ralph-review-trio

Runs a three-tier sequential code review on completed branches using Haiku, Sonnet, and Opus reviewers.

3.4k|487|Updated Jul 25, 2025
One-click install
npx skills add https://github.com/davepoon/buildwithclaude --skill ralph-review-trio
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: ralph-review-trio
Source: https://github.com/davepoon/buildwithclaude/tree/main/plugins/ralph-review-trio/skills/ralph-review-trio
Command: npx skills add https://github.com/davepoon/buildwithclaude --skill ralph-review-trio

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) and assets (resource) components.

What problem does it solve?

Solo developers merging their own branches lack structured pre-merge verification, so surface bugs, logic errors, and architectural drift slip through a single-pass review. This Skill enforces a layered review loop that restarts from the beginning whenever any tier fails.

Core Features & Use Cases

  • Three-tier sequential review: Tier 1 (Haiku) checks surface issues like debug code and magic numbers, Tier 2 (Sonnet) traces cross-file logic and contracts, Tier 3 (Opus) audits architecture, governance drift, and factual claims.
  • Restart-on-fail loop: Any tier failure triggers fixes and a full restart from Tier 1, ensuring later tiers always review the final state of the code.
  • Structured RESULT blocks: Each tier emits a fenced verdict block with findings, file paths, and evidence so the controller can decide pass, restart, or merge.
  • Use Case: After finishing a feature branch that adds a retry wrapper to an HTTP client, invoke the review to catch a leftover print statement at Tier 1, a swallowed exception at Tier 2, and confirm governance cadence at Tier 3 before merging.

Quick Start

Ask Claude to run /ralph-review on your completed feature branch to execute the three-tier pre-merge review.

Frequently Asked Questions about ralph-review-trio

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a multi-tier AI code review on a branch?

Invoke the /ralph-review command on a code-complete branch. It dispatches three reviewer subagents in sequence — Haiku for surface checks, Sonnet for logic tracing, and Opus for deep architectural analysis — and reports a final pass only when all three succeed.

What happens when a review tier fails in ralph-review-trio?

When any tier returns a fail verdict, the main agent applies fixes and the loop restarts from Tier 1 rather than continuing. This ensures every tier re-verifies the final state of the code after changes.

When should I not use a three-tier code review?

Skip it for work-in-progress branches, documentation-only diffs, and hotfixes where speed matters more than verification. The tiers assume the implementation is complete, so mid-development runs produce misleading failures.

What does each review tier check for?

Tier 1 catches debug code, magic numbers, and copy-paste at the diff surface. Tier 2 traces type, schema, and config contracts across files. Tier 3 audits architectural fit, governance drift, dead modules, and unsourced factual claims.

Does the review work when the code graph MCP is unavailable?

Yes, but the tier must retry once and then fall back to an Explore-style subagent performing the equivalent manual audit. Marking the graph unavailable without fallback evidence counts as a silent skip and fails the review.