Review Pr

PR Reviewer

Automated review of SparQ pull requests against project standards defined in CLAUDE.md.

When to Use

• Before pushing a branch or opening a PR
• Reviewing someone else's contribution
• Validating your own changes before requesting review

Workflow

Step 1: Identify Changed Files

1. Determine the base branch — use the branch or PR number provided as argument (e.g., /review-pr feature/xyz or /review-pr 42), the PR target branch if in a PR context, or default to main
2. Run git diff --name-only {base}...HEAD to list all changed files
3. Run git diff --stat {base}...HEAD for a size overview
4. Classify each file into categories:
   • cli: bin/**/*.mjs
   • agent: claude/agents/*.md
   • skill: claude/skills/**
   • reference: claude/skills/sparq-shared/references/**
   • test: test/**/*.test.mjs
   • config: biome.json, mcp/*.json, .github/**
   • docs: root *.md files, docs/**
5. If no changed files found, report "No changes detected" and stop

Step 2: Run Quality Gates

Execute and capture results:

1. npm run check — must exit 0 (runs Biome lint + all tests)
2. For each changed .mjs file: node --check {file} — must exit 0
3. For each new .mjs file in bin/lib/: verify a corresponding test/unit/{name}.test.mjs exists
4. For each new .mjs file in bin/lib/commands/: verify it is registered in constants.mjs and help.mjs

Report pass/fail for each gate. If npm run check fails, include the relevant error output.

Step 3: Category-Specific Review

Load references/review-checklist.md and review each changed file against its category-specific criteria. The checklist covers: CLI code, agents, skills, references, tests, MCP configs, security, and prompt optimization compliance.

For any agent/skill/reference changes with >20 lines of prompt content changed, check for evidence that /sparq:prompt-optimizations was applied. Flag as Warning if no optimization evidence found.

Step 4: Produce Review Report

Output using this structure:

## PR Review: {branch name}

**Files reviewed**: {count} ({category breakdown})
**Quality gates**: {pass/fail summary}

### Critical
{Findings that MUST be fixed — broken gates, missing required sections, real credentials}

### Warnings
{Findings that SHOULD be addressed — line count limits, missing optimization, missing tests}

### Info
{Non-blocking suggestions and observations}

### Verdict
{One of: Ready to merge | Needs fixes ({count} critical, {count} warnings) | Needs rework}

Omit empty severity sections. If all gates pass and no findings exist, output a short "All clear" summary.

References

• CLAUDE.md — canonical code standards and quality gates
• .claude/skills/review-pr/references/review-checklist.md — detailed review criteria by category
• claude/skills/sparq-prompt-optimizations/SKILL.md — prompt optimization requirements