Coding workflow

Pull request review team

Three reviewers check security, performance and tests in parallel; a lead merges, de-duplicates and posts one comment.

ParallelOrchestrator–workersRuns in Claude Code, Codex, GitHub Action

Running it here

1 step needs a connector (github). Run here, those steps are skipped or shown to copy, and the rest still runs. In your own agent with those tools connected, every step runs. Runs here start by hand; to start it from github, run it in your agent or automation tool.

  1. TriggerPull request openedRuns on a pull request's diff ({{diff}}). Paste one to try it.→ github · manual start here
  2. ParallelReview in parallelGive each reviewer {{diff}}. Each returns findings as file:line, severity, why, fix.
    • SecurityInjection, auth checks on new routes, secrets, unsafe deserialization.
    • PerformanceN+1 queries, unbounded loops, missing pagination, large payloads.
    • TestsUntested branches, missing edge cases, flaky patterns.
  3. AgentLead reviewer mergesMerge {{findings}}, drop duplicates and anything you can't point to in {{diff}}. Order by severity. If there are no real findings, say so.
  4. OutputComment on the PRPost {{review}} as one review comment with inline notes.→ github · shown here to copy

Good for

Developers who want a security, performance and test review on a pull request before a human reviewer looks at it.

Last reviewed

Install as a skill

The whole workflow as one SKILL.md. Install it in Claude Code, Codex or Copilot and ask your agent to run “Pull request review team”.

Get the pull-request-review-team skill

What each step does

  1. 1. Pull request opened (Trigger)

    Runs on a pull request's diff ({{diff}}). Paste one to try it.

  2. 2. Review in parallel (Parallel)

    Give each reviewer {{diff}}. Each returns findings as file:line, severity, why, fix.

    • Security: Injection, auth checks on new routes, secrets, unsafe deserialization.
    • Performance: N+1 queries, unbounded loops, missing pagination, large payloads.
    • Tests: Untested branches, missing edge cases, flaky patterns.
  3. 3. Lead reviewer merges (Agent)

    Merge {{findings}}, drop duplicates and anything you can't point to in {{diff}}. Order by severity. If there are no real findings, say so.

  4. 4. Comment on the PR (Output)

    Post {{review}} as one review comment with inline notes.

Use it in your own agent

Fill in the inputs
Open in ClaudeOpen in ChatGPT1 input uses its example until you fill it in.

Save as .claude/skills/pull-request-review-team/SKILL.md for Claude Code, or .agents/skills/pull-request-review-team/SKILL.md for Codex and other Agent Skills tools. Then ask: “run the Pull request review team workflow”.

SKILL.md
---
name: pull-request-review-team
description: Three reviewers check security, performance and tests in parallel; a lead merges, de-duplicates and posts one comment. Use when asked to run the "Pull request review team" workflow.
---

# Pull request review team

Three reviewers check security, performance and tests in parallel; a lead merges, de-duplicates and posts one comment.

## Inputs

Ask for any of these that the user hasn't given:

- `diff` (e.g. "Paste the output of `git diff main...HEAD` (or the PR's .diff)")

## Steps

Run the steps in order. Refer to saved results by name. Don't skip a step that says STOP.

1. **Pull request opened** — Start: Runs on a pull request's diff ({{diff}}). Paste one to try it.
2. **Review in parallel** — Launch these as parallel subagents in one message, then wait for all of them. Give each reviewer {{diff}}. Each returns findings as file:line, severity, why, fix. Save the result as `findings`.
   - **Security**: Injection, auth checks on new routes, secrets, unsafe deserialization.
   - **Performance**: N+1 queries, unbounded loops, missing pagination, large payloads.
   - **Tests**: Untested branches, missing edge cases, flaky patterns.
3. **Lead reviewer merges** — Merge {{findings}}, drop duplicates and anything you can't point to in {{diff}}. Order by severity. If there are no real findings, say so. Save the result as `review`.
4. **Comment on the PR** — Finish: Post {{review}} as one review comment with inline notes.

## Done when

Every step has run, each approval was given, and you've reported what was produced and where it was saved.

Want different steps? Open it in the workflow builder.