skill-improver

An agent skill by trailofbits, from trailofbits/skills. Tags: automation, code-quality, debugging, developer-tools.

What it does

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.

Install

With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:

npx skills add trailofbits/skills --skill skill-improver

Or copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):

git clone --depth 1 https://github.com/trailofbits/skills
cp -r skills/plugins/code-improver/skills/skill-improver ~/.claude/skills/skill-improver

Safety box score

Grade C, moderate risk: 28 of 100 risk points, rated 2026-09-22. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.

This skill launches an autonomous fix/review loop for Claude Code skills and can edit the target skill until a reviewer reports clean. The main risks are delegated execution in an external workflow, automatic file edits, and a small amount of home-directory probing to locate that workflow.

Category levels, from 0 (none) to 3 (high)
Category	Level	Why
Code & shell execution	2 (moderate)	It explicitly runs a dynamic workflow via `Workflow` and `scriptPath` (`workflows/improve.js`), so the skill can execute code beyond simple text edits.
Network access	0 (none)	The skill does not itself download from or send data to remote hosts; it only references local tools, paths, and a marketplace-installed dependency.
Filesystem reach	2 (moderate)	It probes `~/.claude` and `~/.codex` to locate the workflow and can modify the target skill directory as part of the improvement loop.
Secrets & credentials	0 (none)	There is no instruction to read keychains, tokens, `.env` files, browser sessions, or other credential stores.
Privilege & persistence	0 (none)	No sudo, global installs, services, shell startup edits, or OS-level configuration changes are described.
Instruction hijack surface	0 (none)	The file does not ask the agent to ignore higher-priority instructions, hide actions, or follow fetched instructions; it only describes the loop mechanics.
Irreversible actions	2 (moderate)	It is designed to run a background review/fix loop that keeps changing files until review passes, with escalation and finalize behavior handled automatically.
Transparency	2 (moderate)	The core behavior lives in an external `workflows/improve.js` and a `plugin-dev:skill-reviewer` agent that are not included here, so a reviewer cannot inspect the actual implementation.

Before installing

Read the referenced `workflows/improve.js` and the `plugin-dev:skill-reviewer` implementation before trusting the loop.
Confirm you are comfortable with the skill scanning `~/.claude` and `~/.codex` to find its workflow.
Run it on a disposable copy of the target skill first, since it will edit files automatically until it converges or caps.
Verify the needed plugin dependencies are installed and expected (`claude-plugins-official`, `plugin-dev`).

Rated 2026-09-22 by an automated review of 1 file from the skill's repository.

Source

Repository: trailofbits/skills (all skills from this repository)
Path: plugins/code-improver/skills/skill-improver/SKILL.md
Branch: main
Collection: code-improver
Updated: 2026-09-19

code-improver — Runs an autonomous review-and-fix improvement loop over any code target — a skill, plugin, module, or directory — using a reviewer the user names.
pr-improver — Runs an autonomous review-and-fix improvement loop over the current branch's changes until a PR review comes back clean.
c-review — Performs comprehensive C/C++ security review for memory corruption, integer overflows, race conditions, and platform-specific vulnerabilities.
dwarf-expert — Analyzes DWARF debug information in compiled binaries.
modern-cpp — Guides C++ code toward modern idioms (C++20/23/26). Use when writing new C++ code, modernizing legacy patterns, or working on security-critical C++.
modern-python — Configures Python projects with modern tooling (uv, ruff, ty). Use when creating projects, writing standalone scripts, or migrating from pip/Poetry/mypy/black.