Back to search
skillproof
SkillGitHub
Test, score and showcase AI agent skills and prompts. Free, open source, browser-based.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Checked today
Badge for your README
Markdown
[](/listings/piyush2707--skillproof)HTML
<a href="/listings/piyush2707--skillproof"><img src="/api/listings/piyush2707--skillproof/badge.svg" alt="findsafeskills safety rating" /></a>Install
i
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
README.md · README · 256 lines
- <div align="center">
- <img src="assets/skill-logo.jpg" alt="SkillProof: agent skills are easy to write, hard to trust" width="100%">
- <br>
- **Free, open-source tools to check, measure and showcase AI agent skills.**<br>
- Everything runs in your browser. Nothing is uploaded.
- <br>
- [](https://github.com/piyush2707/skillproof/actions/workflows/validate.yml)
- 
- 
- 
- [](https://github.com/piyush2707/skillproof/stargazers)
- <br>
- ### [Open the live site](https://piyush2707.github.io/skillproof/)
- [Context Meter](https://piyush2707.github.io/skillproof/meter/) ·
- [Conflict Map](https://piyush2707.github.io/skillproof/conflicts/) ·
- [Eval Cards](https://piyush2707.github.io/skillproof/evals/) ·
- [Skill Pages](https://piyush2707.github.io/skillproof/skills/)
- </div>
- <br>
- ---
- ## The problem
- A skill is a folder with a `SKILL.md` file that teaches an agent one task. Skills are quick to write, but it is hard to know:
- - Is it **well-formed**?
- - How much **context** does it use before any task starts?
- - Does it **clash** with the other skills you installed?
- - Is there any **evidence** that it actually helps?
- Most skill repositories are collections of instructions with no proof. SkillProof adds measurable checks and visible evidence, using only GitHub and your browser.
- ---
- ## The toolkit
- <table>
- <tr>
- <td width="50%" valign="top">
- ### Context Meter
- Estimate how many tokens your skills use before any task starts.
- [**Open →**](https://piyush2707.github.io/skillproof/meter/)
- </td>
- <td width="50%" valign="top">
- ### Conflict Map
- Find skills whose descriptions overlap and may confuse an agent.
- [**Open →**](https://piyush2707.github.io/skillproof/conflicts/)
- </td>
- </tr>
- <tr>
- <td width="50%" valign="top">
- ### Eval Cards
- A JSON format and viewer for with/without-skill test results.
- [**Open →**](https://piyush2707.github.io/skillproof/evals/)
- </td>
- <td width="50%" valign="top">
- ### Skill Pages
- A page per example skill: cost, install steps and evidence.
- [**Open →**](https://piyush2707.github.io/skillproof/skills/)
- </td>
- </tr>
- <tr>
- <td colspan="2" valign="top">
- ### Validator Action
- A GitHub Action that checks `SKILL.md` files and eval cards on every push and pull request. It also checks that deliberately broken fixtures are rejected, so the validator itself is tested. [**View the code →**](validators/)
- </td>
- </tr>
- </table>
- ---
- ## Try it in 30 seconds
- No install. No account.
- 1. Open the [Context Meter](https://piyush2707.github.io/skillproof/meter/) and tap **Load sample**.
- 2. Paste your own `SKILL.md` and tap **Measure**.
- 3. Paste several skills, separated by a line of `=====`, into the [Conflict Map](https://piyush2707.github.io/skillproof/conflicts/).
- 4. Record test results in an [Eval Card](schema/eval-card.md) and view them on the [Eval Cards page](https://piyush2707.github.io/skillproof/evals/).
- ---
- ## How it fits together
- ```mermaid
- flowchart LR
- A[SKILL.md] --> B[Context Meter]
- A --> C[Conflict Map]
- A --> D[Skill Pages]
- E[Eval Card JSON] --> F[Eval Cards viewer]
- E --> D
- G[Pull request] --> H[Validator Action]
- A --> H
- E --> H
- ```
- Everything on the left is plain text in the repository. Everything on the right is either a web page in your browser or a GitHub Action.
- ---
- ## What the validator checks
- The workflow in [`.github/workflows/validate.yml`](.github/workflows/validate.yml) runs on every push and pull request.
