skill-improver
Runs an autonomous review-and-fix improvement loop over a Claude Code skill until a review comes back clean, with a cross-round findings…
About
Runs an autonomous review-and-fix improvement loop over a Claude Code skill until a review comes back clean, with a cross-round findings ledger, escalation when fixes stop converging, and a mechanical scope guard. Reviews are performed by the plugin-dev skill-reviewer agent. Use to fix skill quality issues, iteratively refine a skill, or resume a loop after an escalation ('fix my skill', 'improve this skill until it passes review', 'skill improvement loop'). NOT for a one-time review — use the plugin-dev skill-reviewer agent directly."
argument-hint: "<SKILL_NAME_OR_PATH> [--max-rounds N]"
allowed-tools: Bash Glob Read TaskOutput TaskStop Workflow
What it can do
- Runs an autonomous review-and-fix improvement loop over a Claude Code skill until a review comes back clean, with a cross-round findings ledger, escalation when fixes stop converging, and a mechanical scope guard. Reviews are performed by the plugin-dev skill-reviewer agent. Use to fix skill quality issues, iteratively refine a skill, or resume a loop after an escalation ('fix my skill', 'improve this skill until it passes review', 'skill improvement loop'). NOT for a one-time review — use the plugin-dev skill-reviewer agent directly."
- argument-hint: "<SKILL_NAME_OR_PATH> [--max-rounds N]"
- all
Setup
- Install:
npx skills add trailofbits/skills --all - The skill auto-loads when relevant, or invoke it by name in your agent.
Prerequisites
A compatible AI coding agent (Claude Code, Cursor, etc.).
Use cases
Works with
Trust score — 9/10
Trail of Bits is a top-tier security research firm; their skills are professional-grade auditing and fuzzing playbooks used by their own engineers. Open source.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.