Merge pull request #33 from madhura68/feat/sprint-worker

PBI-9 + PBI-47: worktree foundation, product-worktrees, P0 fixes, PAUSED flow
2026-05-06 21:35:47 +02:00 · 2026-05-06 21:35:47 +02:00 · eccc75ca56
commit eccc75ca56
parent 4598efde58 f7f5a487ec
29 changed files with 1731 additions and 46 deletions
--- a/README.md
+++ b/README.md
@ -335,3 +335,33 @@ npx @modelcontextprotocol/inspector node dist/index.js
 - **Production database** — verify against a preview database before
  running against prod. The token check enforces user scope but does
  not gate reads of unrelated products you happen to be a member of.
+
+## Worktrees
+
+Scrum4Me-mcp uses git worktrees rooted at `~/.scrum4me-agent-worktrees/` (override via `SCRUM4ME_AGENT_WORKTREE_DIR`).
+
+### Two kinds of worktrees
+
+- **Per-job task-worktrees** (`<jobId>/`) — one per `TASK_IMPLEMENTATION` job. Created at claim, cleaned up on `DONE`/`FAILED`/`CANCELLED` via `cleanup_my_worktrees`.
+- **Persistent product-worktrees** (`_products/<productId>/`) — one per product with `repo_url`, used by `IDEA_GRILL` and `IDEA_MAKE_PLAN`. **Detached HEAD on `origin/main`**, hard-reset at every job start. `.scratch/` holds throw-away work and is wiped on each claim.
+
+### Concurrency: file-locks
+
+Product-worktrees are serialised via `proper-lockfile` on `_products/<productId>.lock`. Two parallel idea-jobs on the same product wait for each other. For multi-product idea-jobs, locks are acquired in alphabetical order to prevent deadlocks.
+
+### Single-host invariant
+
+`proper-lockfile` only works when all MCP-server processes run on the same host. Migrate to Postgres `pg_advisory_lock` when:
+- multiple MCP instances on different machines serve workers, or
+- the worktree directory is shared over NFS/CIFS.
+
+Migration path: replace `acquireFileLock` in `src/git/file-lock.ts` with a `pg_try_advisory_lock(hashtext(path)::bigint)` wrapper via the existing Prisma connection. The API stays identical.
+
+### Manual cleanup
+
+`cleanup_my_worktrees` skips `_products/` and `*.lock` automatically. To clean up a product-worktree manually (after archive or repo-rename):
+
+```bash
+git worktree remove --force ~/.scrum4me-agent-worktrees/_products/<productId>
+rm ~/.scrum4me-agent-worktrees/_products/<productId>.lock  # if still present
+```
--- a/tests/cleanup-my-worktrees.test.ts
+++ b/tests/cleanup-my-worktrees.test.ts
@ -73,6 +73,17 @@ describe('listWorktreeJobIds', () => {
    mockReaddir.mockRejectedValue(Object.assign(new Error('ENOENT'), { code: 'ENOENT' }))
    expect(await listWorktreeJobIds(WORKTREE_PARENT)).toEqual([])
  })
+
+  it('skips _products/ system dir and *.lock files (PBI-9)', async () => {
+    mockReaddir.mockResolvedValue([
+      makeDirent('job-aaa'),
+      makeDirent('_products'),
+      makeDirent('product-abc.lock'),
+      makeDirent('job-bbb'),
+    ])
+    const ids = await listWorktreeJobIds(WORKTREE_PARENT)
+    expect(ids).toEqual(['job-aaa', 'job-bbb'])
+  })
 })

