#7979th of 85 in Skills, plugins and rules
improve
Agent skill that audits a codebase and writes plans for cheaper agents to run
- Stars
- 9.2k
- License
- MIT
- Last commit
- Sep 2026
Missed the latest GitHub check; these facts may be older.
Overview
An Agent Skills-format skill, invoked as /improve, that maps a repo, runs parallel audits across nine categories, re-checks each finding against the cited file and line, and writes self-contained markdown plans into plans/. It never edits source itself. An optional execute mode dispatches a cheaper model in an isolated git worktree and reviews the diff against the plan.
Who it is for: Developers using coding agents who want audited, reviewable work plans
Strengths
- Advisor re-reads every cited file:line before showing findings, dropping false positives
- Plans inline code excerpts, verification commands and STOP conditions for small executor models
- Writes only to plans/; executors work in disposable worktrees and merging stays manual
- Scoped modes: quick, deep, branch, security, plus reconcile to refresh the backlog
Weaknesses
- Needs an agent host that supports the Agent Skills format
- Full audits fan out parallel subagents on a capable model; cost is not documented
- No release tagged and no stated model or version requirements
- Plan quality depends on the advisor model; no benchmarks given
What it needs
- no GPU
- Needs Agent Skills-compatible coding agent, Node.js (npx) for install, git, GitHub (optional, for --issues)
Also in Skills, plugins and rules
See all 85| Rank | Project | Score |
|---|---|---|
| 1 | skillsAgent skills for planning, TDD, debugging and code review | 94 out of 100 |
| 2 | eccAgents, skills, hooks and rules bundle for Claude Code, Codex and other harnesses | 94 out of 100 |
| 3 | cavemanCuts token use in coding agents with terse replies and compressed inputs | 94 out of 100 |
| #4 | ponytailSkill prompt that makes coding agents write less code and add tests | 92 out of 100 |
| #5 | impeccableDesign skill with 24 commands and a detector for AI-generated frontends | 92 out of 100 |