feat(hooks): stop a session following its own PR after handoff

[BigSimmo]

Summary

• Add .claude/hooks/pr-handoff-stop.sh, a two-mode Claude Code hook that ends a session's involvement with its own pull request. The PostToolUse half writes a session-scoped marker when a PR-creating call — gh pr create or any create_pull_request MCP tool — returns a real PR URL, and injects a "the handoff is complete, stop" note into the model's context. A failed create returns no URL and writes no marker, so a session is never stranded with no PR to hand off. The marker lives inside the worktree's git dir and is therefore never committed.
• Add the PreToolUse half, which is where the enforcement actually lives. While the marker exists it denies three classes of call: shell CI polling (gh pr checks|status|view|diff|list, gh run watch|view|list|rerun|download, gh api …actions/runs|check-runs|check-suites|/pulls/, sync:pr-branches); GitHub MCP tools whose names carry pull_request, workflow_run, workflow_job, check_run, check_suite, job_log, or update_branch, so a connector is not a way around the shell rule; and the loop machinery Monitor, ScheduleWakeup, and CronCreate, which is how a session parks itself on a PR without running a single command. Committing, pushing, ledger appends, and PR create/merge stay allowed.
• Register both halves in .claude/settings.json with matchers narrow enough that ordinary Read/Edit/Grep calls never spawn the hook.
• Add the ## Stop when the pull request is open section to AGENTS.md stating the rule, its rationale, what the hook denies, and the two documented unlocks: prefix a shell command with CLAUDE_ALLOW_PR_FOLLOW=1, or delete the marker named in the deny message. Both are meant for an explicit user ask.
• Add a closing step 8 (Stop) to .claude/skills/handoff/SKILL.md so the skill that opens the PR is also the skill that ends there.

Why: a session that opens a PR and then stays attached to it — polling checks, watching workflow runs, re-running failed jobs, re-syncing the branch, answering review bots — spends a long tail of usage on work the user did not ask for. Claude Code on the web is the worst case, because the cloud session keeps running after the PR exists and nothing naturally ends the loop. Prose alone has not held for comparable rules in this repo, so the rule ships with a hook that denies the calls.

Reviewer notes:

• Sessions that never create a PR are untouched, because no marker is ever written in them. Run PR sweeps, pr-ci-fix work, and reviews of someone else's PR keep working exactly as before.
• The write-tool exemption regex is anchored ((create|merge)_?pull_?request$), so create_pull_request and merge_pull_request are allowed while create_pull_request_review is denied — replying to review bots on your own PR is part of the loop being stopped.
• Scope limit worth knowing: .claude/settings.json binds Claude Code sessions only. Codex and Cursor agents get the AGENTS.md prose and no hook enforcement.

Verification

• npm run verify:pr-local

Verification not run: npm run verify:pr-local — this branch is a fresh worktree with no node_modules, and installing there while other worktrees hold test/build leases is the concurrent-npm ci corruption this repo has hit before. The change touches no TypeScript, no source, and no test file, so lint, typecheck, and the unit suite have nothing to exercise. What was run instead, from the pushed worktree:

• node scripts/check-docs-links.mjs — docs link check passed: 1629 repo path references resolve.
• node scripts/ci-change-scope.mjs --self-test — CI change scope self-test passed.
• node scripts/check-gate-manifest.mjs — Gate-manifest OK: all 33 verify:cheap gates are enforced in CI (static-pr + mapped jobs), and the 30 static gates are documented consistently.
• node scripts/check-codex-cloud-setup.mjs — [Codex Cloud Check] PASS: static Cloud contracts match.
• node scripts/check-branch-review-ledger.mjs — Branch review ledger guard passed: 609 live table records + 1206 archived …
• node scripts/check-outstanding-issues.mjs — Outstanding-issues guard passed: 253 rows (122 open, 131 archived) …
• Prettier over the three formatted files (AGENTS.md, .claude/settings.json, .claude/skills/handoff/SKILL.md) — All matched files use Prettier code style!
• bash -n .claude/hooks/pr-handoff-stop.sh — clean.
• classifyPullRequestFiles over the changed paths — clinicalRisk: false, operationalRisk: false, ragRanking: false, ui: false, so no governance preflight or RAG impact: line is required for this diff.