 describe('cleanupWorktrees', () => {
--- a/tests/flow/effects.test.ts
+++ b/tests/flow/effects.test.ts
@ -0,0 +1,22 @@
+import { describe, it, expect } from 'vitest'
+import { executeEffects } from '../../src/flow/effects.js'
+
+describe('effects executor', () => {
+  it('RELEASE_WORKTREE_LOCKS for unknown jobId is a no-op (no throw)', async () => {
+    const out = await executeEffects([{ type: 'RELEASE_WORKTREE_LOCKS', jobId: 'no-such-job' }])
+    expect(out).toEqual([])
+  })
+
+  it('multiple effects execute in order; failure in one is logged but does not abort', async () => {
+    const out = await executeEffects([
+      { type: 'RELEASE_WORKTREE_LOCKS', jobId: 'a' },
+      { type: 'RELEASE_WORKTREE_LOCKS', jobId: 'b' },
+    ])
+    expect(out).toEqual([])
+  })
+
+  it('empty effects array returns empty outcomes', async () => {
+    const out = await executeEffects([])
+    expect(out).toEqual([])
+  })
+})
--- a/tests/flow/pr-flow.test.ts
+++ b/tests/flow/pr-flow.test.ts
@ -0,0 +1,78 @@
+import { describe, it, expect } from 'vitest'
+import { transition, type PrFlowState } from '../../src/flow/pr-flow.js'
+
+describe('pr-flow STORY-mode 3-tasks scenario', () => {
+  it('opens PR early; auto-merge only fires on the last task', () => {
+    let state: PrFlowState = { kind: 'none', strategy: 'STORY' }
+    const allEffects: Array<Record<string, unknown>> = []
+
+    // Task 1 DONE → PR_CREATED
+    let r = transition(state, { type: 'PR_CREATED', prUrl: 'https://github.com/o/r/pull/1' })
+    state = r.nextState
+    allEffects.push(...r.effects)
+    expect(state.kind).toBe('pr_opened')
+    expect(allEffects.filter((e) => e.type === 'ENABLE_AUTO_MERGE')).toHaveLength(0)
+
+    // Task 2 DONE → no STORY_COMPLETED yet, no transition emitted
+    r = transition(state, { type: 'TASK_DONE', taskId: 't2', headSha: 'abc123' })
+    state = r.nextState
+    allEffects.push(...r.effects)
+    expect(state.kind).toBe('pr_opened')
+    expect(allEffects.filter((e) => e.type === 'ENABLE_AUTO_MERGE')).toHaveLength(0)
+
+    // Task 3 DONE = STORY_COMPLETED → ENABLE_AUTO_MERGE with head guard
+    r = transition(state, { type: 'STORY_COMPLETED', storyId: 's1', headSha: 'def456' })
+    state = r.nextState
+    allEffects.push(...r.effects)
+    expect(state.kind).toBe('waiting_for_checks')
+    const enableEffects = allEffects.filter((e) => e.type === 'ENABLE_AUTO_MERGE')
+    expect(enableEffects).toHaveLength(1)
+    expect(enableEffects[0]).toMatchObject({ expectedHeadSha: 'def456' })
+
+    // CI green + merge OK
+    r = transition(state, { type: 'MERGE_RESULT' })
+    state = r.nextState
+    expect(state.kind).toBe('auto_merge_enabled')
+  })
+
+  it('CHECKS_FAILED → checks_failed (no pause)', () => {
+    const state: PrFlowState = {
+      kind: 'waiting_for_checks',
+      strategy: 'STORY',
+      prUrl: 'x',
+      headSha: 'y',
+    }
+    const r = transition(state, { type: 'MERGE_RESULT', reason: 'CHECKS_FAILED' })
+    expect(r.nextState.kind).toBe('checks_failed')
+  })
+
+  it('MERGE_CONFLICT → merge_conflict_paused', () => {
+    const state: PrFlowState = {
+      kind: 'waiting_for_checks',
+      strategy: 'STORY',
+      prUrl: 'x',
+      headSha: 'y',
+    }
+    const r = transition(state, { type: 'MERGE_RESULT', reason: 'MERGE_CONFLICT' })
+    expect(r.nextState.kind).toBe('merge_conflict_paused')
+  })
+})
+
+describe('pr-flow SPRINT-mode', () => {
+  it('draft stays draft until SPRINT_COMPLETED → MARK_PR_READY effect', () => {
+    let state: PrFlowState = { kind: 'none', strategy: 'SPRINT' }
+    let r = transition(state, { type: 'PR_CREATED', prUrl: 'x' })
+    expect(r.nextState.kind).toBe('draft_opened')
+    expect(r.effects).toHaveLength(0)
+
+    state = r.nextState
+    r = transition(state, { type: 'TASK_DONE', taskId: 't1', headSha: 'a' })
+    expect(r.nextState.kind).toBe('draft_opened')
+    expect(r.effects).toHaveLength(0)
+
+    state = r.nextState
+    r = transition(state, { type: 'SPRINT_COMPLETED', sprintRunId: 'sr1' })
+    expect(r.nextState.kind).toBe('ready_for_review')
+    expect(r.effects.filter((e) => e.type === 'MARK_PR_READY')).toHaveLength(1)
+  })
+})
--- a/tests/flow/sprint-run.test.ts
+++ b/tests/flow/sprint-run.test.ts
@ -0,0 +1,82 @@
+import { describe, it, expect } from 'vitest'
+import { transition, type SprintRunState } from '../../src/flow/sprint-run.js'
+
+describe('sprint-run pure transitions', () => {
+  it('queued + CLAIM_FIRST_JOB → running with SET_SPRINT_RUN_STATUS effect', () => {
+    const state: SprintRunState = { kind: 'queued', sprintRunId: 'sr1' }
+    const r = transition(state, { type: 'CLAIM_FIRST_JOB' })
+    expect(r.nextState.kind).toBe('running')
+    expect(r.effects).toEqual([
+      { type: 'SET_SPRINT_RUN_STATUS', sprintRunId: 'sr1', status: 'RUNNING' },
+    ])
+  })
+
+  it('running + MERGE_CONFLICT → paused_merge_conflict + 2 effects in order', () => {
+    const state: SprintRunState = { kind: 'running', sprintRunId: 'sr1' }
+    const r = transition(state, {
+      type: 'MERGE_CONFLICT',
+      prUrl: 'https://github.com/o/r/pull/1',
+      prHeadSha: 'abc123',
+      conflictFiles: ['a.ts', 'b.ts'],
+      resumeInstructions: 'Resolve and push',
+    })
+    expect(r.nextState.kind).toBe('paused_merge_conflict')
+    expect(r.effects).toHaveLength(2)
+    expect(r.effects[0].type).toBe('CREATE_CLAUDE_QUESTION')
+    expect(r.effects[1].type).toBe('SET_SPRINT_RUN_STATUS')
+    if (r.effects[1].type === 'SET_SPRINT_RUN_STATUS') {
+      expect(r.effects[1].status).toBe('PAUSED')
+      expect(r.effects[1].pauseContextDraft).toMatchObject({
+        pause_reason: 'MERGE_CONFLICT',
+        pr_url: 'https://github.com/o/r/pull/1',
+        pr_head_sha: 'abc123',
+        conflict_files: ['a.ts', 'b.ts'],
+      })
+    }
+  })
+
+  it('paused + USER_RESUMED → running + CLOSE_CLAUDE_QUESTION + clear pause_context', () => {
+    const state: SprintRunState = {
+      kind: 'paused_merge_conflict',
+      sprintRunId: 'sr1',
+      pauseContext: {
+        pause_reason: 'MERGE_CONFLICT',
+        pr_url: 'x',
+        pr_head_sha: 'y',
+        conflict_files: [],
+        claude_question_id: 'q1',
+        resume_instructions: 'r',
+        paused_at: new Date().toISOString(),
+      },
+    }
+    const r = transition(state, { type: 'USER_RESUMED' })
+    expect(r.nextState.kind).toBe('running')
+    expect(r.effects[0]).toEqual({ type: 'CLOSE_CLAUDE_QUESTION', questionId: 'q1' })
+    expect(r.effects[1]).toMatchObject({
+      type: 'SET_SPRINT_RUN_STATUS',
+      status: 'RUNNING',
+      clearPauseContext: true,
+    })
+  })
+
+  it('running + TASK_FAILED → failed (no PAUSE)', () => {
+    const state: SprintRunState = { kind: 'running', sprintRunId: 'sr1' }
+    const r = transition(state, { type: 'TASK_FAILED', taskId: 't1', error: 'CI red' })
+    expect(r.nextState.kind).toBe('failed')
+    expect(r.effects[0]).toMatchObject({ status: 'FAILED' })
+  })
+
+  it('running + ALL_DONE → done + SET_SPRINT_RUN_STATUS DONE', () => {
+    const state: SprintRunState = { kind: 'running', sprintRunId: 'sr1' }
+    const r = transition(state, { type: 'ALL_DONE' })
+    expect(r.nextState.kind).toBe('done')
+    expect(r.effects[0]).toMatchObject({ status: 'DONE' })
+  })
+
+  it('forbidden transition (running + CLAIM_FIRST_JOB) keeps state and emits no effects', () => {
+    const state: SprintRunState = { kind: 'running', sprintRunId: 'sr1' }
+    const r = transition(state, { type: 'CLAIM_FIRST_JOB' })
+    expect(r.nextState).toEqual(state)
+    expect(r.effects).toEqual([])
+  })
+})
--- a/tests/flow/worktree-lease.test.ts
+++ b/tests/flow/worktree-lease.test.ts
@ -0,0 +1,82 @@
+import { describe, it, expect } from 'vitest'
+import { transition, type WorktreeLeaseState } from '../../src/flow/worktree-lease.js'
+
+describe('worktree-lease pure transitions', () => {
+  it('idle + JOB_CLAIMED → acquiring_lock, no effects', () => {
+    const r = transition({ kind: 'idle' }, { type: 'JOB_CLAIMED', jobId: 'j1', productIds: ['p1'] })
+    expect(r.nextState.kind).toBe('acquiring_lock')
+    expect(r.effects).toEqual([])
+  })
+
+  it('acquiring_lock + LOCK_ACQUIRED → creating_or_reusing', () => {
+    const state: WorktreeLeaseState = {
+      kind: 'acquiring_lock',
+      jobId: 'j1',
+      productIds: ['p1'],
+    }
+    const r = transition(state, { type: 'LOCK_ACQUIRED' })
+    expect(r.nextState.kind).toBe('creating_or_reusing')
+    expect(r.effects).toEqual([])
+  })
+
+  it('acquiring_lock + LOCK_TIMEOUT → lock_timeout', () => {
+    const state: WorktreeLeaseState = {
+      kind: 'acquiring_lock',
+      jobId: 'j1',
+      productIds: ['p1'],
+    }
+    const r = transition(state, { type: 'LOCK_TIMEOUT' })
+    expect(r.nextState.kind).toBe('lock_timeout')
+  })
+
+  it('creating_or_reusing + WORKTREE_READY → syncing', () => {
+    const r = transition(
+      { kind: 'creating_or_reusing', jobId: 'j1', productIds: ['p1'] },
+      { type: 'WORKTREE_READY' },
+    )
+    expect(r.nextState.kind).toBe('syncing')
+  })
+
+  it('syncing + SYNC_DONE → ready (no release effect yet)', () => {
+    const r = transition(
+      { kind: 'syncing', jobId: 'j1', productIds: ['p1'] },
+      { type: 'SYNC_DONE' },
+    )
+    expect(r.nextState.kind).toBe('ready')
+    expect(r.effects).toEqual([])
+  })
+
+  it('syncing + SYNC_FAILED → sync_failed + RELEASE_WORKTREE_LOCKS effect', () => {
+    const r = transition(
+      { kind: 'syncing', jobId: 'j1', productIds: ['p1'] },
+      { type: 'SYNC_FAILED', error: 'boom' },
+    )
+    expect(r.nextState.kind).toBe('sync_failed')
+    expect(r.effects).toEqual([{ type: 'RELEASE_WORKTREE_LOCKS', jobId: 'j1' }])
+  })
+
+  it('ready + JOB_TERMINAL → releasing + RELEASE_WORKTREE_LOCKS effect', () => {
+    const r = transition(
+      { kind: 'ready', jobId: 'j1', productIds: ['p1'] },
+      { type: 'JOB_TERMINAL', jobId: 'j1' },
+    )
+    expect(r.nextState.kind).toBe('releasing')
+    expect(r.effects).toEqual([{ type: 'RELEASE_WORKTREE_LOCKS', jobId: 'j1' }])
+  })
+
+  it('ready + STALE_RESET → stale_released + RELEASE_WORKTREE_LOCKS effect', () => {
+    const r = transition(
+      { kind: 'ready', jobId: 'j1', productIds: ['p1'] },
+      { type: 'STALE_RESET', jobId: 'j1' },
+    )
+    expect(r.nextState.kind).toBe('stale_released')
+    expect(r.effects).toEqual([{ type: 'RELEASE_WORKTREE_LOCKS', jobId: 'j1' }])
+  })
+
+  it('forbidden transition (idle + LOCK_ACQUIRED) keeps state, no effects', () => {
+    const state: WorktreeLeaseState = { kind: 'idle' }
+    const r = transition(state, { type: 'LOCK_ACQUIRED' })
+    expect(r.nextState).toEqual(state)
+    expect(r.effects).toEqual([])
+  })
+})
--- a/tests/git/file-lock.test.ts
+++ b/tests/git/file-lock.test.ts
@ -0,0 +1,96 @@
+import { describe, it, expect, beforeEach, afterEach } from 'vitest'
+import * as fs from 'node:fs/promises'
+import * as os from 'node:os'
+import * as path from 'node:path'
+import { acquireFileLock, acquireFileLocksOrdered } from '../../src/git/file-lock.js'
+
+describe('file-lock', () => {
+  let tmpDir: string
+
+  beforeEach(async () => {
+    tmpDir = await fs.mkdtemp(path.join(os.tmpdir(), 'file-lock-'))
+  })
+
+  afterEach(async () => {
+    await fs.rm(tmpDir, { recursive: true, force: true })
+  })
+
+  it('acquires and releases a lock; lockfile is gone after release', async () => {
+    const lockPath = path.join(tmpDir, 'a.lock')
+    const release = await acquireFileLock(lockPath)
+    // proper-lockfile creates a directory at <lockPath>.lock for the actual lock
+    const stat = await fs.stat(`${lockPath}.lock`).catch(() => null)
+    expect(stat).not.toBeNull()
+    await release()
+    // After release, the .lock dir should be gone
+    const after = await fs.stat(`${lockPath}.lock`).catch(() => null)
+    expect(after).toBeNull()
+  })
+
+  it('release is idempotent (second call is no-op)', async () => {
+    const lockPath = path.join(tmpDir, 'b.lock')
+    const release = await acquireFileLock(lockPath)
+    await release()
+    await expect(release()).resolves.toBeUndefined()
+  })
+
+  it('second acquire blocks until first release', async () => {
+    const lockPath = path.join(tmpDir, 'c.lock')
+    const release1 = await acquireFileLock(lockPath)
+
+    let secondAcquired = false
+    const second = acquireFileLock(lockPath).then((r) => {
+      secondAcquired = true
+      return r
+    })
+
+    // Give the second acquire a moment to attempt
+    await new Promise((r) => setTimeout(r, 200))
+    expect(secondAcquired).toBe(false)
+
+    await release1()
+    const release2 = await second
+    expect(secondAcquired).toBe(true)
+    await release2()
+  }, 10_000)
+
+  it('acquireFileLocksOrdered sorts paths alphabetically (deadlock-free for crossed sets)', async () => {
+    const a = path.join(tmpDir, 'A.lock')
+    const b = path.join(tmpDir, 'B.lock')
+
+    // Two concurrent multi-locks with crossed orders both sort to [A, B]
+    const r1Promise = acquireFileLocksOrdered([b, a])
+
+    // First should grab both since paths sort the same
+    const r1 = await r1Promise
+
+    let secondAcquired = false
+    const r2Promise = acquireFileLocksOrdered([a, b]).then((r) => {
+      secondAcquired = true
+      return r
+    })
+
+    await new Promise((r) => setTimeout(r, 200))
+    expect(secondAcquired).toBe(false)
+
+    await r1()
+    const r2 = await r2Promise
+    expect(secondAcquired).toBe(true)
+    await r2()
+  }, 15_000)
+
+  it('partial failure releases held locks', async () => {
+    // Force the second acquire to fail by writing a regular file at the lockfile
+    // location proper-lockfile wants to create as a directory.
+    const a = path.join(tmpDir, 'A.lock')
+    const bPath = path.join(tmpDir, 'B.lock')
+    // Create a regular file at `${bPath}.lock` so proper-lockfile's mkdir fails with EEXIST
+    await fs.writeFile(`${bPath}.lock`, 'blocked')
+
+    await expect(acquireFileLocksOrdered([a, bPath])).rejects.toThrow()
+
+    // After failure, A's lock should be released — re-acquire immediately
+    const r = await acquireFileLock(a)
+    await r()
+  }, 90_000)
+})
--- a/tests/git/job-locks.test.ts
+++ b/tests/git/job-locks.test.ts
@ -0,0 +1,121 @@
+import { describe, it, expect, beforeEach, afterEach, vi } from 'vitest'
+import * as fs from 'node:fs/promises'
+import * as os from 'node:os'
+import * as path from 'node:path'
+import { execFile } from 'node:child_process'
+import { promisify } from 'node:util'
+import {
+  registerJobLockReleases,
+  releaseLocksOnTerminal,
+  setupProductWorktrees,
+  _resetJobReleasesForTest,
+} from '../../src/git/job-locks.js'
+
+const exec = promisify(execFile)
+
+describe('job-locks: registerJobLockReleases + releaseLocksOnTerminal', () => {
+  beforeEach(() => _resetJobReleasesForTest())
