🧠 GitHub Skill & Memory Discovery: Full Content for Comparison

Overview

GitHub Skill & Memory Discovery finds every Claude, AI agent, and coding-agent skill or memory repo for a given domain on GitHub, and extracts the full text of each skill-defining file (SKILL.md, CLAUDE.md, AGENTS.md, .cursorrules, mcp.json), plus maturity signals (stars, last commit). Give it "sales outreach" or "graphic design for brands" and get back the actual content, not just metadata, so a human or an LLM can compare candidate skills side by side and pick or synthesise the best one.

Built lean and HTTP-only against GitHub's official Search, REST, and raw-content APIs. Ideal for AI agent builders, skill authors, prompt engineers, and Claude teams that need to survey the state of the art on a topic before writing their own skill.

Reliability posture: blocked, empty, or failed extractions are never charged. Charging only happens after a repo's metadata, skill files, and README have been fetched successfully and the record is on its way to your dataset.

✅ No browser | ✅ Two search modes | ✅ Full file content | ✅ MCP-ready for AI agents

Features

Discovery + extraction in one run, both repo search and code search. Full text of every skill file, not just metadata, for real comparison. Repo maturity signals: stars, last commit date, and match technique. Optional githubToken unlocks precise code search across skill files. Same output whether you consume it via CSV, JSON, Excel, API, or MCP tool call.

How it works

Finding the right skill or agent config for a domain is a survey problem. There are usually a dozen or more candidate repos on GitHub, and reading each one manually is slow. This actor combines two GitHub search modes: code search (finds repos whose skill files actually mention your topic, requires a token) and repository search (finds repos whose name, description, or README mention your topic, works without a token). Code search results are ranked first because they are more precise.

For every match, the actor pulls the full text of any file matching the skill patterns above, plus the README as fallback context, plus the maturity signals a reviewer would look at first. Deciding which repo is best or synthesising a new skill from the strongest pieces is a judgment call, and that is exactly what your downstream LLM or reviewer is set up to do with this data.

🧾 Input configuration

{
  "query": "sales outreach",
  "maxRepos": 15,
  "githubToken": null
}

📤 Output format

This is a real record captured from a live run (query: "claude code skill"), with the readme field shortened for space (the actual dataset row carries the full text):

{
  "owner": "vinta",
  "repo": "awesome-python",
  "fullName": "vinta/awesome-python",
  "url": "https://github.com/vinta/awesome-python",
  "description": "An opinionated list of Python frameworks, libraries, tools, and resources",
  "matchedVia": "repo-search",
  "stars": 308221,
  "updatedAt": "2026-07-15T04:01:20Z",
  "lastCommitAt": "2026-07-14T07:27:15Z",
  "skillFiles": [
    {
      "path": "CLAUDE.md",
      "content": "# CLAUDE.md\n\n## Repository Overview\n\nAn opinionated list of Python frameworks, libraries, tools, and resources. Published at [awesome-python.com](https://awesome-python.com/).\n\n## Entry Guidelines\n\n**Refer to [CONTRIBUTING.md](CONTRIBUTING.md)** for acceptance criteria, quality requirements, rejection rules, and entry format...\n\n## Key Rules\n\n- Alphabetical ordering within categories is mandatory.\n- Quality over quantity. Only \"awesome\" projects.\n- One project per PR.\n- README.md is the single source of content truth."
    },
    {
      "path": "AGENTS.md",
      "content": "# AGENTS.md\n\n## Repository Overview\n\nAn opinionated list of Python frameworks, libraries, tools, and resources. Published at [awesome-python.com](https://awesome-python.com/).\n\n## Entry Guidelines\n\n**Refer to [CONTRIBUTING.md](CONTRIBUTING.md)** for acceptance criteria, quality requirements, rejection rules, and entry format...\n\n## Key Rules\n\n- Alphabetical ordering within categories is mandatory.\n- Quality over quantity. Only \"awesome\" projects.\n- One project per PR."
    }
  ],
  "readme": "# [Awesome Python](https://awesome-python.com/)\n\nAn opinionated guide to the best Python frameworks, libraries, tools, and resources.\n\n**Visit the [website](https://awesome-python.com/) to search and filter projects more easily.**\n\n## Categories\n\n**AI & ML**\n\n- [AI and Agents](#ai-and-agents)\n- [Deep Learning](#deep-learning)... (truncated here, full README text is delivered in the dataset row)",
  "charged": false,
  "scraped_at": "2026-07-15T04:15:58.085Z"
}