- | Check | Looks at |
- |---|---|
- | Frontmatter | Exists, with a `name` and `description` |
- | Name | Lowercase letters, numbers and hyphens only |
- | Description | Long enough to be useful and says when to use the skill |
- | File links | Links inside the skill point to files that exist |
- | Length | Warns on very long skill bodies |
- | Eval cards | Required fields, valid results and unique case ids |
- Skill rules live in [`validators/rules.json`](validators/rules.json).
- ---
- ## Eval Cards
- An Eval Card is a file named `<skill>.eval.json` in [`evals/`](evals/). Each case records a prompt, what a good answer needs, and the result **without** and **with** the skill.
- | Result | Meaning |
- |---|---|
- | `pass` | The answer met the expectation |
- | `partial` | Partly met |
- | `fail` | Did not meet it |
- | `untested` | Not run |
- The `evidence_type` field says whether a card is `example`, `manual` or `automated`, so sample data is never mistaken for a real test. See the [format reference](schema/eval-card.md).
- ---
- ## Repository layout
- ```
- skillproof/
- ├── docs/ Website (GitHub Pages)
- │ ├── index.html Home page
- │ ├── meter/ Context Meter
- │ ├── conflicts/ Conflict Map
- │ ├── evals/ Eval Cards viewer
- │ ├── skills/ Skill Pages
- │ └── assets/ Banner image
- ├── skills-examples/ Example skills
- ├── evals/ Eval cards
- ├── schema/ Eval Card format reference
- ├── validators/ Validator scripts and rules
- ├── tests/fixtures/ Broken files the validator must reject
- └── .github/ Workflows and issue templates
- ```
- ---
- ## Honest limits
- - **Token counts are estimates**, at about 4 characters per token. A real tokenizer will differ.
- - **Conflict scores are word overlap.** They are a warning to review, not proof that an agent will get confused.
- - **SkillProof does not run any agent.** Eval Cards record evidence that a person ran and logged.
- - **Compatibility:** SkillProof reads the `SKILL.md` format (name and description in frontmatter). Support for specific agents will be listed here only after it has been tested.
- - The example eval card for `hello-skill` is sample data and is labelled that way.
- ---
- ## Roadmap
- - [x] Validator Action
- - [x] Context Meter
- - [x] Conflict Map
- - [x] Eval Cards
- - [x] Skill Pages
- - [ ] RAG Failure Attribution skill
- - [ ] RepoReel: turn a GitHub repo into a short explainer video in the browser
- - [ ] More validator checks and community-submitted skills
- Full details in [`ROADMAP.md`](ROADMAP.md).
- ---
- ## FAQ
- <details>
- <summary><b>Does SkillProof upload my skills anywhere?</b></summary>
- <br>
- No. The web tools run entirely in your browser. What you paste stays on your device.
- </details>
- <details>
- <summary><b>Does it work with my agent?</b></summary>
- <br>
- SkillProof reads the <code>SKILL.md</code> format. It does not run any agent, so it checks your files rather than an agent's behavior. Agent-specific compatibility will be listed only after testing.
- </details>
- <details>
- <summary><b>Why are token counts approximate?</b></summary>
- <br>
- They use about 4 characters per token, which is a rough rule. Different models count tokens differently, so treat the numbers as a guide for comparing skills.
- </details>
- <details>
- <summary><b>Can I add my own skill?</b></summary>
- <br>
- Yes. Add a folder under <code>skills-examples/</code> with a <code>SKILL.md</code> and open a pull request. See <a href="CONTRIBUTING.md">CONTRIBUTING.md</a>.
- </details>
- <details>
- <summary><b>Can I trust the example eval card?</b></summary>
- <br>
- It is labelled "example" because it is sample data showing the format, not a real test run. Real results use <code>manual</code> or <code>automated</code>.
- </details>
- ---
- ## Contributing
- Skills, eval cards and fixes are welcome. Read [`CONTRIBUTING.md`](CONTRIBUTING.md) first and follow the [Code of Conduct](CODE_OF_CONDUCT.md). Report vulnerabilities privately as described in [`SECURITY.md`](SECURITY.md).
- If SkillProof is useful to you, a star on the repository helps others find it.
- ---
- ## License
- MIT. See [`LICENSE`](LICENSE).
- <br>
- <div align="center">
- Made by [Piyush](https://github.com/piyush2707) at **Welvox**
- </div>
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.