+
+  it('releaseLocksOnTerminal for unknown job is a no-op', async () => {
+    await expect(releaseLocksOnTerminal('nonexistent')).resolves.toBeUndefined()
+  })
+
+  it('runs registered releases and clears the entry', async () => {
+    const release = vi.fn().mockResolvedValue(undefined)
+    registerJobLockReleases('job-1', [release])
+    await releaseLocksOnTerminal('job-1')
+    expect(release).toHaveBeenCalledTimes(1)
+    // Second call → no-op (cleared)
+    await releaseLocksOnTerminal('job-1')
+    expect(release).toHaveBeenCalledTimes(1)
+  })
+
+  it('failures in one release do not abort others', async () => {
+    const r1 = vi.fn().mockRejectedValue(new Error('boom'))
+    const r2 = vi.fn().mockResolvedValue(undefined)
+    registerJobLockReleases('job-2', [r1, r2])
+    await expect(releaseLocksOnTerminal('job-2')).resolves.toBeUndefined()
+    expect(r1).toHaveBeenCalled()
+    expect(r2).toHaveBeenCalled()
+  })
+
+  it('append-mode: multiple registers accumulate', async () => {
+    const r1 = vi.fn().mockResolvedValue(undefined)
+    const r2 = vi.fn().mockResolvedValue(undefined)
+    registerJobLockReleases('job-3', [r1])
+    registerJobLockReleases('job-3', [r2])
+    await releaseLocksOnTerminal('job-3')
+    expect(r1).toHaveBeenCalledTimes(1)
+    expect(r2).toHaveBeenCalledTimes(1)
+  })
+})
+
+describe('job-locks: setupProductWorktrees', () => {
+  let tmpRoot: string
+  let originalEnv: string | undefined
+  let bareRepo: string
+  let originRepo: string
+
+  beforeEach(async () => {
+    _resetJobReleasesForTest()
+    tmpRoot = await fs.mkdtemp(path.join(os.tmpdir(), 'job-locks-'))
+    originalEnv = process.env.SCRUM4ME_AGENT_WORKTREE_DIR
+    process.env.SCRUM4ME_AGENT_WORKTREE_DIR = path.join(tmpRoot, 'agent-worktrees')
+
+    // Set up a bare repo as origin and a clone with origin/main
+    bareRepo = path.join(tmpRoot, 'origin.git')
+    await exec('git', ['init', '--bare', '-b', 'main', bareRepo])
+
+    originRepo = path.join(tmpRoot, 'work')
+    await exec('git', ['init', '-b', 'main', originRepo])
+    await exec('git', ['config', 'user.email', 't@t.local'], { cwd: originRepo })
+    await exec('git', ['config', 'user.name', 'Test'], { cwd: originRepo })
+    await exec('git', ['remote', 'add', 'origin', bareRepo], { cwd: originRepo })
+    await fs.writeFile(path.join(originRepo, 'README.md'), '# init\n')
+    await exec('git', ['add', '-A'], { cwd: originRepo })
+    await exec('git', ['commit', '-m', 'init'], { cwd: originRepo })
+    await exec('git', ['push', '-u', 'origin', 'main'], { cwd: originRepo })
+  })
+
+  afterEach(async () => {
+    if (originalEnv) process.env.SCRUM4ME_AGENT_WORKTREE_DIR = originalEnv
+    else delete process.env.SCRUM4ME_AGENT_WORKTREE_DIR
+    await fs.rm(tmpRoot, { recursive: true, force: true })
+  })
+
+  it('returns empty when productIds is empty', async () => {
+    const result = await setupProductWorktrees('j1', [], async () => null)
+    expect(result).toEqual([])
+  })
+
+  it('creates a product-worktree, registers a lock-release, and releases it', async () => {
+    const result = await setupProductWorktrees('j2', ['prod-a'], async () => originRepo)
+    expect(result).toHaveLength(1)
+    expect(result[0].productId).toBe('prod-a')
+    expect(result[0].worktreePath).toContain('_products/prod-a')
+
+    // Worktree dir exists with detached HEAD on origin/main
+    const stat = await fs.stat(result[0].worktreePath)
+    expect(stat.isDirectory()).toBe(true)
+
+    // Lockfile is held during the job (proper-lockfile creates a .lock dir)
+    const lockDir = path.join(
+      process.env.SCRUM4ME_AGENT_WORKTREE_DIR!,
+      '_products',
+      'prod-a.lock.lock',
+    )
+    const lockStat = await fs.stat(lockDir).catch(() => null)
+    expect(lockStat).not.toBeNull()
+
+    await releaseLocksOnTerminal('j2')
+    const lockAfter = await fs.stat(lockDir).catch(() => null)
+    expect(lockAfter).toBeNull()
+  })
+
+  it('skips products where resolveRepoRoot returns null', async () => {
+    const result = await setupProductWorktrees('j3', ['no-repo'], async () => null)
+    expect(result).toEqual([])
+    // Lock was still acquired and registered — release cleans up
+    await releaseLocksOnTerminal('j3')
+  })
+})
--- a/tests/git/pr-enable-auto-merge.test.ts
+++ b/tests/git/pr-enable-auto-merge.test.ts
@ -0,0 +1,75 @@
+import { describe, it, expect, vi, beforeEach } from 'vitest'
+
+// Mock node:child_process before importing the module under test
+vi.mock('node:child_process', () => ({
+  execFile: vi.fn(),
+}))
+
+import { execFile } from 'node:child_process'
+import { enableAutoMergeOnPr } from '../../src/git/pr.js'
+
+const mockExecFile = vi.mocked(execFile) as unknown as ReturnType<typeof vi.fn>
+
+function mockGhFailure(stderr: string) {
+  mockExecFile.mockImplementation(((_cmd: string, _args: string[], _opts: unknown, cb: any) => {
+    cb(Object.assign(new Error('gh exit'), { stderr }))
+  }) as never)
+}
+
+function mockGhSuccess() {
+  mockExecFile.mockImplementation(((_cmd: string, _args: string[], _opts: unknown, cb: any) => {
+    cb(null, { stdout: '', stderr: '' })
+  }) as never)
+}
+
+describe('enableAutoMergeOnPr — typed errors (PBI-47 C2 layer 1)', () => {
+  beforeEach(() => {
+    mockExecFile.mockReset()
+  })
+
+  it('returns ok:true on green merge', async () => {
+    mockGhSuccess()
+    const result = await enableAutoMergeOnPr({ prUrl: 'x', expectedHeadSha: 'sha1' })
+    expect(result.ok).toBe(true)
+  })
+
+  it('classifies GH_AUTH_ERROR for 401/403 / permission strings', async () => {
+    mockGhFailure('gh: HTTP 403: permission denied')
+    const result = await enableAutoMergeOnPr({ prUrl: 'x', expectedHeadSha: 'sha1' })
+    expect(result.ok).toBe(false)
+    if (!result.ok) expect(result.reason).toBe('GH_AUTH_ERROR')
+  })
+
+  it('classifies AUTO_MERGE_NOT_ALLOWED for repo-setting refusal', async () => {
+    mockGhFailure('auto-merge is not allowed for this repository')
+    const result = await enableAutoMergeOnPr({ prUrl: 'x', expectedHeadSha: 'sha1' })
+    expect(result.ok).toBe(false)
+    if (!result.ok) expect(result.reason).toBe('AUTO_MERGE_NOT_ALLOWED')
+  })
+
+  it('classifies MERGE_CONFLICT for dirty merge state', async () => {
+    mockGhFailure('pull request is not in a mergeable state (dirty)')
+    const result = await enableAutoMergeOnPr({ prUrl: 'x', expectedHeadSha: 'sha1' })
+    expect(result.ok).toBe(false)
+    if (!result.ok) expect(result.reason).toBe('MERGE_CONFLICT')
+  })
+
+  it('classifies UNKNOWN for unrecognised stderr', async () => {
+    mockGhFailure('unexpected gh error')
+    const result = await enableAutoMergeOnPr({ prUrl: 'x', expectedHeadSha: 'sha1' })
+    expect(result.ok).toBe(false)
+    if (!result.ok) expect(result.reason).toBe('UNKNOWN')
+  })
+
+  it('passes --match-head-commit when expectedHeadSha provided', async () => {
+    mockGhSuccess()
+    await enableAutoMergeOnPr({ prUrl: 'pr-url', expectedHeadSha: 'abc123' })
+    const callArgs = mockExecFile.mock.calls[0]
+    expect(callArgs[0]).toBe('gh')
+    const args = callArgs[1] as string[]
+    expect(args).toContain('--match-head-commit')
+    expect(args).toContain('abc123')
+    expect(args).toContain('--auto')
+    expect(args).toContain('--squash')
+  })
+})
--- a/tests/verify/classify-delete.test.ts
+++ b/tests/verify/classify-delete.test.ts
@ -0,0 +1,59 @@
+import { describe, it, expect } from 'vitest'
+import { classifyDiffAgainstPlan } from '../../src/verify/classify.js'
+
+describe('classify — delete-only commits (PBI-47 C5)', () => {
+  it('returns ALIGNED when the deleted path is in the plan', () => {
+    const diff = `diff --git a/app/todos/page.tsx b/app/todos/page.tsx
+deleted file mode 100644
+index 1234567..0000000
+--- a/app/todos/page.tsx
+++ /dev/null
+@@ -1,3 +0,0 @@
+-export default function TodosPage() {
+-  return null
+-}`
+
+    const plan = '- Verwijder `app/todos/page.tsx`\n- Verwijder gerelateerde imports'
+
+    const result = classifyDiffAgainstPlan({ diff, plan })
+    expect(result.result).toBe('ALIGNED')
+  })
+
+  it('returns ALIGNED for multi-file delete-only when both paths in plan', () => {
+    const diff = `diff --git a/app/todos/page.tsx b/app/todos/page.tsx
+deleted file mode 100644
+--- a/app/todos/page.tsx
+++ /dev/null
+@@ -1,2 +0,0 @@
+-line 1
+-line 2
+diff --git a/components/todo-list.tsx b/components/todo-list.tsx
+deleted file mode 100644
+--- a/components/todo-list.tsx
+++ /dev/null
+@@ -1,1 +0,0 @@
+-line`
+
+    const plan = '- `app/todos/page.tsx`\n- `components/todo-list.tsx`'
+    const result = classifyDiffAgainstPlan({ diff, plan })
+    expect(result.result).toBe('ALIGNED')
+  })
+
+  it('returns PARTIAL when only some plan deletes appear in the diff', () => {
+    const diff = `diff --git a/a.ts b/a.ts
+deleted file mode 100644
+--- a/a.ts
+++ /dev/null
+@@ -1,1 +0,0 @@
+-x`
+
+    const plan = '- `a.ts`\n- `b.ts`' // b.ts missing
+    const result = classifyDiffAgainstPlan({ diff, plan })
+    expect(result.result).toBe('PARTIAL')
+  })
+
+  it('returns EMPTY for a no-op diff', () => {
+    const result = classifyDiffAgainstPlan({ diff: '', plan: 'irrelevant' })
+    expect(result.result).toBe('EMPTY')
+  })
+})
--- a/tests/verify/verify-scope.test.ts
+++ b/tests/verify/verify-scope.test.ts
@ -0,0 +1,55 @@
+import { describe, it, expect, beforeAll, afterAll } from 'vitest'
+import * as fs from 'node:fs/promises'
+import * as path from 'node:path'
+import * as os from 'node:os'
+import { execFile } from 'node:child_process'
+import { promisify } from 'node:util'
+import { getDiffInWorktree } from '../../src/tools/verify-task-against-plan.js'
+
+const exec = promisify(execFile)
+
+describe('verify scope per-job (PBI-47 P0)', () => {
+  let tmpRepo: string
+  let baseSha: string
+  let task1Sha: string
+
+  beforeAll(async () => {
+    tmpRepo = await fs.mkdtemp(path.join(os.tmpdir(), 'verify-scope-'))
+    await exec('git', ['init', '-b', 'main'], { cwd: tmpRepo })
+    await exec('git', ['config', 'user.email', 't@t.local'], { cwd: tmpRepo })
+    await exec('git', ['config', 'user.name', 'Test'], { cwd: tmpRepo })
+    await fs.writeFile(path.join(tmpRepo, 'README.md'), '# init\n')
+    await exec('git', ['add', '-A'], { cwd: tmpRepo })
+    await exec('git', ['commit', '-m', 'init'], { cwd: tmpRepo })
+    const baseRev = await exec('git', ['rev-parse', 'HEAD'], { cwd: tmpRepo })
+    baseSha = baseRev.stdout.trim()
+
+    // Simulate task 1: add a.ts
+    await fs.writeFile(path.join(tmpRepo, 'a.ts'), 'task 1\n')
+    await exec('git', ['add', '-A'], { cwd: tmpRepo })
+    await exec('git', ['commit', '-m', 'task 1'], { cwd: tmpRepo })
+    const t1Rev = await exec('git', ['rev-parse', 'HEAD'], { cwd: tmpRepo })
+    task1Sha = t1Rev.stdout.trim()
+
+    // Simulate task 2: add b.ts
+    await fs.writeFile(path.join(tmpRepo, 'b.ts'), 'task 2\n')
+    await exec('git', ['add', '-A'], { cwd: tmpRepo })
+    await exec('git', ['commit', '-m', 'task 2'], { cwd: tmpRepo })
+  })
+
+  afterAll(async () => {
+    await fs.rm(tmpRepo, { recursive: true, force: true })
+  })
+
+  it('diff vs base = origin/main → both task 1 and task 2 visible', async () => {
+    const diff = await getDiffInWorktree(tmpRepo, baseSha)
+    expect(diff).toContain('a.ts')
+    expect(diff).toContain('b.ts')
+  })
+
+  it('diff vs base = task1_sha → only task 2 visible', async () => {
+    const diff = await getDiffInWorktree(tmpRepo, task1Sha)
+    expect(diff).not.toContain('a.ts')
+    expect(diff).toContain('b.ts')
+  })
+})
--- a/package-lock.json
+++ b/package-lock.json
@ -13,7 +13,9 @@
        "@modelcontextprotocol/sdk": "^1.29.0",
        "@prisma/adapter-pg": "^7.8.0",
        "@prisma/client": "^7.8.0",
+        "@types/proper-lockfile": "^4.1.4",
        "pg": "^8.13.1",
+        "proper-lockfile": "^4.1.2",
        "yaml": "^2.8.4",
        "zod": "^4.0.0"
      },
@ -1327,6 +1329,15 @@
        "pg-types": "^2.2.0"
      }
    },
+    "node_modules/@types/proper-lockfile": {
+      "version": "4.1.4",
+      "resolved": "https://registry.npmjs.org/@types/proper-lockfile/-/proper-lockfile-4.1.4.tgz",
+      "integrity": "sha512-uo2ABllncSqg9F1D4nugVl9v93RmjxF6LJzQLMLDdPaXCUIDPeOJ21Gbqi43xNKzBi/WQ0Q0dICqufzQbMjipQ==",
+      "license": "MIT",
+      "dependencies": {
+        "@types/retry": "*"
+      }
+    },
    "node_modules/@types/react": {
      "version": "19.2.14",
      "resolved": "https://registry.npmjs.org/@types/react/-/react-19.2.14.tgz",
@ -1338,6 +1349,12 @@
        "csstype": "^3.2.2"
      }
    },
+    "node_modules/@types/retry": {
+      "version": "0.12.5",
+      "resolved": "https://registry.npmjs.org/@types/retry/-/retry-0.12.5.tgz",
+      "integrity": "sha512-3xSjTp3v03X/lSQLkczaN9UIEwJMoMCA1+Nb5HfbJEQWogdeQIyVtTvxPXDQjZ5zws8rFQfVfRdz03ARihPJgw==",
+      "license": "MIT"
+    },
    "node_modules/@vitest/expect": {
      "version": "4.1.5",
      "resolved": "https://registry.npmjs.org/@vitest/expect/-/expect-4.1.5.tgz",
@ -2332,7 +2349,6 @@
      "version": "4.2.11",
      "resolved": "https://registry.npmjs.org/graceful-fs/-/graceful-fs-4.2.11.tgz",
      "integrity": "sha512-RbJ5/jmFcNNCcDV5o9eTnBLJ/HszWV0P73bc+Ff4nS/rJj+YaS6IGyiOL0VoBYX+l1Wrl3k63h/KrH+nhJ0XvQ==",
-      "devOptional": true,
      "license": "ISC"
    },
    "node_modules/grammex": {
@ -3277,7 +3293,6 @@
      "version": "4.1.2",
      "resolved": "https://registry.npmjs.org/proper-lockfile/-/proper-lockfile-4.1.2.tgz",
      "integrity": "sha512-TjNPblN4BwAWMXU8s9AEz4JmQxnD1NNL7bNOY/AKUzyamc379FWASUhc/K1pL2noVb+XmZKLL68cjzLsiOAMaA==",
-      "devOptional": true,
      "license": "MIT",
      "dependencies": {
        "graceful-fs": "^4.2.4",
@ -3289,7 +3304,6 @@
      "version": "3.0.7",
      "resolved": "https://registry.npmjs.org/signal-exit/-/signal-exit-3.0.7.tgz",
      "integrity": "sha512-wnD2ZE+l+SPC/uoS0vXeE9L1+0wuaMqKlfz9AMUo38JsyLSBWSFcHR1Rri62LZc12vLr1gb3jl7iwQhgwpAbGQ==",
-      "devOptional": true,
      "license": "ISC"
    },
    "node_modules/proxy-addr": {
@ -3444,7 +3458,6 @@
      "version": "0.12.0",
      "resolved": "https://registry.npmjs.org/retry/-/retry-0.12.0.tgz",
      "integrity": "sha512-9LkiTwjUh6rT555DtE9rTX+BKByPfrMzEAtnlEtdEwr3Nkffwiihqe2bWADg+OQRjt9gl6ICdmB/ZFDCGAtSow==",
-      "devOptional": true,
      "license": "MIT",
      "engines": {
        "node": ">= 4"
--- a/package.json
+++ b/package.json
@ -32,7 +32,9 @@
    "@modelcontextprotocol/sdk": "^1.29.0",
    "@prisma/adapter-pg": "^7.8.0",
    "@prisma/client": "^7.8.0",
+    "@types/proper-lockfile": "^4.1.4",
    "pg": "^8.13.1",
+    "proper-lockfile": "^4.1.2",
    "yaml": "^2.8.4",
    "zod": "^4.0.0"
  },
--- a/prisma/schema.prisma
+++ b/prisma/schema.prisma
@ -2,7 +2,6 @@ generator client {
  provider = "prisma-client-js"
 }

