review-and-improve

Audits repository documentation against code and auto-fixes mechanical drift before release.

Updated Sep 9, 2026
One-click install
npx skills add https://github.com/martin-cousseau/refund-arena --skill review-and-improve-martin-cousseau
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: review-and-improve
Source: https://github.com/martin-cousseau/refund-arena/tree/main/backend/.agents/skills/review-and-improve
Command: npx skills add https://github.com/martin-cousseau/refund-arena --skill review-and-improve-martin-cousseau

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Documentation drifts out of sync with code as a repository evolves: env vars go undocumented, agents get added but never registered, doc paths break, and architecture diagrams go stale. This Skill sweeps the whole repo for that drift, fixes the mechanical issues in place, and flags everything that needs a human decision. ## Core Features & Use Cases - Consistency sweep: Cross-checks agent/team/workflow registrations in app/main.py against files on disk, the app/config.yaml manifest, env vars against example.env and AGENTS.md, and doc paths against the filesystem. - Auto-fix with guardrails: Repairs stale doc paths, missing env var entries, broken links, and missing manifest entries, while explicitly refusing to touch db/, compose.yaml, Dockerfile, or pyproject.toml. - Live smoke testing: Verifies the running container serves this repo's components, smokes every code agent and team via the AgentOS HTTP API, and runs scripts/mcp_check.sh. - Use Case: Before tagging a public release of an Agno-based agent repo, run the sweep to catch an agent file that was never imported in app/main.py and an env var missing from example.env, then commit the fixes as one chore commit. ## Quick Start Ask the agent to run the review-and-improve sweep across the repo and report what it fixed and what needs your decision.

Frequently Asked Questions about review-and-improve

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

FAQPage Schema
How do I check that repo documentation matches the code before a release?▼

Run the review-and-improve sweep, which inventories registered agents, env vars, doc paths, and scripts, then diffs them against README.md, AGENTS.md, and example.env. It auto-fixes mechanical drift and reports anything needing a human decision.

How do I verify every agent is registered in an Agno AgentOS app?▼

The sweep compares agent, team, and workflow files on disk against imports in app/main.py and entries in app/config.yaml. It also queries the live /agents and /teams endpoints to confirm the running container serves this repo's components.

What kinds of drift does the sweep fix automatically versus flag?▼

It auto-fixes stale doc paths, missing or stale example.env entries, broken links, missing manifest entries, and single-line contradictions. It flags section-level rewrites, code changes beyond imports, and anything in db/, compose.yaml, Dockerfile, or pyproject.toml.

Does the review sweep require a running backend?▼

Yes, the live smoke step expects http://localhost:8000/health to return 200 so it can verify registered components and smoke each agent. If another checkout owns port 8000, the sweep stops and reports the mismatch.

When should I not run the eval step in the review workflow?▼

Skip the evals step when you want to avoid OpenAI costs or latency, since it runs python -m evals --tag release for 1-3 minutes. The sweep asks for confirmation first and routes failures to the eval-and-improve follow-up.