Hook behaviour was exercised directly by piping representative hook payloads into the script. Denied: Monitor, ScheduleWakeup, CronCreate, mcp__github__get_pull_request, mcp__github__list_workflow_runs, mcp__github__get_job_logs, mcp__github__update_pull_request_branch, mcp__github__create_pull_request_review, gh pr checks --watch. Allowed: mcp__github__create_pull_request, mcp__github__merge_pull_request, Read, Edit, npm run ledger:append, git push, gh pr merge --squash --auto, and any command carrying the CLAUDE_ALLOW_PR_FOLLOW=1 prefix. Marker written for both a shell create and an MCP create that returned a PR URL; no marker written for a create whose output carried no URL; a different session_id passed straight through. The jq-less fallback path produced the same JSON in both modes.

• npm run verify:ui when UI, routing, styling, browser behavior, reduced-motion, or forced-colors behavior changed

UI verification not run: no UI, routing, styling, or browser behaviour is touched by this diff.

Risk and rollout

• Risk: the failure mode to watch is over-blocking — a denied PreToolUse call that a session legitimately needed. The deny list is deliberately narrow (PR/CI following only) and fires solely in a session that itself created a PR, so a session doing ordinary work, reviewing someone else's PR, or running a Run PR sweep is never affected. Every deny message names both unlocks. The hook exits 0 without a decision on any parse problem, so a malformed payload leaves the tool call exactly as it was rather than failing it.
• Rollback: revert this commit, or remove the PreToolUse/PostToolUse blocks from .claude/settings.json to disable enforcement while keeping the documentation. No state outside the hook's own marker file is involved, and that file lives in the git dir and disappears with the worktree.
• Provider or production effects: None. No application code, no schema, no provider call, no CI workflow change.

Notes

• Branch note: this work was originally staged on claude/cloud-pr-loop-prevention-bc052b, but that branch was 110 commits behind main and carried two unmerged commits from an earlier session (chore: complete sentry setup hardening, chore: complete sentry logging and environment hardening, ~389 lines across src/sentry.*.config.ts, src/lib/env.ts, and src/lib/supabase/client.tsx). Pushing it would have bundled that provider/production work into this PR, so the change was re-applied on a fresh branch cut from current origin/main. Those commits remain untouched on the original branch.

🤖 Generated with Claude Code

---

Summary by CodeRabbit

• New Features

  • Sessions now report the pull request URL and stop follow-up monitoring by default.
  • Added safeguards against post-creation polling, CI supervision, branch synchronization, and review-bot interactions.
  • Explicit overrides remain available when continued follow-up is required.

• Documentation

  • Documented the updated pull request handoff workflow, restrictions, and override options.

• Tests

  • Added coverage for pull request detection, session handling, fallback behavior, cleanup, and blocked follow-up actions.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.

[supabase]

This pull request has been ignored for the connected project sjrfecxgysukkwxsowpy because there are no changes detected in supabase directory. You can change this behaviour in Project Integrations Settings ↗︎.

---

Preview Branches by Supabase.
Learn more about Supabase Branching ↗︎.

[coderabbitai]

Warning

Review limit reached

You’ve reached a temporary PR review limit under our Fair Usage Limits Policy.

Your recent review volume is higher than typical usage, so adaptive limits are currently applied.

Next review available in:46 minutes

Your organization has reached its usage spending cap. Adjust your spending cap in the billing tab.

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: b1e5c794-3450-43b8-ade4-eea11a4863b9

📥 Commits

Reviewing files that changed from the base of the PR and between ed21f6f and 43250d4.

📒 Files selected for processing (1)

• docs/branch-review-ledger.md

📝 Walkthrough

Walkthrough

The PR adds a session-scoped hook that records successful GitHub pull request creation and blocks prohibited follow-up actions. Configuration, handoff guidance, ledger records, and Vitest coverage support the hook.

Changes

PR handoff stop enforcement

Layer / File(s)	Summary
Detect PR creation and track session state .claude/hooks/pr-handoff-stop.sh	The hook parses tool payloads, detects successful GitHub PR creation, validates session IDs, and writes session markers.
Enforce the post-PR stop state .claude/hooks/pr-handoff-stop.sh, .claude/settings.json, AGENTS.md	Pre-tool enforcement blocks PR, CI, monitoring, polling, review, and branch-sync follow-up actions while allowing PR writes and an explicit bypass.
Document and validate handoff behavior .claude/skills/handoff/SKILL.md, tests/pr-handoff-stop.test.ts, docs/branch-review-ledger.md	Handoff guidance defines the stop rule. Tests cover marker handling, fallback parsing, denied actions, and bypass handling. The ledger adds six PR review records.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Sequence Diagram(s)

sequenceDiagram
participant Session
participant pr_handoff_stop
participant SessionMarker
participant GitHubTools
Session->>pr_handoff_stop: create pull request
pr_handoff_stop->>GitHubTools: inspect PR creation output
GitHubTools-->>pr_handoff_stop: GitHub PR URL
pr_handoff_stop->>SessionMarker: write session marker
Session->>pr_handoff_stop: request follow-up check
pr_handoff_stop->>SessionMarker: inspect active marker
pr_handoff_stop-->>Session: deny follow-up operation