-
 datasource db {
  provider = "postgresql"
 }
@ -314,6 +313,7 @@ model SprintRun {
  failure_reason String?
  failed_task    Task?           @relation("SprintRunFailedTask", fields: [failed_task_id], references: [id], onDelete: SetNull)
  failed_task_id String?
+  pause_context  Json?
  created_at     DateTime        @default(now())
  updated_at     DateTime        @updatedAt
  jobs           ClaudeJob[]
@ -385,6 +385,8 @@ model ClaudeJob {
  cache_read_tokens   Int?
  cache_write_tokens  Int?
  plan_snapshot       String?
+  base_sha            String?
+  head_sha            String?
  branch              String?
  pr_url              String?
  summary             String?
--- a/src/cancel/pbi-cascade.ts
+++ b/src/cancel/pbi-cascade.ts
@ -13,6 +13,7 @@ import {
  getPullRequestState,
 } from '../git/pr.js'
 import { deleteRemoteBranch } from '../git/push.js'
+import { releaseLocksOnTerminal } from '../git/job-locks.js'

 export type CascadeOutcome = {
  cancelled_job_ids: string[]
@ -88,6 +89,9 @@ async function runCascade(failedJobId: string): Promise<CascadeOutcome> {
        error: 'cancelled_by_pbi_failure',
      },
    })
+    // PBI-9: release product-worktree locks for cancelled jobs.
+    // No-op for jobs without registered locks (TASK_IMPLEMENTATION).
+    for (const j of eligible) await releaseLocksOnTerminal(j.id)
  }

