Full code review lifecycle for AI agents — multi-agent parallel analysis with confidence scoring, anti-sycophancy feedback processing, and disciplined fix implementation.

• Performing reviews: Spawns 5 parallel reviewer agents (correctness, security, style, architecture, test coverage), each scoring issues ≥80% confidence before reporting
• Filtering false positives: Excludes pre-existing issues, auto-generated code, vendored deps, personal style
• Receiving reviews: Anti-sycophancy framework — READ → UNDERSTAND → VERIFY → EVALUATE → RESPOND → IMPLEMENT
• Push-back discipline: Concrete criteria for when to accept, negotiate, or reject reviewer feedback
• Fix ordering: Blocking bugs first, then correctness, then style — never batch unrelated fixes

# OpenClaw
clawhub install complete-code-review

# Claude Code / Cursor / any agent
git clone https://github.com/jeremyknows/complete-code-review.git
# Drop the folder in your agent's skills directory, then reference SKILL.md

• Parallel agent spawn capability (Agent tool in Cowork, Task tool in CC, sessions_spawn in OpenClaw)
• gh CLI — for PR history, past comments, blame context (optional but recommended)
• git CLI — for history/blame in multi-agent context

Natural language triggers:

• "Do a code review of this PR"
• "Review my changes"
• "I got feedback on my PR — help me process it"
• "Should I push back on this review comment?"

• references/agent-spawn-templates.md — Exact prompts for each of the 5 reviewer agents
• references/review-quality-checklist.md — Scoring rubric for review quality
• references/openclaw.md — OpenClaw-specific autoresearch scoring loop

Multi-agent review system from Claude Plugins by Boris Cherny.
Anti-sycophancy framework adapted from obra/superpowers by Jesse Vincent.