SkillAdam plugin
Plugin for Codex, Claude Code, Cursor and Copilot that tests your SKILL.md on tasks and edits it after validation.
- 109 ★ on the repository
- Last commit
Extended access
Why this level
- The installer creates a Python environment and registers a skill in the agent, and runs consume account quota.
- Skill and task content is sent to the model provider.
Install
- Works in
Manual install
git clone https://github.com/ruc-datalab/SkillAdam.git && cd SkillAdam && ./integrations/claude-code/install.shClone and installer for Claude Code. Restart the agent after installing.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The installer creates a Python environment and registers a skill in the agent, and runs consume account quota.
- Skill and task content is sent to the model provider.
This is third-party code. Review the repository files before installing.
What it does
SkillAdam improves one existing `SKILL.md` toward a stated goal. The agent prepares relevant tasks and scoring rules, runs the current skill, proposes edits and validates the revised version. By default edits are chosen automatically, and the file is updated only if the new version passes validation. The installer creates a dedicated Python environment and registers the `skilladam-optimize` skill in the chosen agent. Model calls use the sign-in of an agent you already set up, so no separate key or GPU is needed. An interrupted run resumes from its run directory, and edits can be reviewed and accepted one by one before applying.
Who it is for. Skill authors who want to measure rather than guess whether `SKILL.md` improved after an edit.
Good fit when
- A code review skill gives long explanations without a verdict and must be tightened using examples.
- Before publishing a skill its edits should be checked on a task set.
- A run was interrupted and must continue from the same run directory.
- A team wants to accept skill edits selectively and see each change.
Not a fit when
- Several skills must be edited at once; one run handles a single `SKILL.md`.
- The skill relies on neighboring files the agent will not see in test runs; they are not attached automatically.
- Skill and task content must not be sent to a model provider.
Example request
Use SkillAdam to improve `/path/to/SKILL.md` for concise code reviews with a concrete recommendation and show the edits before applying.Limitations
Python 3.10 or newer, Git and the CLI of the chosen agent (`codex`, `claude`, `cursor-agent` or `copilot`) with an active sign-in are required. Copilot needs the `code` CLI by default, otherwise run the installer with `--skip-vscode-registration`. Test runs consume account quota, and skill and task content goes to the provider. Supporting skill resources and workspace files are not attached to runs automatically. Keep the skill under version control so changes can be rolled back.
How to disable. Remove the registered `skilladam-optimize` skill from your agent skills directory, the clone folder and the created Python environment. Git history restores the previous `SKILL.md`.
FAQ
Which agents are supported?
Codex, Claude Code, Cursor Agent and GitHub Copilot. Each has its own installer under `integrations/`.
Is a separate model key needed?
No. By default it uses the sign-in of the agent you already configured, though calls draw on your subscription quota.
Can I pick edits manually?
Yes. Add a request to show edits before applying; each can be accepted or rejected.
How do I resume a run?
Ask the agent to continue the optimization in the same run directory.
How many skills does it process at a time?
One. Give the absolute path to `SKILL.md` and the improvement goal in the request, plus a task example if needed.
Discussion
No questions or reviews yet.
Related
PluginsA plugin and skill that pick three vetted skills for a task from an offline registry, with a safety grade and reasons.
A skill for creating skills that proves them by behavior: before-and-after runs, regression evals and ecosystem audit
A skill that designs, tests, evaluates and packages other agent skills, starting from architecture
A plugin of seven skills for Claude Code and Codex: prompts, subagents, skill authoring, editing, web search and a browser.
Runs a coding agent over tasks with known answers, studies failures and writes new skills and prompts, versioned in Git.
A meta-skill that folds rules scattered across AGENTS.md, CLAUDE.md and Cursor files into one project skill with task routes.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| SkillAdam (this card) | no | 109 | |
| skill-creator: skills from OpenAPI, GraphQL and MCP | no | 611 | creates skills; SkillAdam edits a finished `SKILL.md` from run results |
| Functional Skill Creator | no | 256 | helps write a new skill; here an existing file is tested on tasks and improved |
README in short
The method is described in an arXiv paper on stable and efficient skill evolution. Separate configurations and a guide exist for reproducing experiments.