  const outcome: CascadeOutcome = {
--- a/src/flow/effects.ts
+++ b/src/flow/effects.ts
@ -0,0 +1,192 @@
+// PBI-9 + PBI-47: declarative effects produced by pure transitions.
+// Executor handles each effect idempotently; failures are logged, not thrown.
+
+export type PauseContext = {
+  pause_reason: 'MERGE_CONFLICT'
+  pr_url: string
+  pr_head_sha: string
+  conflict_files: string[]
+  claude_question_id: string
+  resume_instructions: string
+  paused_at: string
+}
+
+export type FlowEffect =
+  | { type: 'RELEASE_WORKTREE_LOCKS'; jobId: string }
+  | { type: 'ENABLE_AUTO_MERGE'; prUrl: string; expectedHeadSha: string }
+  | { type: 'MARK_PR_READY'; prUrl: string }
+  | {
+      type: 'CREATE_CLAUDE_QUESTION'
+      sprintRunId: string
+      prUrl: string
+      files: string[]
+    }
+  | { type: 'CLOSE_CLAUDE_QUESTION'; questionId: string }
+  | {
+      type: 'SET_SPRINT_RUN_STATUS'
+      sprintRunId: string
+      status: 'QUEUED' | 'RUNNING' | 'PAUSED' | 'DONE' | 'FAILED' | 'CANCELLED'
+      pauseContextDraft?: Omit<PauseContext, 'claude_question_id'>
+      clearPauseContext?: boolean
+    }
+
+export type AutoMergeOutcome =
+  | { effect: 'ENABLE_AUTO_MERGE'; ok: true }
+  | {
+      effect: 'ENABLE_AUTO_MERGE'
+      ok: false
+      reason: 'CHECKS_FAILED' | 'MERGE_CONFLICT' | 'GH_AUTH_ERROR' | 'AUTO_MERGE_NOT_ALLOWED' | 'UNKNOWN'
+      stderr: string
+    }
+
+/**
+ * Execute a list of effects in order. Returns outcome objects only for
+ * effects whose result the caller needs to react to (auto-merge fail
+ * triggers MERGE_CONFLICT-event in update-job-status). Other failures
+ * are logged but swallowed.
+ *
+ * CREATE_CLAUDE_QUESTION → SET_SPRINT_RUN_STATUS chains: the question_id
+ * created in the first effect is injected into the pause_context of the
+ * second.
+ */
+export async function executeEffects(
+  effects: FlowEffect[],
+): Promise<AutoMergeOutcome[]> {
+  const outcomes: AutoMergeOutcome[] = []
+  let lastQuestionId: string | undefined
+  for (const effect of effects) {
+    try {
+      if (effect.type === 'CREATE_CLAUDE_QUESTION') {
+        lastQuestionId = await createOrReuseClaudeQuestion(effect)
+        continue
+      }
+      if (effect.type === 'SET_SPRINT_RUN_STATUS') {
+        await applySprintRunStatus(effect, lastQuestionId)
+        continue
+      }
+      const outcome = await executeEffect(effect)
+      if (outcome) outcomes.push(outcome)
+    } catch (err) {
+      console.warn(`[effects] effect ${effect.type} failed (idempotent skip):`, err)
+    }
+  }
+  return outcomes
+}
+
+async function executeEffect(effect: FlowEffect): Promise<AutoMergeOutcome | undefined> {
+  switch (effect.type) {
+    case 'RELEASE_WORKTREE_LOCKS': {
+      const { releaseLocksOnTerminal } = await import('../git/job-locks.js')
+      await releaseLocksOnTerminal(effect.jobId)
+      return undefined
+    }
+    case 'ENABLE_AUTO_MERGE': {
+      const { enableAutoMergeOnPr } = await import('../git/pr.js')
+      const result = await enableAutoMergeOnPr({
+        prUrl: effect.prUrl,
+        expectedHeadSha: effect.expectedHeadSha,
+      })
+      if (result.ok) return { effect: 'ENABLE_AUTO_MERGE', ok: true }
+      return { effect: 'ENABLE_AUTO_MERGE', ok: false, reason: result.reason, stderr: result.stderr }
+    }
+    case 'MARK_PR_READY': {
+      const { markPullRequestReady } = await import('../git/pr.js')
+      const result = await markPullRequestReady({ prUrl: effect.prUrl })
+      if ('error' in result) {
+        console.warn(`[effects] MARK_PR_READY failed for ${effect.prUrl}: ${result.error}`)
+      }
+      return undefined
+    }
+    case 'CLOSE_CLAUDE_QUESTION': {
+      const { prisma } = await import('../prisma.js')
+      await prisma.claudeQuestion.updateMany({
+        where: { id: effect.questionId, status: 'open' },
+        data: { status: 'closed' },
+      })
+      return undefined
+    }
+    // CREATE_CLAUDE_QUESTION + SET_SPRINT_RUN_STATUS handled in executeEffects.
+    case 'CREATE_CLAUDE_QUESTION':
+    case 'SET_SPRINT_RUN_STATUS':
+      return undefined
+  }
+}
+
+async function createOrReuseClaudeQuestion(effect: {
+  sprintRunId: string
+  prUrl: string
+  files: string[]
+}): Promise<string> {
+  const { prisma } = await import('../prisma.js')
+
+  // Reuse existing open question for the same SprintRun + PR if present.
+  const existing = await prisma.claudeQuestion.findFirst({
+    where: {
+      status: 'open',
+      options: { path: ['sprint_run_id'], equals: effect.sprintRunId } as never,
+    },
+    orderBy: { created_at: 'desc' },
+    select: { id: true },
+  })
+  if (existing) return existing.id
+
+  // Need product_id + asker (user) to create. Resolve via SprintRun.
+  const sprintRun = await prisma.sprintRun.findUnique({
+    where: { id: effect.sprintRunId },
+    select: {
+      started_by_id: true,
+      sprint: { select: { product_id: true } },
+    },
+  })
+  if (!sprintRun) {
+    throw new Error(`SprintRun ${effect.sprintRunId} not found`)
+  }
+
+  const fileList =
+    effect.files.length === 0
+      ? '(unknown files — check the PR)'
+      : effect.files.slice(0, 5).join(', ')
+        + (effect.files.length > 5 ? ` + ${effect.files.length - 5} more` : '')
+
+  const created = await prisma.claudeQuestion.create({
+    data: {
+      product_id: sprintRun.sprint.product_id,
+      asked_by: sprintRun.started_by_id,
+      question:
+        `Merge-conflict on ${effect.prUrl}. Conflict files: ${fileList}. `
+        + `Resolve on the branch and push, then resume the sprint.`,
+      options: {
+        sprint_run_id: effect.sprintRunId,
+        pr_url: effect.prUrl,
+        conflict_files: effect.files,
+      },
+      status: 'open',
+      expires_at: new Date(Date.now() + 7 * 24 * 60 * 60 * 1000), // 7 days
+    },
+    select: { id: true },
+  })
+  return created.id
+}
+
+async function applySprintRunStatus(
+  effect: Extract<FlowEffect, { type: 'SET_SPRINT_RUN_STATUS' }>,
+  lastQuestionId: string | undefined,
+): Promise<void> {
+  const { prisma, Prisma } = await (async () => {
+    const mod = await import('../prisma.js')
+    const prismaPkg = await import('@prisma/client')
+    return { prisma: mod.prisma, Prisma: prismaPkg.Prisma }
+  })()
+
+  const data: Record<string, unknown> = { status: effect.status }
+  if (effect.pauseContextDraft && lastQuestionId) {
+    data.pause_context = {
+      ...effect.pauseContextDraft,
+      claude_question_id: lastQuestionId,
+    }
+  }
+  if (effect.clearPauseContext) {
+    data.pause_context = Prisma.JsonNull
+  }
+  await prisma.sprintRun.update({ where: { id: effect.sprintRunId }, data })
+}
--- a/src/flow/pr-flow.ts
+++ b/src/flow/pr-flow.ts
@ -0,0 +1,110 @@
+import type { FlowEffect } from './effects.js'
+import type { AutoMergeFailReason } from '../git/pr.js'
+
+export type PrStrategy = 'STORY' | 'SPRINT'
+
+export type PrFlowState =
+  | { kind: 'none'; strategy: PrStrategy }
+  | { kind: 'branch_pushed'; strategy: PrStrategy; prUrl?: string }
+  | { kind: 'pr_opened'; strategy: 'STORY'; prUrl: string }
+  | { kind: 'draft_opened'; strategy: 'SPRINT'; prUrl: string }
+  | { kind: 'waiting_for_checks'; strategy: 'STORY'; prUrl: string; headSha: string }
+  | { kind: 'auto_merge_enabled'; strategy: 'STORY'; prUrl: string; headSha: string }
+  | { kind: 'ready_for_review'; strategy: 'SPRINT'; prUrl: string }
+  | { kind: 'merged'; strategy: PrStrategy; prUrl: string }
+  | { kind: 'checks_failed'; strategy: PrStrategy; prUrl: string }
+  | { kind: 'merge_conflict_paused'; strategy: PrStrategy; prUrl: string; headSha: string }
+
+export type PrFlowEvent =
+  | { type: 'PR_CREATED'; prUrl: string }
+  | { type: 'TASK_DONE'; taskId: string; headSha: string }
+  | { type: 'STORY_COMPLETED'; storyId: string; headSha: string }
+  | { type: 'SPRINT_COMPLETED'; sprintRunId: string }
+  | { type: 'MERGE_RESULT'; reason?: AutoMergeFailReason }
+
+export type TransitionResult = { nextState: PrFlowState; effects: FlowEffect[] }
+
+export function transition(state: PrFlowState, event: PrFlowEvent): TransitionResult {
+  if (state.strategy === 'STORY') {
+    switch (state.kind) {
+      case 'none':
+      case 'branch_pushed':
+        if (event.type === 'PR_CREATED') {
+          return {
+            nextState: { kind: 'pr_opened', strategy: 'STORY', prUrl: event.prUrl },
+            effects: [],
+          }
+        }
+        break
+      case 'pr_opened':
+        if (event.type === 'STORY_COMPLETED') {
+          return {
+            nextState: {
+              kind: 'waiting_for_checks',
+              strategy: 'STORY',
+              prUrl: state.prUrl,
+              headSha: event.headSha,
+            },
+            effects: [
+              { type: 'ENABLE_AUTO_MERGE', prUrl: state.prUrl, expectedHeadSha: event.headSha },
+            ],
+          }
+        }
+        break
+      case 'waiting_for_checks':
+        if (event.type === 'MERGE_RESULT' && !event.reason) {
+          return {
+            nextState: {
+              kind: 'auto_merge_enabled',
+              strategy: 'STORY',
+              prUrl: state.prUrl,
+              headSha: state.headSha,
+            },
+            effects: [],
+          }
+        }
+        if (event.type === 'MERGE_RESULT' && event.reason === 'MERGE_CONFLICT') {
+          return {
+            nextState: {
+              kind: 'merge_conflict_paused',
+              strategy: 'STORY',
+              prUrl: state.prUrl,
+              headSha: state.headSha,
+            },
+            effects: [],
+          }
+        }
+        if (event.type === 'MERGE_RESULT' && event.reason === 'CHECKS_FAILED') {
+          return {
+            nextState: { kind: 'checks_failed', strategy: 'STORY', prUrl: state.prUrl },
+            effects: [],
+          }
+        }
+        break
+    }
+  }
+
+  if (state.strategy === 'SPRINT') {
+    switch (state.kind) {
+      case 'none':
+      case 'branch_pushed':
+        if (event.type === 'PR_CREATED') {
+          return {
+            nextState: { kind: 'draft_opened', strategy: 'SPRINT', prUrl: event.prUrl },
+            effects: [],
+          }
+        }
+        break
+      case 'draft_opened':
+        if (event.type === 'SPRINT_COMPLETED') {
+          return {
+            nextState: { kind: 'ready_for_review', strategy: 'SPRINT', prUrl: state.prUrl },
+            effects: [{ type: 'MARK_PR_READY', prUrl: state.prUrl }],
+          }
+        }
+        break
+    }
+  }
+
+  return { nextState: state, effects: [] }
+}
--- a/src/flow/sprint-run.ts
+++ b/src/flow/sprint-run.ts
@ -0,0 +1,136 @@
+import type { FlowEffect, PauseContext } from './effects.js'
+
+export type SprintRunStateKind =
+  | 'queued'
+  | 'running'
+  | 'paused_merge_conflict'
+  | 'done'
+  | 'failed'
+  | 'cancelled'
+
+export type SprintRunState = {
+  kind: SprintRunStateKind
+  sprintRunId: string
+  pauseContext?: PauseContext
+}
+
+export type SprintRunEvent =
+  | { type: 'CLAIM_FIRST_JOB' }
+  | { type: 'TASK_DONE'; taskId: string }
+  | { type: 'TASK_FAILED'; taskId: string; error: string }
+  | {
+      type: 'MERGE_CONFLICT'
+      prUrl: string
+      prHeadSha: string
+      conflictFiles: string[]
+      resumeInstructions: string
+    }
+  | { type: 'USER_RESUMED' }
+  | { type: 'USER_CANCELLED' }
+  | { type: 'ALL_DONE' }
+
+export type TransitionResult = { nextState: SprintRunState; effects: FlowEffect[] }
+
+export function transition(state: SprintRunState, event: SprintRunEvent): TransitionResult {
+  switch (state.kind) {
+    case 'queued':
+      if (event.type === 'CLAIM_FIRST_JOB') {
+        return {
+          nextState: { ...state, kind: 'running' },
+          effects: [
+            { type: 'SET_SPRINT_RUN_STATUS', sprintRunId: state.sprintRunId, status: 'RUNNING' },
+          ],
+        }
+      }
+      break
+    case 'running':
+      if (event.type === 'TASK_DONE') {
+        return { nextState: state, effects: [] }
+      }
+      if (event.type === 'TASK_FAILED') {
+        return {
+          nextState: { ...state, kind: 'failed' },
+          effects: [
+            { type: 'SET_SPRINT_RUN_STATUS', sprintRunId: state.sprintRunId, status: 'FAILED' },
+          ],
+        }
+      }
+      if (event.type === 'ALL_DONE') {
+        return {
+          nextState: { ...state, kind: 'done' },
+          effects: [
+            { type: 'SET_SPRINT_RUN_STATUS', sprintRunId: state.sprintRunId, status: 'DONE' },
+          ],
+        }
+      }
+      if (event.type === 'MERGE_CONFLICT') {
+        const pauseContextDraft: Omit<PauseContext, 'claude_question_id'> = {
+          pause_reason: 'MERGE_CONFLICT',
+          pr_url: event.prUrl,
+          pr_head_sha: event.prHeadSha,
+          conflict_files: event.conflictFiles,
+          resume_instructions: event.resumeInstructions,
+          paused_at: new Date().toISOString(),
+        }
+        return {
+          nextState: { ...state, kind: 'paused_merge_conflict' },
+          effects: [
+            {
+              type: 'CREATE_CLAUDE_QUESTION',
+              sprintRunId: state.sprintRunId,
+              prUrl: event.prUrl,
+              files: event.conflictFiles,
+            },
+            {
+              type: 'SET_SPRINT_RUN_STATUS',
+              sprintRunId: state.sprintRunId,
+              status: 'PAUSED',
+              pauseContextDraft,
+            },
+          ],
+        }
+      }
+      if (event.type === 'USER_CANCELLED') {
+        return {
+          nextState: { ...state, kind: 'cancelled' },
+          effects: [
+            { type: 'SET_SPRINT_RUN_STATUS', sprintRunId: state.sprintRunId, status: 'CANCELLED' },
+          ],
+        }
+      }
+      break
+    case 'paused_merge_conflict':
+      if (event.type === 'USER_RESUMED') {
+        const closeQuestionEffects: FlowEffect[] = state.pauseContext
+          ? [{ type: 'CLOSE_CLAUDE_QUESTION', questionId: state.pauseContext.claude_question_id }]
+          : []
+        return {
+          nextState: { ...state, kind: 'running', pauseContext: undefined },
+          effects: [
+            ...closeQuestionEffects,
+            {
+              type: 'SET_SPRINT_RUN_STATUS',
+              sprintRunId: state.sprintRunId,
+              status: 'RUNNING',
+              clearPauseContext: true,
+            },
+          ],
+        }
+      }
+      if (event.type === 'USER_CANCELLED') {
+        return {
+          nextState: { ...state, kind: 'cancelled', pauseContext: undefined },
+          effects: [
+            {
+              type: 'SET_SPRINT_RUN_STATUS',
+              sprintRunId: state.sprintRunId,
+              status: 'CANCELLED',
+              clearPauseContext: true,
+            },
+          ],
+        }
+      }
+      break
+  }
+  return { nextState: state, effects: [] }
+}
--- a/src/flow/worktree-lease.ts
+++ b/src/flow/worktree-lease.ts
@ -0,0 +1,103 @@
+import type { FlowEffect } from './effects.js'
+
+export type WorktreeLeaseState =
+  | { kind: 'idle' }
+  | { kind: 'acquiring_lock'; jobId: string; productIds: string[] }
+  | { kind: 'creating_or_reusing'; jobId: string; productIds: string[] }
+  | { kind: 'syncing'; jobId: string; productIds: string[] }
+  | { kind: 'ready'; jobId: string; productIds: string[] }
+  | { kind: 'releasing'; jobId: string }
+  | { kind: 'released'; jobId: string }
+  | { kind: 'lock_timeout'; jobId: string; productIds: string[] }
+  | { kind: 'sync_failed'; jobId: string; productIds: string[]; error: string }
+  | { kind: 'stale_released'; jobId: string }
+
+export type WorktreeLeaseEvent =
+  | { type: 'JOB_CLAIMED'; jobId: string; productIds: string[] }
+  | { type: 'LOCK_ACQUIRED' }
+  | { type: 'LOCK_TIMEOUT' }
+  | { type: 'WORKTREE_READY' }
+  | { type: 'SYNC_DONE' }
+  | { type: 'SYNC_FAILED'; error: string }
+  | { type: 'JOB_TERMINAL'; jobId: string }
+  | { type: 'STALE_RESET'; jobId: string }
+
+export type TransitionResult = {
+  nextState: WorktreeLeaseState
+  effects: FlowEffect[]
+}
+
+export function transition(
+  state: WorktreeLeaseState,
+  event: WorktreeLeaseEvent,
+): TransitionResult {
+  switch (state.kind) {
+    case 'idle':
+      if (event.type === 'JOB_CLAIMED') {
+        return {
+          nextState: { kind: 'acquiring_lock', jobId: event.jobId, productIds: event.productIds },
+          effects: [],
+        }
+      }
+      break
+    case 'acquiring_lock':
+      if (event.type === 'LOCK_ACQUIRED') {
+        return {
+          nextState: { kind: 'creating_or_reusing', jobId: state.jobId, productIds: state.productIds },
+          effects: [],
+        }
+      }
+      if (event.type === 'LOCK_TIMEOUT') {
+        return {
+          nextState: { kind: 'lock_timeout', jobId: state.jobId, productIds: state.productIds },
+          effects: [],
+        }
+      }
+      break
+    case 'creating_or_reusing':
+      if (event.type === 'WORKTREE_READY') {
+        return {
+          nextState: { kind: 'syncing', jobId: state.jobId, productIds: state.productIds },
+          effects: [],
+        }
+      }
+      break
+    case 'syncing':
+      if (event.type === 'SYNC_DONE') {
+        return {
+          nextState: { kind: 'ready', jobId: state.jobId, productIds: state.productIds },
+          effects: [],
+        }
+      }
+      if (event.type === 'SYNC_FAILED') {
+        return {
+          nextState: {
+            kind: 'sync_failed',
+            jobId: state.jobId,
+            productIds: state.productIds,
+            error: event.error,
+          },
+          effects: [{ type: 'RELEASE_WORKTREE_LOCKS', jobId: state.jobId }],
+        }
+      }
+      break
+    case 'ready':
+      if (event.type === 'JOB_TERMINAL') {
+        return {
+          nextState: { kind: 'releasing', jobId: state.jobId },
+          effects: [{ type: 'RELEASE_WORKTREE_LOCKS', jobId: state.jobId }],
+        }
+      }
+      if (event.type === 'STALE_RESET') {
+        return {
+          nextState: { kind: 'stale_released', jobId: state.jobId },
+          effects: [{ type: 'RELEASE_WORKTREE_LOCKS', jobId: state.jobId }],
+        }
+      }
+      break
+    case 'releasing':
+      return { nextState: { kind: 'released', jobId: state.jobId }, effects: [] }
+  }
+  // Unknown or forbidden transition — keep current state, no effects
+  return { nextState: state, effects: [] }
+}
--- a/src/git/file-lock.ts
+++ b/src/git/file-lock.ts
@ -0,0 +1,38 @@
+import lockfile from 'proper-lockfile'
+
+export async function acquireFileLock(lockPath: string): Promise<() => Promise<void>> {
+  const release = await lockfile.lock(lockPath, {
+    realpath: false,
+    stale: 30_000,
+    update: 5_000,
+    retries: { retries: 60, factor: 1, minTimeout: 1_000, maxTimeout: 1_000 },
+  })
+  let released = false
+  return async () => {
+    if (released) return
+    released = true
+    await release()
+  }
+}
+
+export async function acquireFileLocksOrdered(
+  lockPaths: string[],
+): Promise<() => Promise<void>> {
+  const sorted = [...lockPaths].sort()
+  const releases: Array<() => Promise<void>> = []
+  try {
+    for (const p of sorted) {
+      releases.push(await acquireFileLock(p))
+    }
+  } catch (err) {
+    for (const r of releases.reverse()) {
+      await r().catch(() => {})
+    }
+    throw err
+  }
+  return async () => {
+    for (const r of releases.reverse()) {
+      await r().catch(() => {})
+    }
+  }
+}
--- a/src/git/job-locks.ts
+++ b/src/git/job-locks.ts
@ -0,0 +1,69 @@
+import * as fs from 'node:fs/promises'
+import * as path from 'node:path'
+import { acquireFileLocksOrdered } from './file-lock.js'
+import {
+  getProductWorktreeLockPath,
+  getWorktreeRoot,
+} from './worktree-paths.js'
+import {
+  getOrCreateProductWorktree,
+  syncProductWorktree,
+} from './product-worktree.js'
+
+type JobReleases = Map<string, Array<() => Promise<void>>>
+const jobReleases: JobReleases = new Map()
+
+export async function setupProductWorktrees(
+  jobId: string,
+  productIds: string[],
+  resolveRepoRoot: (productId: string) => Promise<string | null>,
+): Promise<Array<{ productId: string; worktreePath: string }>> {
+  if (productIds.length === 0) return []
+
+  // Ensure parent dir exists so lockfile creation succeeds
+  await fs.mkdir(path.join(getWorktreeRoot(), '_products'), { recursive: true })
+
+  // Lock-first, alphabetically sorted (deadlock prevention for multi-product idea-jobs)
+  const sorted = [...productIds].sort()
+  const lockPaths = sorted.map(getProductWorktreeLockPath)
+  const releaseAll = await acquireFileLocksOrdered(lockPaths)
+  registerJobLockReleases(jobId, [releaseAll])
+
+  // After lock-acquire, create/reuse worktrees and sync
+  const out: Array<{ productId: string; worktreePath: string }> = []
+  for (const productId of sorted) {
+    const repoRoot = await resolveRepoRoot(productId)
+    if (!repoRoot) continue
+    const { worktreePath } = await getOrCreateProductWorktree({ repoRoot, productId })
+    await syncProductWorktree({ worktreePath })
+    out.push({ productId, worktreePath })
+  }
+
+  return out
+}
+
+export function registerJobLockReleases(
+  jobId: string,
+  releases: Array<() => Promise<void>>,
+): void {
+  const existing = jobReleases.get(jobId) ?? []
+  jobReleases.set(jobId, [...existing, ...releases])
+}
+
+export async function releaseLocksOnTerminal(jobId: string): Promise<void> {
+  const releases = jobReleases.get(jobId)
+  if (!releases) return // idempotent — already released or never locked
+  jobReleases.delete(jobId)
+  for (const release of releases) {
+    try {
+      await release()
+    } catch (err) {
+      console.warn(`[job-locks] release failed for job ${jobId}:`, err)
+    }
+  }
+}
+
+// For tests
+export function _resetJobReleasesForTest(): void {
+  jobReleases.clear()
+}
--- a/src/git/pr.ts
+++ b/src/git/pr.ts
@ -1,7 +1,7 @@
 import { execFile } from 'node:child_process'
 import { promisify } from 'node:util'
 import * as path from 'node:path'
-import * as os from 'node:os'
+import { getWorktreeRoot } from './worktree-paths.js'

 const exec = promisify(execFile)

@ -12,10 +12,17 @@ export async function createPullRequest(opts: {
  body: string
  /** Open as draft PR (mens moet 'm later ready-for-review zetten). Default false. */
  draft?: boolean
-  /** Schakel auto-merge (squash) in. Default true. Voor sprint-mode: false. */
+  /**
+   * PBI-47 (P0): default changed to false. Auto-merge is now enabled
+   * separately via `enableAutoMergeOnPr` only on the **last task** of a
+   * STORY-mode story, with a head-SHA guard to prevent racing earlier
+   * task merges. Callers may still pass `true` for one-off PRs that
+   * are immediately ready to merge; in that case we use the new typed
+   * helper rather than the previous fire-and-forget gh call.
+   */
  enableAutoMerge?: boolean
 }): Promise<{ url: string } | { error: string }> {
-  const { worktreePath, branchName, title, body, draft = false, enableAutoMerge = true } = opts
+  const { worktreePath, branchName, title, body, draft = false, enableAutoMerge = false } = opts

  let url: string
  try {
@ -40,21 +47,14 @@ export async function createPullRequest(opts: {
    return { error: `gh pr create failed: ${msg.slice(0, 300)}` }
  }

-  // Best-effort: enable auto-merge (squash) on the freshly created PR. If the
-  // repo doesn't have "Allow auto-merge" turned on, or the token lacks scope,
-  // gh exits non-zero and we just log. The PR is still valid; auto-merge can
-  // be turned on manually. We do NOT fail the whole createPullRequest call —
-  // the URL was successfully obtained which is the contract this returns.
-  // Bij draft + sprint-flow slaan we dit over: de PR moet eerst handmatig of
-  // via markPullRequestReady ready-for-review worden gezet.
+  // Legacy opt-in: enableAutoMerge=true and not draft → fire the new typed
+  // helper without head-SHA guard (caller didn't supply one). Result is
+  // logged but not propagated — same shape as before.
  if (enableAutoMerge && !draft) {
-    try {
-      await exec('gh', ['pr', 'merge', '--auto', '--squash', url], { cwd: worktreePath })
-    } catch (err) {
-      const stderr =
-        (err as { stderr?: string }).stderr ?? (err as Error).message ?? ''
+    const result = await enableAutoMergeOnPr({ prUrl: url, cwd: worktreePath })
+    if (!result.ok) {
      console.warn(
-        `[createPullRequest] auto-merge enable failed for ${url}: ${stderr.slice(0, 200)}`,
+        `[createPullRequest] auto-merge enable failed for ${url}: ${result.reason} ${result.stderr.slice(0, 200)}`,
      )
    }
  }
@ -62,6 +62,51 @@ export async function createPullRequest(opts: {
  return { url }
 }

+export type AutoMergeFailReason =
+  | 'CHECKS_FAILED'
+  | 'MERGE_CONFLICT'
+  | 'GH_AUTH_ERROR'
+  | 'AUTO_MERGE_NOT_ALLOWED'
+  | 'UNKNOWN'
+
+export type EnableAutoMergeResult =
+  | { ok: true }
+  | { ok: false; reason: AutoMergeFailReason; stderr: string }
+
+function classifyAutoMergeError(stderr: string): AutoMergeFailReason {
+  if (/conflict|not in mergeable state|dirty/i.test(stderr)) return 'MERGE_CONFLICT'
+  if (/checks? failed|status check|required check/i.test(stderr)) return 'CHECKS_FAILED'
+  if (/authentication|HTTP 401|HTTP 403|permission|gh auth/i.test(stderr)) return 'GH_AUTH_ERROR'
+  if (/auto-?merge.*not.*allowed|auto-?merge.*disabled/i.test(stderr)) return 'AUTO_MERGE_NOT_ALLOWED'
+  return 'UNKNOWN'
+}
+
+/**
+ * Enable auto-merge (squash) on a PR with an optional head-SHA guard.
+ *
+ * PBI-47 (P0): when `expectedHeadSha` is provided we pass `--match-head-commit`
+ * so GitHub only activates auto-merge if the remote head still matches the
+ * SHA the caller observed. This prevents racing late pushes from another
+ * worker triggering a merge of a different commit set.
+ */
+export async function enableAutoMergeOnPr(opts: {
+  prUrl: string
+  expectedHeadSha?: string
+  cwd?: string
+}): Promise<EnableAutoMergeResult> {
+  try {
+    const args = ['pr', 'merge', '--auto', '--squash']
+    if (opts.expectedHeadSha) args.push('--match-head-commit', opts.expectedHeadSha)
+    args.push(opts.prUrl)
+    await exec('gh', args, opts.cwd ? { cwd: opts.cwd } : {})
+    return { ok: true }
+  } catch (err) {
+    const stderr =
+      (err as { stderr?: string }).stderr ?? (err as Error).message ?? ''
+    return { ok: false, reason: classifyAutoMergeError(stderr), stderr: stderr.slice(0, 500) }
+  }
+}
+
 // Zet een draft-PR over naar "ready for review". Gebruikt bij sprint-mode
 // wanneer alle stories in de SprintRun DONE zijn — mens reviewt en mergt zelf.
 export async function markPullRequestReady(opts: {
@ -163,8 +208,7 @@ export async function createRevertPullRequest(opts: {
    pbiCode,
  } = opts

