PR Review

Comprehensive pull request review using specialized agents in parallel.

Workflow Overview

1. Get PR Diff -> 2. Analyze Changes -> 3. Select Agents -> 4. Parallel Reviews -> 5. Aggregate -> 6. Verdict

Step 1: Get PR Diff

# Get PR information
gh pr view <PR_NUMBER> --json title,body,files,additions,deletions

# Get full diff
gh pr diff <PR_NUMBER>

# Get changed files list
gh pr view <PR_NUMBER> --json files --jq '.files[].path'

Step 2: Analyze Changes

Categorize changed files:

File Pattern	Category	Relevant Agent
*.py (src/)	Python code	reviewer
*.py (tests/)	Tests	test-architect
requirements*.txt, pyproject.toml	Dependencies	analyzer
Dockerfile, docker-compose*	Infrastructure	(manual review)
*.md	Documentation	(manual review)
*.html, *.css, *.js	Frontend	(manual review)

Step 3: Select Agents

Based on changed file categories, select agents:

Always Run (for code changes)

• reviewer -- Security, types, logging, privacy, best practices

Conditional

Condition	Agent
New dependencies / performance-sensitive code / architecture changes	analyzer
Tests added/modified	test-architect

Step 4: Parallel Reviews

Agent Loading

Read the relevant .claude/agents/*.md files, then spawn explore agents in parallel:

# Step 1: Read agent files (parallel)
Read(".claude/agents/reviewer.md")
Read(".claude/agents/analyzer.md")
# + any situational agents (test-architect.md)

# Step 2: Spawn agents in parallel (single message, multiple Task calls)
Task(
  subagent_type="explore",
  prompt="""You are acting as the reviewer agent.

  <agent-instructions>
  [content from reviewer.md]
  </agent-instructions>

  Review this PR diff:

  <diff>
  [PR diff content]
  </diff>

  PR Title: [title]
  PR Description: [description]

  Focus on: security, types, logging, privacy, best practices, bugs, logic errors.
  Return findings by severity: CRITICAL > HIGH > MEDIUM > LOW.
  Reference specific file paths and line numbers from the diff.
  """
)

Step 5: Aggregate Findings

Collect all agent results and organize:

## PR Review: [PR Title] (#[number])

### Summary
- Files changed: X
- Additions: +Y / Deletions: -Z
- Agents run: [list]

---

### CRITICAL (Must Fix Before Merge)
1. [Finding + agent source + file:line]

### HIGH (Should Fix)
1. [Finding + agent source + file:line]

### MEDIUM (Recommended)
1. [Finding + agent source + file:line]

### LOW (Suggestions)
1. [Finding]

### Positive Observations
- [What's good about this PR]

Step 6: Verdict

Condition	Verdict
Any CRITICAL finding	REQUEST CHANGES
2+ HIGH findings	REQUEST CHANGES
Only MEDIUM/LOW	APPROVE with comments
No findings	APPROVE

### Verdict: [APPROVE / REQUEST CHANGES / COMMENT]

**Reason:** [Brief explanation]

**Required before merge:**
1. [Action item if any]

**Suggestions (non-blocking):**
1. [Optional improvement]

Posting Review

# Post review comment
gh pr review <PR_NUMBER> --comment --body "$(cat <<'EOF'
## PR Review

[Aggregated findings here]
EOF
)"

# Or request changes
gh pr review <PR_NUMBER> --request-changes --body "..."

# Or approve
gh pr review <PR_NUMBER> --approve --body "..."

Important Notes

• Agent loading: Always read .claude/agents/*.md files and inject their prompts into explore Task agents
• Parallelization: Spawn all review agents simultaneously
• Match language: Use the language the user speaks
• Constructive: Don't just list problems -- suggest fixes
• Prioritize: Focus on CRITICAL/HIGH first
• Context: Read CLAUDE.md for project-specific patterns before reviewing