Loading

Possibly related PRs

• BigSimmo/Database#594: Introduces the handoff workflow that this PR extends with enforced post-PR stopping rules.
• BigSimmo/Database#1447: Contains related guidance for GitHub PR, review, and Actions operations.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: stopping a session after it hands off its own pull request.
Description check	✅ Passed	The description covers the change, verification results and exceptions, risk, rollback, effects, and notes with sufficient detail.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch claude/pr-handoff-stop-hook

---

_{Comment @coderabbitai help to get the list of available commands.}

[BigSimmo]

Addressed the remaining review notes in 5d96617 (beyond the two inline Devin threads):

• PowerShell tool payloads — command_text now also reads .tool_input.script / .code / .input (jq path and jq-less field extract), so PowerShell create/deny detection is not limited to .command.
• JSON escaping — json_escape folds CR/LF/tab to spaces and strips other C0 controls before quote/backslash escaping.
• Marker accumulation — post prunes claude-pr-handoff-* files older than a day under the git dir.
• pr_status matcher drift — added .*pr_status.* to the PreToolUse matcher in .claude/settings.json so it stays in sync with the script deny regex.
• Shell substring deny — left as designed (cheap, no shell parse). False positives that only mention blocked tokens already have CLAUDE_ALLOW_PR_FOLLOW=1; documented that trade-off in a comment next to the filter.

Also merged origin/main (clean) so the branch is no longer behind. Prior CI reds on this PR were GitHub Actions Service Unavailable during action download (Change scope / PR policy / Semgrep), not product failures — this push should re-run them.

[devin-ai-integration]

Devin Review found 1 new potential issue.

View 3 additional findings in Devin Review.

[BigSimmo]

PR #1649 review-and-fix summary

Tip:935671ca956755ce208726144bb389e01a3b94d2 (product fix 057a5579e538fcacc15feb8a0ac7fa580b3decb4)
Branch:claude/pr-handoff-stop-hook → main
Sync / merge-tree: behind 0 / ahead 9; git merge-tree --write-tree origin/main <tip> clean (exit 0). No sync needed. Auto-merge not armed.

Fixed

• Bugbot high — cross-session prune disarm: removed age-based sibling claude-pr-handoff-* prune from post-mode (every Bash PostToolUse was able to delete another session’s day-old marker when that session only used Read/Edit and never refreshed mtime).
• Marker write fail-open: post-mode emits handoff additionalContext only after a successful marker write ([ -f "$marker" ]); failed writes no longer claim tools are denied.
• Review-bot shell loop: pre-mode now denies gh pr comment / gh pr review (AGENTS.md loop), with tests.
• Devin ledger thread: superseded stacked Run PR rows that pointed at unresolvable HEADs (fff524d7 / f67e5c91 / 3403126b) via ledger:append --supersede to resolvable product tip 057a5579. Thread replied + resolved.

Dispositioned (no code)

• Security review: no P0/P1 (path injection / JSON deny / unlock prefix already controlled).
• Bugbot medium — incomplete gh blocklist beyond comment/review (e.g. gh workflow view, raw curl): residual policy incompleteness; substring deny is best-effort by design.
• Prior Devin/CodeRabbit findings on this PR were already fixed on earlier heads and threads were already resolved.

Threads

All review threads resolved (0 unresolved).

Required CI

Re-triggered by this push. Prior tip 42ae4f37 had required CI still pending/queued (Actions slow / not settled) — not treated as green. Missing checks while dirty/pending ≠ green. Will not babysit the new run.

Local gates (decisive lines)

• npx vitest run tests/pr-handoff-stop.test.ts → Test Files 1 passed (1) / Tests 11 passed (11)
• npm run verify:cheap → Test Files 516 passed (516) / Tests 5457 passed | 1 skipped (5458) (exit 0)
• npm run verify:pr-local → same unit suite pass + Offline RAG fixture and manifest validation passed (36 golden cases, 23 suites). + All matched files use Prettier code style! (exit 0)
• bash -n .claude/hooks/pr-handoff-stop.sh → clean

Residual risks

• Hook enforcement remains best-effort (substring deny; agents can still reach GitHub outside listed tools/commands).
• Markers are no longer pruned across sessions (tiny files under the git dir / worktree lifetime).
• No RAG / clinical / UI / provider surfaces touched.
• Merge left to you.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.