repo-to-skill
Skill9 starsChanged 8 months agoMITView on GitHub
Generate an agent skill from a GitHub open-source repository. Use when asked to create a skill from a repo URL, turn a GitHub project into a usable skill, or generate skill files for a CLI/library/framework. Triggers on phrases like "create a skill from this repo", "make a skill for vllm", "turn this GitHub project into a skill", "generate a skill from repo". Also use when asked to "skillify" a tool or generate usage instructions from source code. Command usage — `/skill repo-to-skill <github-repo-url>`.
- Installs packages
3 other projects use a version of this.
Adopted by
Most recently taken up by
- imMamdouhaboammar/repo-to-skilltheir own version, first seen Sep 26, 2026
- louiswang524/repo-to-skilltheir own version, first seen Mar 23, 2026
- NeuroAIHub/awesome_cognitive_and_neuroscience_skillstheir own version, first seen Mar 2, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- NeuroAIHub/awesome_cognitive_and_neuroscience_skills★ 39same name
- louiswang524/repo-to-skill★ 4same name
- imMamdouhaboammar/repo-to-skill★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
5 rules. 0 are common, 3 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 68
Create the skill in the user's preferred location (default: ~/.agents/skills/<tool-name>/).
- Distinctive: line 122
Verify the skill works by actually using the tool:
- Distinctive: line 128
If tests fail, update the skill with corrections.
- Distinctive: line 151
After generating and testing, run a structured evaluation.
- Very specific: line 155
Write 3-5 realistic prompts a user would send to an agent with this skill.
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Repo-to-Skill Generator
Generate a complete, tested, and evaluated agent skill from any GitHub repository.
Command
/skill repo-to-skill <github-repo-url>Example:
/skill repo-to-skill https://github.com/vllm-project/vllmOverview
Given a GitHub repo URL, this skill:
- Clones and analyzes the repository
- Extracts usage patterns from README, docs, examples, CLI help
- Generates a working skill (SKILL.md + resources)
- Tests the skill by running basic operations
- Evaluates the skill with test prompts and assertions
Process
Step 1: Clone & Analyze
# Clone repo (shallow for speed)
git clone --depth 1 <repo-url> /tmp/repo-to-skill/<repo-name>Analyze in this order (stop when you have enough):
- README.md — primary source for installation, quickstart, features
- docs/ or documentation/ — detailed usage guides
- examples/ — concrete usage patterns (high value)
- CLI help — if it's a CLI tool, check
--helpoutput - setup.py / pyproject.toml / package.json — dependencies and entry points
- Source code — only if docs are insufficient; focus on public API
Extract:
- What it does (one sentence)
- Installation method (pip, npm, cargo, etc.)
- Core commands/API (the 5-10 most common operations)
- Configuration (env vars, config files, required setup)
- Input/Output formats (what goes in, what comes out)
- Common patterns (from examples/ or README)
Step 2: Classify the Tool
Determine the tool type to shape the skill structure:
| Type | Skill Focus | Example |
| CLI tool | Commands, flags, common workflows | vllm, ffmpeg, gh |
| Python library | API patterns, code snippets | transformers, pandas |
| Framework | Project setup, config, patterns | FastAPI, Next.js |
| Service | API endpoints, auth, integration | Stripe, OpenAI |
Step 3: Generate Skill Structure
Create the skill in the user's preferred location (default: ~/.agents/skills/<tool-name>/).
<tool-name>/
├── SKILL.md # Core instructions
├── references/
│ ├── api-reference.md # Full API/CLI reference (if large)
│ ├── examples.md # Curated usage examples
│ └── configuration.md # Config options (if complex)
└── scripts/
└── setup.sh # Installation/setup script (if needed)SKILL.md Template
---
name: <tool-name>
description: <what it does and when to trigger — be specific and slightly pushy>
---
# <Tool Name>
<One-line description>
## Installation
<Installation command>
## Quick Start
<Minimal working example — the "hello world">
## Core Operations
<The 5-10 most common operations with examples>
<Use imperative form: "Run X to do Y">
## Common Patterns
<2-3 real-world workflow examples>
## Troubleshooting
<Top 3 gotchas or common errors>
## References
- For full API details, see [references/api-reference.md](references/api-reference.md)
- For more examples, see [references/examples.md](references/examples.md)Step 4: Test the Skill
Verify the skill works by actually using the tool:
- Install test — run the installation command
- Smoke test — run the quickstart example
- Feature test — try 2-3 core operations from the skill
If tests fail, update the skill with corrections.
Document test results:
✅ Installation: pip install vllm → success
✅ Quick start: vllm serve model → server started
❌ Feature: offline batching → fixed: added --dtype auto flagStep 5: Validate & Deliver
Before delivering:
- SKILL.md under 500 lines
- Frontmatter has name + description
- Description includes trigger phrases
- All referenced files exist
- Examples are tested and working
- No secrets or credentials in skill files
Present the skill to the user with a summary of what it covers.
Step 6: Evaluate the Skill
After generating and testing, run a structured evaluation. See references/eval-schemas.md for full JSON schemas.
Create Test Prompts
Write 3-5 realistic prompts a user would send to an agent with this skill. Save to evals/evals.json:
{
"skill_name": "vllm",
"evals": [
{
"id": 1,
"prompt": "Serve Llama-3-8B with vllm on port 8000",
"expected_output": "Working vllm serve command with correct model and port",
"expectations": [
"Command includes 'vllm serve' or 'python -m vllm.entrypoints'",
"Port 8000 is specified",
"Model name is correct"
]
}
]
}Run Evals (with-skill vs baseline)
For each test prompt, spawn two runs in parallel:
- With-skill run: Agent has the generated skill loaded
- Baseline run: Same prompt, no skill
Save outputs to <skill-name>-workspace/iteration-<N>/eval-<ID>/with_skill/ and without_skill/.
Grade Results
For each run, evaluate assertions and produce grading.json:
{
"expectations": [
{
"text": "Command includes 'vllm serve'",
"passed": true,
"evidence": "Output contains: vllm serve meta-llama/Llama-3-8B"
}
],
"summary": {
"passed": 3,
"failed": 0,
"total": 3,
"pass_rate": 1.0
}
}For assertions that can be checked programmatically (file exists, command runs, output matches pattern), write and run a script instead of eyeballing.
Aggregate & Report
Produce a benchmark comparing with-skill vs baseline:
- pass_rate: mean ± stddev across runs
- time_seconds: execution time
- tokens: token usage
- delta: improvement from skill
Present results to user. If pass_rate < 0.7, iterate on the skill.
Guidelines
- Be concise — only include what the model doesn't already know
- Prefer examples over explanations — show, don't tell
- Test everything — never include untested commands
- Progressive disclosure — keep SKILL.md lean, put details in references/
- Version-aware — note the repo version/commit analyzed
- Installation-first — always verify the tool actually installs cleanly
- Evaluate — always run at least 3 test prompts before delivering
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

