Babysitter code-review

Structured code quality assessment with Conventional Comments format, scaled review depth, and soft-gating verdicts preserving user autonomy.

install

source · Clone the upstream repo

git clone https://github.com/a5c-ai/babysitter

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/a5c-ai/babysitter "$T" && mkdir -p ~/.claude/skills && cp -r "$T/library/methodologies/rpikit/skills/code-review" ~/.claude/skills/a5c-ai-babysitter-code-review-9ed438 && rm -rf "$T"

manifest: library/methodologies/rpikit/skills/code-review/SKILL.md

Code Review

Overview

Assess code quality, design, correctness, and maintainability through a structured 9-step review workflow. Uses Conventional Comments format with file-specific references.

When to Use

After implementation phase completes
When reviewing code changes before merge
As part of the /review-code command

Process

Identify modified files via git
Assess change magnitude for review depth
Execute 9-step review: context, correctness, design, testing, security flags, operations, maintainability
Synthesize findings in standardized report
Deliver verdict with rationale

Review Depth Scaling

Under 200 lines: full detail review
200-1000 lines: focused review on critical areas
Over 1000 lines: architectural-level review only

Verdicts

APPROVE: Ready for security review
APPROVE WITH NITS: Non-blocking suggestions only
REQUEST CHANGES: Blocking issues exist (user may override)

Key Rules

Provide specific file paths and line numbers
Include at least one positive comment per review
Use Conventional Comments format with decorations
Explain reasoning, not just observations
Limit critical issues to top 5 per category
Reviews are soft gates preserving user autonomy

Tool Use

Invoke via babysitter process:

methodologies/rpikit/rpikit-review