Batch Review

Batch Review (claude-batch-toolkit)

Integration: Load via @Skill:batch-review.md in DECISION LOOP step 5
RTK Integration: Supports @hooks-session-start.md, @PreCompact.md

Overview

Send non-urgent analysis to Anthropic's Batch API at 50% cost reduction. Ideal for code reviews, documentation, security audits, and architecture analysis that can wait ~1 hour.

When to Activate

Use this skill for batch processing when:

• Code reviews or security audits (non-urgent)
• Documentation generation
• Architecture analysis
• Test generation for existing code
• Refactoring plans
• Changelog generation
• Any task where 50% cost savings > immediate results

Detection Keywords

"batch", "review", "audit", "analyze", "documentation", "architecture", "changelog", "refactoring plan"

Protocol Integration

DECISION LOOP Position

This skill applies at step 5 of the DECISION LOOP:

1. CLASSIFY  -> complexity? backup needed? tools?
2. PROTECT   -> cp file file.bak (for configs, DBs)
3. MEMORY    -> query relevant context + past failures
4. AGENTS    -> check overlaps (if multi-agent)
5. SKILLS    -> @Skill:batch-review.md for domain-specific guidance
6. WORK      -> implement (ALWAYS use worktree for ANY file changes)
7. REVIEW    -> self-review diff before testing
8. TEST      -> completion gates pass
9. LEARN     -> store outcome in memory

Required Pre-Checks

Before using batch review:

• MANDATORY: Worktree created (uap worktree create <slug>)
• Task is non-urgent (can wait ~1 hour)
• Context is self-contained (no interactive debugging needed)
• Full file contents gathered (batch model has NO codebase access)

Setup

git clone git@github.com:s2-streamstore/claude-batch-toolkit.git
cd claude-batch-toolkit && ./install.sh --api-key $ANTHROPIC_API_KEY

Good Batch Candidates

• ✅ Code reviews and security audits
• ✅ Documentation generation
• ✅ Architecture analysis
• ✅ Test generation for existing code
• ✅ Refactoring plans
• ✅ Changelog generation

Bad Candidates (use real-time instead)

• ❌ Anything needed right now
• ❌ Interactive debugging
• ❌ Quick questions
• ❌ Tasks requiring back-and-forth

Usage Pattern

1. Gather all context into a self-contained prompt file
2. Include full file contents (batch model has NO codebase access)
3. Submit via MCP tool send_to_batch with packet_path
4. Check results: batch_list then read from ~/.claude/batches/results/

Integration with Parallel Review Protocol

Instead of running security-auditor/code-quality-guardian synchronously as subagents, submit reviews as batch jobs for 50% cost savings when time allows:

# Assemble review prompt with full diff
git diff --staged > /tmp/review_diff.txt
cat > /tmp/batch_prompt.md << 'EOF'
Review the following diff for security vulnerabilities and code quality issues.
## Diff
EOF
cat /tmp/review_diff.txt >> /tmp/batch_prompt.md

# Submit (via MCP or CLI)
uv run ~/.claude/mcp/claude_batch_mcp.py submit --packet-path /tmp/batch_prompt.md --label "pr-review"

Cost Reference

UAP Protocol Compliance

Completion Gates Checklist

Before reporting task complete:

[x] Schema diffed against test expectations
[x] Tests: X/Y (must be 100%, run 3+ times)
[x] Outputs verified: ls -la
[x] Worktree created and PR prepared
[x] MANDATORY cleanup after PR merge

Verifier-First Integration

If tests are involved:

1. Schema Diff Gate: Compare output schema against test expectations BEFORE running verifier
2. Environment Check: python3 --version; which pip3 gcc make node npm
3. Run Verifier: After each change, minimum 3 runs with TESTS output format

Memory Integration

Store Lessons Learned

uap memory store "Batch review completed: <task_type> saved $<cost> vs real-time" \
  --tags batch-review,<category> --importance 7

Query Relevant Context

uap memory query "batch review <task_type>"

Common Pitfalls

1. Never skip worktree: ALWAYS use uap worktree create <slug> before any file changes
2. Never batch urgent tasks: Batch takes ~1 hour, use real-time for urgent needs
3. Never forget context: Batch model has NO codebase access, include all file contents
4. Never forget cleanup: Run uap worktree cleanup <id> after PR merge

Success Criteria

• All tests pass (100%, 3+ runs)
• Schema verified against test expectations
• Worktree cleanup completed
• Lessons stored in memory
• No debug artifacts left behind
• Cost savings achieved vs real-time review

References

• CLAUDE.md v2.3.0: Universal Agent Patterns (P1-P39)
• Claude Batch API: https://docs.anthropic.com/en/docs/build-with-claude/batch-processing
• Pattern P8: CLI over Libraries
• Pattern P13: Iterative Refinement Loop

Last Updated: 2026-03-09
Compatibility: CLAUDE.md v2.3.0+
RTK Includes: @hooks-session-start.md, @PreCompact.md