Community Skills (32): - jat: jat-start, jat-verify, jat-complete - pi-mono: codex-cli, codex-5.3-prompting, interactive-shell - picoclaw: github, weather, tmux, summarize, skill-creator - dyad: 18 skills (swarm-to-plan, multi-pr-review, fix-issue, lint, etc.) - dexter: dcf valuation skill Agents (23): - pi-mono subagents: scout, planner, reviewer, worker - toad: 19 agent configs (Claude, Codex, Gemini, Copilot, OpenCode, etc.) System Prompts (91): - Anthropic: 15 Claude prompts (opus-4.6, code, cowork, etc.) - OpenAI: 49 GPT prompts (gpt-5 series, o3, o4-mini, tools) - Google: 13 Gemini prompts (2.5-pro, 3-pro, workspace, cli) - xAI: 5 Grok prompts - Other: 9 misc prompts (Notion, Raycast, Warp, Kagi, etc.) Hooks (9): - JAT hooks for session management, signal tracking, activity logging Prompts (6): - pi-mono templates for PR review, issue analysis, changelog audit Sources analyzed: jat, ralph-desktop, toad, pi-mono, cmux, pi-interactive-shell, craft-agents-oss, dexter, picoclaw, dyad, system_prompts_leaks, Prometheus, zed, clawdbot, OS-Copilot, and more
10 KiB
name, description
| name | description |
|---|---|
| dyad:swarm-pr-review | Team-based PR review using Claude Code swarm. Spawns three specialized teammates (correctness expert, code health expert, UX wizard) who review the PR diff, discuss findings with each other, and reach consensus on real issues. Posts a summary with merge verdict and inline comments for HIGH/MEDIUM issues. |
Swarm PR Review
This skill uses Claude Code's agent team (swarm) functionality to perform a collaborative PR review with three specialized reviewers who discuss and reach consensus.
Overview
- Fetch PR diff and existing comments
- Create a review team with 3 specialized teammates
- Each teammate reviews the diff from their expert perspective
- Teammates discuss findings to reach consensus on real issues
- Team lead compiles final review with merge verdict
- Post summary comment + inline comments to GitHub
Team Members
| Name | Role | Focus |
|---|---|---|
correctness-reviewer |
Correctness & Debugging Expert | Bugs, edge cases, control flow, security, error handling |
code-health-reviewer |
Code Health Expert | Dead code, duplication, complexity, meaningful comments, abstractions |
ux-reviewer |
UX Wizard | User experience, consistency, accessibility, error states, delight |
Workflow
Step 1: Determine PR Number and Repo
Parse the PR number and repo from the user's input. If not provided, try to infer from the current git context:
# Get current repo
gh repo view --json nameWithOwner -q '.nameWithOwner'
# If user provides a PR URL, extract the number
# If user just says "review this PR", check for current branch PR
gh pr view --json number -q '.number'
Step 2: Fetch PR Diff and Context
IMPORTANT: Always save files to the current working directory (e.g. ./pr_diff.patch), never to /tmp/ or other directories outside the repo. In CI, only the repo working directory is accessible.
# Save the diff to current working directory (NOT /tmp/ or $SCRATCHPAD)
gh pr diff <PR_NUMBER> --repo <OWNER/REPO> > ./pr_diff.patch
# Get PR metadata
gh pr view <PR_NUMBER> --repo <OWNER/REPO> --json title,body,files,headRefOid
# Fetch existing comments to avoid duplicates
gh api repos/<OWNER/REPO>/pulls/<PR_NUMBER>/comments --paginate
gh api repos/<OWNER/REPO>/issues/<PR_NUMBER>/comments --paginate
Save the diff content and existing comments for use in the review.
Step 3: Create the Review Team
Use TeamCreate to create the team:
TeamCreate:
team_name: "pr-review-<PR_NUMBER>"
description: "Code review for PR #<PR_NUMBER>"
Step 4: Create Review Tasks
Create 4 tasks:
- "Review PR for correctness issues" - Assigned to correctness-reviewer
- "Review PR for code health issues" - Assigned to code-health-reviewer
- "Review PR for UX issues" - Assigned to ux-reviewer
- "Discuss and reach consensus on findings" - Blocked by tasks 1-3, no owner (team-wide)
Step 5: Spawn Teammates
Spawn all 3 teammates in parallel using the Task tool with team_name set to the team name. Each teammate should be a general-purpose subagent.
IMPORTANT: Each teammate's prompt must include:
- Their role description (from the corresponding file in
references/) - The full PR diff content (inline, NOT a file path - teammates cannot read files from the team lead's scratchpad)
- The list of existing PR comments (so they can avoid duplicates)
- Instructions to send their findings back as a structured message
Teammate Prompt Template
For each teammate, the prompt should follow this structure:
You are the [ROLE NAME] on a PR review team. Read your role description carefully:
<role>
[Contents of references/<role>.md]
</role>
You are reviewing PR #<NUMBER> in <REPO>: "<PR TITLE>"
<pr_description>
[PR body/description]
</pr_description>
Here is the diff to review:
<diff>
[Full diff content]
</diff>
Here are existing PR comments (do NOT flag issues already commented on):
<existing_comments>
[Existing comment data]
</existing_comments>
## Instructions
1. Read your role description carefully and review the diff from your expert perspective.
2. For each issue you find, classify it as HIGH, MEDIUM, or LOW severity using the guidelines in your role description.
3. Send your findings to the team lead using SendMessage with this format:
FINDINGS:
```json
[
{
"file": "path/to/file.ts",
"line_start": 42,
"line_end": 45,
"severity": "MEDIUM",
"category": "category-name",
"title": "Brief title",
"description": "Clear description of the issue and its impact",
"suggestion": "How to fix (optional)"
}
]
- After sending your initial findings, wait for the team lead to share other reviewers' findings.
- When you receive other reviewers' findings, discuss them:
- ENDORSE issues you agree with (even if you missed them)
- CHALLENGE issues you think are false positives or wrong severity
- ADD context from your expertise that strengthens or weakens an issue
- Send your discussion responses to the team lead.
Be thorough but focused. Only flag real issues, not nitpicks disguised as issues.
IMPORTANT: Cross-reference infrastructure changes (DB migrations, new tables/columns, API endpoints, config entries) against actual usage in the diff. If a migration creates a table but no code in the PR reads from or writes to it, that's dead infrastructure and should be flagged.
### Step 6: Collect Initial Reviews
Wait for all 3 teammates to send their initial findings. Parse the JSON from each teammate's message.
### Step 7: Facilitate Discussion
Once all initial reviews are in:
1. Send each teammate a message with ALL findings from all reviewers (labeled by who found them)
2. Ask them to discuss: endorse, challenge, or add context
3. Wait for discussion responses
The message to each teammate should look like:
All initial reviews are in. Here are the findings from all three reviewers:
Correctness Reviewer Findings:
[list of issues]
Code Health Reviewer Findings:
[list of issues]
UX Reviewer Findings:
[list of issues]
Please review the other reviewers' findings from YOUR expert perspective:
- ENDORSE issues you agree are real problems (say "ENDORSE: