresh_token).toBe(RefreshToken.make("rt_new"))
+  }),
+)
+
 it.live("config sends the selected org header", () =>
  Effect.gen(function* () {
    const id = AccountID.make("user-1")
@@ -20,6 +20,13 @@ type Hit = {
  body: Record<string, unknown>
 }

+type Match = (hit: Hit) => boolean
+
+type Queue = {
+  item: Item
+  match?: Match
+}
+
 type Wait = {
  count: number
  ready: Deferred.Deferred<void>
@@ -420,7 +427,7 @@ const reset = Effect.fn("TestLLMServer.reset")(function* (item: Sse) {
    for (const part of item.tail) res.write(line(part))
    res.destroy(new Error("connection reset"))
  })
-  yield* Effect.never
+  return yield* Effect.never
 })

 function fail(item: HttpError) {
@@ -581,6 +588,9 @@ namespace TestLLMServer {
  export interface Service {
    readonly url: string
    readonly push: (...input: (Item | Reply)[]) => Effect.Effect<void>
+    readonly pushMatch: (match: Match, ...input: (Item | Reply)[]) => Effect.Effect<void>
+    readonly textMatch: (match: Match, value: string, opts?: { usage?: Usage }) => Effect.Effect<void>
+    readonly toolMatch: (match: Match, name: string, input: unknown) => Effect.Effect<void>
    readonly text: (value: string, opts?: { usage?: Usage }) => Effect.Effect<void>
    readonly tool: (name: string, input: unknown) => Effect.Effect<void>
    readonly toolHang: (name: string, input: unknown) => Effect.Effect<void>
@@ -605,11 +615,15 @@ export class TestLLMServer extends ServiceMap.Service<TestLLMServer, TestLLMServ
      const router = yield* HttpRouter.HttpRouter

      let hits: Hit[] = []
-      let list: Item[] = []
+      let list: Queue[] = []
      let waits: Wait[] = []

      const queue = (...input: (Item | Reply)[]) => {
-        list = [...list, ...input.map(item)]
+        list = [...list, ...input.map((value) => ({ item: item(value) }))]
+      }
+
+      const queueMatch = (match: Match, ...input: (Item | Reply)[]) => {
+        list = [...list, ...input.map((value) => ({ item: item(value), match }))]
      }

      const notify = Effect.fnUntraced(function* () {
@@ -619,19 +633,21 @@ export class TestLLMServer extends ServiceMap.Service<TestLLMServer, TestLLMServ
        yield* Effect.forEach(ready, (item) => Deferred.succeed(item.ready, void 0))
      })

-      const pull = () => {
-        const first = list[0]
-        if (!first) return
-        list = list.slice(1)
-        return first
+      const pull = (hit: Hit) => {
+        const index = list.findIndex((entry) => !entry.match || entry.match(hit))
+        if (index === -1) return
+        const first = list[index]
+        list = [...list.slice(0, index), ...list.slice(index + 1)]
+        return first.item
      }

      const handle = Effect.fn("TestLLMServer.handle")(function* (mode: "chat" | "responses") {
        const req = yield* HttpServerRequest.HttpServerRequest
-        const next = pull()
-        if (!next) return HttpServerResponse.text("unexpected request", { status: 500 })
        const body = yield* req.json.pipe(Effect.orElseSucceed(() => ({})))
-        hits = [...hits, hit(req.originalUrl, body)]
+        const current = hit(req.originalUrl, body)
+        const next = pull(current)
+        if (!next) return HttpServerResponse.text("unexpected request", { status: 500 })
+        hits = [...hits, current]
        yield* notify()
        if (next.type !== "sse") return fail(next)
        if (mode === "responses") return send(responses(next, modelFrom(body)))
@@ -655,6 +671,21 @@ export class TestLLMServer extends ServiceMap.Service<TestLLMServer, TestLLMServ
        push: Effect.fn("TestLLMServer.push")(function* (...input: (Item | Reply)[]) {
          queue(...input)
        }),
+        pushMatch: Effect.fn("TestLLMServer.pushMatch")(function* (match: Match, ...input: (Item | Reply)[]) {
+          queueMatch(match, ...input)
+        }),
+        textMatch: Effect.fn("TestLLMServer.textMatch")(function* (
+          match: Match,
+          value: string,
+          opts?: { usage?: Usage },
+        ) {
+          const out = reply().text(value)
+          if (opts?.usage) out.usage(opts.usage)
+          queueMatch(match, out.stop().item())
+        }),
+        toolMatch: Effect.fn("TestLLMServer.toolMatch")(function* (match: Match, name: string, input: unknown) {
+          queueMatch(match, reply().tool(name, input).item())
+        }),
        text: Effect.fn("TestLLMServer.text")(function* (value: string, opts?: { usage?: Usage }) {
          const out = reply().text(value)
          if (opts?.usage) out.usage(opts.usage)
@@ -1,6 +1,7 @@
 import { NodeFileSystem } from "@effect/platform-node"
 import { expect, spyOn } from "bun:test"
 import { Cause, Effect, Exit, Fiber, Layer } from "effect"
+import path from "path"
 import z from "zod"
 import type { Agent } from "../../src/agent/agent"
 import { Agent as AgentSvc } from "../../src/agent/agent"
@@ -887,6 +888,79 @@ unix("shell captures stdout and stderr in completed tool output", () =>
  ),
 )