Every skill-repo record contains these fields:

Field	Description
👤 owner	Repo owner
📦 repo	Repo name
🏷️ fullName	owner/repo
🔗 url	Canonical GitHub URL
📝 description	Repo description
🎯 matchedVia	code-search or repo-search
⭐ stars	Star count
🔄 updatedAt	Last update datetime
⏰ lastCommitAt	Last commit datetime on default branch
📄 skillFiles	Array of { path, content } for every matched skill file
📖 readme	Full decoded README text
💰 charged	Whether the pay-per-event charge for this record went through
🕒 scraped_at	ISO timestamp of capture

💼 Common use cases

Building a new Claude or AI agent skill Survey the state of the art on your topic in ten minutes instead of a week. Feed the extracted content into an LLM to synthesise a stronger skill.

Prompt engineering research Compare how different authors structure a skill file for the same problem. Reuse tested patterns and avoid known bad structures.

Team standardisation Benchmark your team's SKILL.md and AGENTS.md conventions against public best practice. Build a private catalog of vetted skills for your organisation.

Landscape and market intelligence Map every public AI agent skill in a category with maturity signals. Track which skill topics are gaining author attention over time.

🚀 Getting started

1. Open the actor and set query to your domain in plain words (e.g. "graphic design for brands").
2. Optionally paste a githubToken to unlock precise code search.
3. Set maxRepos to cap how many candidates you extract (default 15).
4. Click Start, then download the dataset as JSON, CSV, or Excel, or pull it via API or MCP.
5. Feed the result to Claude or ChatGPT for side-by-side comparison and synthesis.

FAQ

Do I need a GitHub token? Not required. Without a token you get repo-search results (README and name matches). With a token you also get code-search results (matches inside skill files themselves), which is a much stronger signal.

How is this different from GitHub's own search? GitHub search returns links to repos. This actor returns the actual content of every matched skill file, ready for an LLM to read and compare.

Why not score and rank the results here? Judging which skill is best depends on the exact use case and taste. Raw extracted content lets an LLM reviewer make the call in context, which is more useful than a rigid formula baked into the actor.

How much does it cost? $5 per 1,000 repos extracted. You pay only for repos actually extracted and returned. There is no free tier.

Can I use it inside an AI agent? Yes. It is exposed as an MCP tool. See below.

Use in Claude, ChatGPT & any MCP agent

https://mcp.apify.com/?tools=themineworks/github-skill-discovery

Or call it programmatically with the Apify client:

import { ApifyClient } from 'apify-client';

const client = new ApifyClient({ token: 'YOUR_APIFY_TOKEN' });

const run = await client.actor('themineworks/github-skill-discovery').call({
  query: 'sales outreach',
  maxRepos: 15,
});

const { items } = await client.dataset(run.defaultDatasetId).listItems();
console.log(items);

🛠️ Complete your OSS intelligence pipeline

Pair this actor with the rest of the themineworks GitHub suite:

• GitHub Repo Intelligence: pull deep metadata for the repos you shortlist.
• GitHub Trending Scraper: discover new skill-authoring projects gaining momentum.
• RAG Website Crawler: turn any project's docs site into embedding-ready Markdown.

Typical flow: github-skill-discovery gathers the candidate skills, github-repo-intelligence checks their maturity, and an LLM does the synthesis into your own tuned version.

Questions or need a custom field set? Reach out through the Apify profile.