-  const worktreeDir =
-    process.env.SCRUM4ME_AGENT_WORKTREE_DIR ?? path.join(os.homedir(), '.scrum4me-agent-worktrees')
+  const worktreeDir = getWorktreeRoot()
  const wtPath = path.join(worktreeDir, `revert-${jobId}`)
  const revertBranch = `revert/${originalBranch}-${jobId.slice(-8)}`

--- a/src/git/product-worktree.ts
+++ b/src/git/product-worktree.ts
@ -0,0 +1,66 @@
+import { execFile } from 'node:child_process'
+import { promisify } from 'node:util'
+import * as fs from 'node:fs/promises'
+import * as path from 'node:path'
+import { getProductWorktreePath } from './worktree-paths.js'
+
+const exec = promisify(execFile)
+
+export async function getOrCreateProductWorktree(opts: {
+  repoRoot: string
+  productId: string
+}): Promise<{ worktreePath: string; created: boolean }> {
+  const worktreePath = getProductWorktreePath(opts.productId)
+  await fs.mkdir(path.dirname(worktreePath), { recursive: true })
+
+  try {
+    await fs.access(worktreePath)
+    return { worktreePath, created: false }
+  } catch {
+    // Path bestaat niet — aanmaken
+  }
+
+  await exec('git', ['fetch', 'origin', '--prune'], { cwd: opts.repoRoot })
+  await exec('git', ['worktree', 'add', '--detach', worktreePath, 'origin/main'], {
+    cwd: opts.repoRoot,
+  })
+
+  // Resolve REAL exclude-pad (linked worktree heeft .git als file, niet directory)
+  const { stdout } = await exec('git', ['rev-parse', '--git-path', 'info/exclude'], {
+    cwd: worktreePath,
+  })
+  const excludePath = path.resolve(worktreePath, stdout.trim())
+  const existing = await fs.readFile(excludePath, 'utf8').catch(() => '')
+  if (!existing.split('\n').includes('.scratch/')) {
+    const sep = existing === '' || existing.endsWith('\n') ? '' : '\n'
+    await fs.appendFile(excludePath, `${sep}.scratch/\n`)
+  }
+
+  return { worktreePath, created: true }
+}
+
+export async function syncProductWorktree(opts: { worktreePath: string }): Promise<void> {
+  const { worktreePath } = opts
+  await exec('git', ['fetch', 'origin', '--prune'], { cwd: worktreePath })
+  await exec('git', ['reset', '--hard', 'origin/main'], { cwd: worktreePath })
+  await exec('git', ['clean', '-fd', '-e', '.scratch/'], { cwd: worktreePath })
+  // Wis .scratch/ inhoud, behoud de map
+  const scratch = path.join(worktreePath, '.scratch')
+  await fs.rm(scratch, { recursive: true, force: true })
+  await fs.mkdir(scratch, { recursive: true })
+}
+
+export async function removeProductWorktree(opts: {
+  repoRoot: string
+  productId: string
+}): Promise<{ removed: boolean }> {
+  const worktreePath = getProductWorktreePath(opts.productId)
+  try {
+    await exec('git', ['worktree', 'remove', '--force', worktreePath], {
+      cwd: opts.repoRoot,
+    })
+    return { removed: true }
+  } catch {
+    return { removed: false }
+  }
+}
--- a/src/git/worktree-paths.ts
+++ b/src/git/worktree-paths.ts
@ -0,0 +1,19 @@
+import * as os from 'node:os'
+import * as path from 'node:path'
+
+export const SYSTEM_WORKTREE_DIRS = new Set(['_products'])
+
+export function getWorktreeRoot(): string {
+  return (
+    process.env.SCRUM4ME_AGENT_WORKTREE_DIR
+    ?? path.join(os.homedir(), '.scrum4me-agent-worktrees')
+  )
+}
+
+export function getProductWorktreePath(productId: string): string {
+  return path.join(getWorktreeRoot(), '_products', productId)
+}
+
+export function getProductWorktreeLockPath(productId: string): string {
+  return path.join(getWorktreeRoot(), '_products', `${productId}.lock`)
+}
--- a/src/git/worktree.ts
+++ b/src/git/worktree.ts
@ -1,8 +1,8 @@
 import { execFile } from 'node:child_process'
 import { promisify } from 'node:util'
 import * as path from 'node:path'
-import * as os from 'node:os'
 import * as fs from 'node:fs/promises'
+import { getWorktreeRoot } from './worktree-paths.js'

 const exec = promisify(execFile)

@ -50,9 +50,7 @@ export async function createWorktreeForJob(opts: {
  const { repoRoot, jobId, baseRef = 'origin/main', reuseBranch = false } = opts
  let { branchName } = opts

-  const parent =
-    process.env.SCRUM4ME_AGENT_WORKTREE_DIR ??
-    path.join(os.homedir(), '.scrum4me-agent-worktrees')
+  const parent = getWorktreeRoot()

  await fs.mkdir(parent, { recursive: true })

@ -121,9 +119,7 @@ export async function removeWorktreeForJob(opts: {
 }): Promise<{ removed: boolean }> {
  const { repoRoot, jobId, keepBranch = false } = opts

-  const parent =
-    process.env.SCRUM4ME_AGENT_WORKTREE_DIR ??
-    path.join(os.homedir(), '.scrum4me-agent-worktrees')
+  const parent = getWorktreeRoot()

  const worktreePath = path.join(parent, jobId)

--- a/src/tools/cleanup-my-worktrees.ts
+++ b/src/tools/cleanup-my-worktrees.ts
@ -1,12 +1,11 @@
 import { z } from 'zod'
 import * as fs from 'node:fs/promises'
-import * as path from 'node:path'
-import * as os from 'node:os'
 import type { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js'
 import { prisma } from '../prisma.js'
 import { requireWriteAccess } from '../auth.js'
 import { toolJson, withToolErrors } from '../errors.js'
 import { removeWorktreeForJob } from '../git/worktree.js'
+import { getWorktreeRoot, SYSTEM_WORKTREE_DIRS } from '../git/worktree-paths.js'
 import { resolveRepoRoot } from './wait-for-job.js'

 const TERMINAL_STATUSES = new Set(['DONE', 'FAILED', 'CANCELLED'])
@ -15,16 +14,20 @@ const ACTIVE_STATUSES = new Set(['QUEUED', 'CLAIMED', 'RUNNING'])
 const inputSchema = z.object({})

 export async function getWorktreeParent(): Promise<string> {
-  return (
-    process.env.SCRUM4ME_AGENT_WORKTREE_DIR ??
-    path.join(os.homedir(), '.scrum4me-agent-worktrees')
-  )
+  return getWorktreeRoot()
 }

 export async function listWorktreeJobIds(worktreeParent: string): Promise<string[]> {
  try {
    const entries = await fs.readdir(worktreeParent, { withFileTypes: true })
-    return entries.filter((e) => e.isDirectory()).map((e) => e.name)
+    return entries
+      .filter(
+        (e) =>
+          e.isDirectory()
+          && !SYSTEM_WORKTREE_DIRS.has(e.name)
+          && !e.name.endsWith('.lock'),
+      )
+      .map((e) => e.name)
  } catch {
    return []
  }
--- a/src/tools/update-job-status.ts
+++ b/src/tools/update-job-status.ts
@ -11,11 +11,30 @@ import { prisma } from '../prisma.js'
 import { requireWriteAccess } from '../auth.js'
 import { toolJson, toolError, withToolErrors } from '../errors.js'
 import { removeWorktreeForJob } from '../git/worktree.js'
+import { getWorktreeRoot } from '../git/worktree-paths.js'
+import { releaseLocksOnTerminal } from '../git/job-locks.js'
 import { resolveRepoRoot } from './wait-for-job.js'
 import { pushBranchForJob } from '../git/push.js'
 import { createPullRequest, markPullRequestReady } from '../git/pr.js'
 import { cancelPbiOnFailure } from '../cancel/pbi-cascade.js'
 import { propagateStatusUpwards } from '../lib/tasks-status-update.js'
+import { transition as prFlowTransition } from '../flow/pr-flow.js'
+import { transition as sprintRunTransition } from '../flow/sprint-run.js'
+import { executeEffects } from '../flow/effects.js'
+import { execFile as execFileCb } from 'node:child_process'
+import { promisify } from 'node:util'
+
+const execGh = promisify(execFileCb)
+
+async function fetchConflictFiles(prUrl: string): Promise<string[]> {
+  try {
+    const { stdout } = await execGh('gh', ['pr', 'view', prUrl, '--json', 'files'])
+    const parsed = JSON.parse(stdout) as { files?: Array<{ path: string }> }
+    return parsed.files?.map((f) => f.path) ?? []
+  } catch {
+    return []
+  }
+}

 const inputSchema = z.object({
  job_id: z.string().min(1),
@ -85,26 +104,37 @@ export type DoneUpdatePlan = {
  branchOverride: string | undefined
  errorOverride: string | undefined
  skipWorktreeCleanup: boolean
+  headSha: string | undefined
 }

 export async function prepareDoneUpdate(
  jobId: string,
  branch: string | undefined,
 ): Promise<DoneUpdatePlan> {
-  const worktreeDir =
-    process.env.SCRUM4ME_AGENT_WORKTREE_DIR ?? path.join(os.homedir(), '.scrum4me-agent-worktrees')
+  const worktreeDir = getWorktreeRoot()
  const worktreePath = path.join(worktreeDir, jobId)
  const branchName = branch ?? `feat/job-${jobId.slice(-8)}`

  const pushResult = await pushBranchForJob({ worktreePath, branchName })

  if (pushResult.pushed) {
+    let headSha: string | undefined
+    try {
+      const { execFile } = await import('node:child_process')
+      const { promisify } = await import('node:util')
+      const exec = promisify(execFile)
+      const { stdout } = await exec('git', ['rev-parse', 'HEAD'], { cwd: worktreePath })
+      headSha = stdout.trim()
+    } catch (err) {
+      console.warn(`[prepareDoneUpdate] failed to resolve HEAD sha for job ${jobId}:`, err)
+    }
    return {
      dbStatus: 'DONE',
      pushedAt: new Date(),
      branchOverride: branchName,
      errorOverride: undefined,
      skipWorktreeCleanup: false,
+      headSha,
    }
  }

@ -115,6 +145,7 @@ export async function prepareDoneUpdate(
      branchOverride: undefined,
      errorOverride: undefined,
      skipWorktreeCleanup: false,
+      headSha: undefined,
    }
  }

@ -126,6 +157,7 @@ export async function prepareDoneUpdate(
    branchOverride: undefined,
    errorOverride: `push failed (${pushResult.reason}): ${snippet}`,
    skipWorktreeCleanup: true,
+    headSha: undefined,
  }
 }

@ -376,6 +408,7 @@ export function registerUpdateJobStatusTool(server: McpServer) {
        let branchToWrite = branch
        let errorToWrite = error
        let skipWorktreeCleanup = false
+        let headShaToWrite: string | undefined

        if (status === 'done') {
          // M12: idea-jobs hebben geen task/plan_snapshot/branch — skip de
@ -401,6 +434,7 @@ export function registerUpdateJobStatusTool(server: McpServer) {
            if (plan.branchOverride !== undefined) branchToWrite = plan.branchOverride
            if (plan.errorOverride !== undefined) errorToWrite = plan.errorOverride
            skipWorktreeCleanup = plan.skipWorktreeCleanup
+            headShaToWrite = plan.headSha
          }
        }

