Skip to content
SkillModel-led1,603 tokens in use

Skill Improver

Loops the plugin-dev skill-reviewer agent and a fixer over a Claude Code skill until its review comes back clean, with a ledger and scope guard.

8,319 stars on the repository CC-BY-SA-4.0trailofbits/skills · plugins/code-improver/skills/skill-improver
/plugin marketplace add trailofbits/skills
View source

Install command as written in the repository's README

Skill label

measured from its files

What installing Skill Improver costs

114tokens idle
1,603tokens in use

Idle is what sits in context while the skill is installed. In use is its instructions, loaded when it runs. Counts are approximate, made offline with cl100k_base.

Model dependence Model-led

Script-ledMixedModel-led

Mostly instructions, so how well it works depends on the model that follows them.

This describes the skill's structure, how much of it is scripts. It does not say how well it works.

Touches

shelldeclared

Declared tags come from the skill's frontmatter; detected ones from its scripts, its code blocks and the commands it names in inline code. This is detected, not audited.

How it scores

Quality

How well the skill is written, counted from its own files: its frontmatter, its steps, examples, stated output and constraints, its size and whether every file it ships is one it names.How we score it →
Excellent
  • Frontmatter8/15
  • Structure25/25
  • Examples9/15
  • Output15/15
  • Size15/15
  • Resources10/15

Security

Starts at 100 and loses points for each thing in the skill's files that a reader should look at first.How we score it →
Excellent

No findings.

Repository health

Pre-computed score based on development activity, maintenance, community, maturity, and trend momentum.How we score it →
Good
  • Development activity100
  • Maintenance52
  • Community64
  • Maturity32
  • Momentum40

Skill Improver refines a Claude Code skill by running the improve workflow with plugin-dev:skill-reviewer as the reviewer. A fixer subagent edits the skill, the reviewer checks it again, and the loop ends when a review reports zero critical or major findings. It requires the plugin-dev plugin from the claude-plugins-official marketplace.

You give a skill name or path. The skill resolves it to a directory, starts the workflow in the background, and relays whether it converged, hit the round cap, escalated or halted. By default the scope is the skill’s plugin directory.

What You Get

  • A loop of review and fix capped at five rounds unless you ask for another number
  • A findings ledger that survives an escalation and is reloaded on the next run
  • A finalize pass that strips narration comments, collapses version changes to one bump and checks docs against code
  • A report of the outcome and the paths of the ledger and metrics

Common Use Cases

  • Fixing quality problems a skill reviewer found in your skill
  • Refining a new skill until it passes review
  • Resuming after an escalation with your ruling as the decision
  • Narrowing or widening the scope globs the loop may edit

What Makes It Unique

It is wired to one reviewer, the plugin-dev skill-reviewer, so skill authors need no setup beyond installing that plugin. The loop verifies each fix on the next review, guards scope by git diff and by content hash for untracked files, and refuses to fall back to inline edits.

Under The Hood

The skill finds the target with Glob and shell checks, then starts a Workflow script from its plugin. Inside, the reviewer agent and a fixer subagent read and edit files in the skill’s plugin directory, and a ledger and metrics are written to disk. Git diffs and content hashes check scope after each round, and a version bump may touch plugin.json and the marketplace entry. Nothing is committed, and the skill does not fetch web pages or call outside APIs by itself.

Join founders building with open source

Opinionated takes, migration guides, cost-saving tips, and insights from the open source ecosystem.

Subscribe on SubstackJoin 750+ subscribers