+unix("shell completes a fast command on the preferred shell", () =>
+  provideTmpdirInstance(
+    (dir) =>
+      Effect.gen(function* () {
+        const { prompt, chat } = yield* boot()
+        const result = yield* prompt.shell({
+          sessionID: chat.id,
+          agent: "build",
+          command: "pwd",
+        })
+
+        expect(result.info.role).toBe("assistant")
+        const tool = completedTool(result.parts)
+        if (!tool) return
+
+        expect(tool.state.input.command).toBe("pwd")
+        expect(tool.state.output).toContain(dir)
+        expect(tool.state.metadata.output).toContain(dir)
+        yield* prompt.assertNotBusy(chat.id)
+      }),
+    { git: true, config: cfg },
+  ),
+)
+
+unix("shell lists files from the project directory", () =>
+  provideTmpdirInstance(
+    (dir) =>
+      Effect.gen(function* () {
+        const { prompt, chat } = yield* boot()
+        yield* Effect.promise(() => Bun.write(path.join(dir, "README.md"), "# e2e\n"))
+
+        const result = yield* prompt.shell({
+          sessionID: chat.id,
+          agent: "build",
+          command: "command ls",
+        })
+
+        expect(result.info.role).toBe("assistant")
+        const tool = completedTool(result.parts)
+        if (!tool) return
+
+        expect(tool.state.input.command).toBe("command ls")
+        expect(tool.state.output).toContain("README.md")
+        expect(tool.state.metadata.output).toContain("README.md")
+        yield* prompt.assertNotBusy(chat.id)
+      }),
+    { git: true, config: cfg },
+  ),
+)
+
+unix("shell captures stderr from a failing command", () =>
+  provideTmpdirInstance(
+    (dir) =>
+      Effect.gen(function* () {
+        const { prompt, chat } = yield* boot()
+        const result = yield* prompt.shell({
+          sessionID: chat.id,
+          agent: "build",
+          command: "command -v __nonexistent_cmd_e2e__ || echo 'not found' >&2; exit 1",
+        })
+
+        expect(result.info.role).toBe("assistant")
+        const tool = completedTool(result.parts)
+        if (!tool) return
+
+        expect(tool.state.output).toContain("not found")
+        expect(tool.state.metadata.output).toContain("not found")
+        yield* prompt.assertNotBusy(chat.id)
+      }),
+    { git: true, config: cfg },
+  ),
+)
+
 unix(
  "shell updates running metadata before process exit",
  () =>
@@ -15,9 +15,17 @@ To create a new `AGENTS.md` file, you can run the `/init` command in opencode.
 You should commit your project's `AGENTS.md` file to Git.
 :::

-This will scan your project and all its contents to understand what the project is about and generate an `AGENTS.md` file with it. This helps opencode to navigate the project better.
+`/init` scans the important files in your repo, may ask a couple of targeted questions when the codebase cannot answer them, and then creates or updates `AGENTS.md` with concise project-specific guidance.

-If you have an existing `AGENTS.md` file, this will try to add to it.
+It focuses on the things future agent sessions are most likely to need:
+
+- build, lint, and test commands
+- command order and focused verification steps when they matter
+- architecture and repo structure that are not obvious from filenames alone
+- project-specific conventions, setup quirks, and operational gotchas
+- references to existing instruction sources like Cursor or Copilot rules
+
+If you already have an `AGENTS.md`, `/init` will improve it in place instead of blindly replacing it.

 ---

@@ -153,7 +153,7 @@ Show the help dialog.

 ### init

-Create or update `AGENTS.md` file. [Learn more](/docs/rules).
+Guided setup for creating or updating `AGENTS.md`. [Learn more](/docs/rules).

 ```bash frame="none"
 /init
Author	SHA1	Message	Date
Kit Langton	839efafad6	refactor(summary): remove unnecessary Layer.unwrap from defaultLayer	2026-04-01 14:21:29 -04:00
Kit Langton	c619caefdd	fix(account): coalesce concurrent console token refreshes (#20503 )	2026-04-01 13:16:35 -04:00
Kit Langton	c559af51ce	test(app): migrate more e2e suites to isolated backend (#20505 )	2026-04-01 13:15:42 -04:00
github-actions[bot]	d1e0a4640c	Update VOUCHED list https://github.com/anomalyco/opencode/issues/20482#issuecomment-4171492178	2026-04-01 16:50:21 +00:00
opencode-agent[bot]	f9e71ec515	chore: update nix node_modules hashes	2026-04-01 16:47:33 +00:00
opencode-agent[bot]	ef538c9707	chore: generate	2026-04-01 16:14:37 +00:00
Kit Langton	2f405daa98	refactor: use Effect services instead of async facades in provider, auth, and file (#20480 )	2026-04-01 16:13:13 +00:00
Kit Langton	a9c85b7c27	refactor(shell): use Effect ChildProcess for shell command execution (#20494 )	2026-04-01 12:07:57 -04:00
Shoubhit Dash	897d83c589	refactor(init): tighten AGENTS guidance (#20422 )	2026-04-01 21:37:25 +05:30
opencode-agent[bot]	0a125e5d4d	chore: generate	2026-04-01 15:59:28 +00:00
Kit Langton	38d2276592	test(e2e): isolate prompt tests with per-worker backend (#20464 )	2026-04-01 15:58:11 +00:00