@ -414,9 +448,7 @@ export function registerUpdateJobStatusTool(server: McpServer) {
          job.kind === 'TASK_IMPLEMENTATION' &&
          job.task_id
        ) {
-          const worktreeDir =
-            process.env.SCRUM4ME_AGENT_WORKTREE_DIR ??
-            path.join(os.homedir(), '.scrum4me-agent-worktrees')
+          const worktreeDir = getWorktreeRoot()
          prUrl = await maybeCreateAutoPr({
            jobId: job_id,
            productId: job.product_id,
@ -443,6 +475,7 @@ export function registerUpdateJobStatusTool(server: McpServer) {
            ...(summary !== undefined ? { summary } : {}),
            ...(errorToWrite !== undefined ? { error: errorToWrite } : {}),
            ...(prUrl !== null ? { pr_url: prUrl } : {}),
+            ...(headShaToWrite !== undefined ? { head_sha: headShaToWrite } : {}),
            ...(model_id !== undefined ? { model_id } : {}),
            ...(input_tokens !== undefined ? { input_tokens } : {}),
            ...(output_tokens !== undefined ? { output_tokens } : {}),
@ -468,6 +501,7 @@ export function registerUpdateJobStatusTool(server: McpServer) {
        // sibling-cancel binnen dezelfde SprintRun af bij FAILED.
        // Idea-jobs hebben geen task_id en worden hier overgeslagen.
        let sprintRunBecameDone = false
+        let storyBecameDone = false
        if (
          (actualStatus === 'done' || actualStatus === 'failed') &&
          job.kind === 'TASK_IMPLEMENTATION' &&
@ -479,6 +513,7 @@ export function registerUpdateJobStatusTool(server: McpServer) {
              actualStatus === 'done' ? 'DONE' : 'FAILED',
            )
            sprintRunBecameDone = actualStatus === 'done' && propagation.sprintRunChanged
+            storyBecameDone = actualStatus === 'done' && propagation.storyChanged
          } catch (err) {
            console.warn(
              `[update_job_status] propagateStatusUpwards error for task ${job.task_id}:`,
@ -487,6 +522,72 @@ export function registerUpdateJobStatusTool(server: McpServer) {
          }
        }

+        // PBI-47 (P0): STORY-mode auto-merge timing fix.
+        // Only enable auto-merge when this DONE was the *last* task of a STORY
+        // (story.status flipped to DONE) and pr_strategy === STORY. The
+        // pr-flow transition emits ENABLE_AUTO_MERGE with the head_sha guard.
+        if (
+          storyBecameDone &&
+          updated.pr_url &&
+          headShaToWrite &&
+          job.kind === 'TASK_IMPLEMENTATION'
+        ) {
+          const storyCtx = await prisma.claudeJob.findUnique({
+            where: { id: job_id },
+            select: {
+              task: { select: { story: { select: { status: true } } } },
+              sprint_run: { select: { pr_strategy: true } },
+            },
+          })
+          if (
+            storyCtx?.sprint_run?.pr_strategy === 'STORY'
+            && storyCtx.task?.story.status === 'DONE'
+          ) {
+            const result = prFlowTransition(
+              { kind: 'pr_opened', strategy: 'STORY', prUrl: updated.pr_url },
+              {
+                type: 'STORY_COMPLETED',
+                storyId: '',
+                headSha: headShaToWrite,
+              },
+            )
+            const outcomes = await executeEffects(result.effects)
+            // PBI-47 (C2): route MERGE_CONFLICT to sprint-run flow → PAUSED.
+            // Other reasons (CHECKS_FAILED, GH_AUTH_ERROR, AUTO_MERGE_NOT_ALLOWED, UNKNOWN)
+            // remain warnings; CHECKS_FAILED is already covered by the task-FAIL cascade.
+            for (const o of outcomes) {
+              if (o.effect === 'ENABLE_AUTO_MERGE' && !o.ok) {
+                console.warn(
+                  `[update_job_status] auto-merge fail for ${updated.pr_url}: ${o.reason} ${o.stderr.slice(0, 200)}`,
+                )
+                if (o.reason === 'MERGE_CONFLICT') {
+                  const sprintRunId = await prisma.claudeJob
+                    .findUnique({
+                      where: { id: job_id },
+                      select: { sprint_run_id: true },
+                    })
+                    .then((j) => j?.sprint_run_id)
+                  if (sprintRunId) {
+                    const conflictFiles = await fetchConflictFiles(updated.pr_url)
+                    const conflictResult = sprintRunTransition(
+                      { kind: 'running', sprintRunId },
+                      {
+                        type: 'MERGE_CONFLICT',
+                        prUrl: updated.pr_url,
+                        prHeadSha: headShaToWrite ?? '',
+                        conflictFiles,
+                        resumeInstructions:
+                          'Resolve the conflict on this branch, push, then resume the sprint via the UI.',
+                      },
+                    )
+                    await executeEffects(conflictResult.effects)
+                  }
+                }
+              }
+            }
+          }
+        }
+
        // SPRINT-mode: bij sprint-DONE de draft-PR ready-for-review zetten.
        // Mens reviewt + mergt zelf — geen auto-merge in deze modus.
        if (sprintRunBecameDone && updated.pr_url) {
@ -581,6 +682,12 @@ export function registerUpdateJobStatusTool(server: McpServer) {
          await cancelPbiOnFailure(job_id)
        }

+        // PBI-9: release product-worktree locks on terminal transitions.
+        // No-op for jobs without registered locks (i.e. TASK_IMPLEMENTATION).
+        if (actualStatus === 'done' || actualStatus === 'failed') {
+          await releaseLocksOnTerminal(job_id)
+        }
+
        const queueCount = await prisma.claudeJob.count({
          where: { user_id: userId, status: 'QUEUED' },
        })
--- a/src/tools/verify-task-against-plan.ts
+++ b/src/tools/verify-task-against-plan.ts
@ -15,8 +15,15 @@ const inputSchema = z.object({
  worktree_path: z.string().min(1),
 })

-export async function getDiffInWorktree(worktreePath: string): Promise<string> {
-  const { stdout } = await exec('git', ['diff', 'origin/main...HEAD'], { cwd: worktreePath })
+export async function getDiffInWorktree(
+  worktreePath: string,
+  baseSha?: string,
+): Promise<string> {
+  // PBI-47 (P0): when base_sha is provided, diff against the per-job base
+  // captured at claim-time so verify only sees the current task's changes.
+  // Falls back to origin/main only for legacy callers without base_sha.
+  const range = baseSha ? `${baseSha}...HEAD` : 'origin/main...HEAD'
+  const { stdout } = await exec('git', ['diff', range], { cwd: worktreePath })
  return stdout
 }

@ -58,7 +65,7 @@ export function registerVerifyTaskAgainstPlanTool(server: McpServer) {
              where: { status: { in: ['CLAIMED', 'RUNNING'] } },
              orderBy: { created_at: 'desc' },
              take: 1,
-              select: { id: true, plan_snapshot: true },
+              select: { id: true, plan_snapshot: true, base_sha: true },
            },
          },
        })
@ -67,9 +74,19 @@ export function registerVerifyTaskAgainstPlanTool(server: McpServer) {

        const activeJob = task.claude_jobs[0] ?? null

+        // PBI-47 (P0): require base_sha so diff is scoped to this job's work,
+        // not the full origin/main...HEAD which would include sibling commits
+        // on a reused story/sprint branch.
+        if (activeJob && !activeJob.base_sha) {
+          return toolError(
+            'MISSING_BASE_SHA: This claim has no base_sha. '
+              + 'Re-claim the task (cancel + wait_for_job) so a fresh base_sha is captured.',
+          )
+        }
+
        let diff: string
        try {
-          diff = await getDiffInWorktree(worktree_path)
+          diff = await getDiffInWorktree(worktree_path, activeJob?.base_sha ?? undefined)
        } catch (err) {
          return toolError(
            `git diff failed in worktree (${worktree_path}): ${(err as Error).message ?? 'unknown error'}`,
--- a/src/tools/wait-for-job.ts
+++ b/src/tools/wait-for-job.ts
@ -7,10 +7,15 @@ import { Client } from 'pg'
 import * as fs from 'node:fs/promises'
 import * as os from 'node:os'
 import * as path from 'node:path'
+import { execFile } from 'node:child_process'
+import { promisify } from 'node:util'
 import { prisma } from '../prisma.js'
+
+const execFileP = promisify(execFile)
 import { requireWriteAccess } from '../auth.js'
 import { toolJson, toolError, withToolErrors } from '../errors.js'
 import { createWorktreeForJob } from '../git/worktree.js'
+import { setupProductWorktrees, releaseLocksOnTerminal } from '../git/job-locks.js'

 /** Parse `https://github.com/<owner>/<name>(.git)?` → `<name>`. */
 export function repoNameFromUrl(repoUrl: string | null | undefined): string | null {
@ -181,6 +186,26 @@ export async function attachWorktreeToJob(
      branchName,
      reuseBranch: reused,
    })
+
+    // PBI-47 (P0): capture base_sha so verify_task_against_plan can diff
+    // against the claim-time HEAD instead of origin/main. For reused branches
+    // (siblings already pushed), base_sha = SHA of the worktree HEAD now.
+    // For fresh branches, base_sha = origin/main HEAD which createWorktreeForJob
+    // just checked out.
+    let baseSha: string | null = null
+    try {
+      const { stdout } = await execFileP('git', ['rev-parse', 'HEAD'], { cwd: worktreePath })
+      baseSha = stdout.trim()
+    } catch (err) {
+      console.warn(`[attachWorktreeToJob] failed to resolve base_sha for ${jobId}:`, err)
+    }
+    if (baseSha) {
+      await prisma.claudeJob.update({
+        where: { id: jobId },
+        data: { base_sha: baseSha },
+      })
+    }
+
    return { worktree_path: worktreePath, branch_name: actualBranch, reused_branch: reused }
  } catch (err) {
    await rollbackClaim(jobId)
@ -234,6 +259,11 @@ export async function resetStaleClaimedJobs(userId: string): Promise<void> {

  if (failedRows.length === 0 && requeuedRows.length === 0) return

+  // PBI-9: release any product-worktree locks held by these stale jobs.
+  // No-op for jobs without registered locks (TASK_IMPLEMENTATION).
+  for (const j of failedRows) await releaseLocksOnTerminal(j.id)
+  for (const j of requeuedRows) await releaseLocksOnTerminal(j.id)
+
  // Notify UI via SSE for each transition (best-effort)
  try {
    const pg = new Client({ connectionString: process.env.DATABASE_URL })
@ -371,6 +401,9 @@ async function getFullJobContext(jobId: string) {
      idea: {
        include: {
          pbi: { select: { id: true, code: true, title: true } },
+          secondary_products: {
+            include: { product: { select: { id: true, repo_url: true } } },
+          },
        },
      },
      product: { select: { id: true, name: true, repo_url: true, definition_of_done: true } },
@ -384,6 +417,21 @@ async function getFullJobContext(jobId: string) {
    if (!job.idea) return null
    const { idea } = job
    const { getIdeaPromptText } = await import('../lib/idea-prompts.js')
+
+    // Setup persistent product-worktrees for this idea-job (PBI-9).
+    // Primary product is gated by repo_url via resolveRepoRoot returning null.
+    // Secondary products from IdeaProduct[] need explicit repo_url filter.
+    const involvedProductIds: string[] = []
+    if (idea.product_id) involvedProductIds.push(idea.product_id)
+    for (const ip of idea.secondary_products ?? []) {
+      if (ip.product?.repo_url && !involvedProductIds.includes(ip.product_id)) {
+        involvedProductIds.push(ip.product_id)
+      }
+    }
+    const worktrees = involvedProductIds.length > 0
+      ? await setupProductWorktrees(job.id, involvedProductIds, (pid) => resolveRepoRoot(pid))
+      : []
+
    return {
      job_id: job.id,
      kind: job.kind,
@ -408,6 +456,11 @@ async function getFullJobContext(jobId: string) {
      repo_url: job.product.repo_url,
      prompt_text: getIdeaPromptText(job.kind),
      branch_suggestion: `feat/idea-${idea.code.toLowerCase()}-${job.kind === 'IDEA_GRILL' ? 'grill' : 'plan'}`,
+      product_worktrees: worktrees.map((w) => ({
+        product_id: w.productId,
+        worktree_path: w.worktreePath,
+      })),
+      primary_worktree_path: worktrees[0]?.worktreePath ?? null,
    }
  }