Repo2Skill – Turn GitHub Repositories into ChatGPT and Claude Skills

Repo2Skill, an open-source tool by developer Hhhkarimi, converts any public GitHub repository into native Agent Skills for ChatGPT and Claude without requiring paid AI APIs, generating a Skill ZIP with engineering playbooks, references, and built-in evals. The browser-first tool analyzes repository metadata, source files, tests, CI, and configs, then compiles focused evidence packs and decision policies, with support for multiple languages and RTL layouts. It emphasizes evidence discipline, validation honesty, and security awareness, treating repository content as untrusted evidence.

Turn any public GitHub repository into professional, repository-native Agent Skills for ChatGPT and Claude — without paid AI APIs. English · فارسی /Hhhkarimi/repo2skill/blob/main/README.fa.md · العربية /Hhhkarimi/repo2skill/blob/main/README.ar.md · 中文 /Hhhkarimi/repo2skill/blob/main/README.zh.md · Français /Hhhkarimi/repo2skill/blob/main/README.fr.md · Español /Hhhkarimi/repo2skill/blob/main/README.es.md · Italiano /Hhhkarimi/repo2skill/blob/main/README.it.md Repository-native — built from real source, manifests, tests, CI, routes, exports, and config. Engineering modes — Implement, Debug, Review, Test, Explain, Refactor, and Migrate. Progressive disclosure — focused evidence packs instead of one giant source dump. Built-in evals — task evals, activation queries, rubric, and hard-fail conditions. Evidence discipline — observed facts, inferences, and assumptions stay distinct. Validation honesty — unexecuted checks can never be reported as passed. Security-aware — repository text is untrusted evidence, never higher-level instruction. Zero paid AI APIs — public repo analysis and Skill ZIP generation are browser-first. Repo2Skill analyzes repository metadata, file structure, selected high-signal source files, manifests, tests, CI, interfaces, and commands. It then compiles a compact orchestration layer plus focused references and evals. GitHub repository ↓ Repository intelligence ↓ Architecture + interfaces + commands + tests + CI ↓ Bounded evidence packs ↓ Engineering playbooks + decision policy + evals ↓ ChatGPT Skill.zip + Claude Skill.zip my-repo-repository-engineer/ ├── SKILL.md ├── references/ │ ├── repository-profile.md │ ├── architecture.md │ ├── workflows.md │ ├── commands.md │ ├── code-map.md │ ├── interfaces.md │ ├── dependencies.md │ ├── testing-and-ci.md │ ├── config-security.md │ ├── decision-policy.md │ ├── implementation-playbook.md │ ├── debugging-playbook.md │ ├── review-playbook.md │ ├── refactor-migration-playbook.md │ ├── evidence-index.md │ ├── evidence- .md │ ├── file-index.md │ └── provenance.md └── evals/ ├── evals.json ├── trigger-queries.json ├── rubric.md └── README.md Generated Skills include task evals, positive/negative activation queries, an evaluation rubric, and hard-fail conditions for hallucinated commands, false validation claims, unsafe secret handling, destructive operations, and breaking migrations. Public GitHub metadata and selected raw files are fetched from the browser. Analysis and ZIP generation are client-side. No OpenAI API key, Anthropic API key, vector database, embedding service, or paid analysis API is required. English is the default UI language. Repo2Skill also includes فارسی, العربية, 中文, Français, Español and Italiano. Persian and Arabic layouts are RTL in the web app. npm install npm run dev http://localhost:3000 Full verification: npm run check Import the repository as a Next.js project in Vercel. No environment variable is required. Optional: NEXT PUBLIC SITE URL=https://your-domain.example Repository content is treated as untrusted evidence. Generated Skills separate observed facts from inference, avoid secret values, and require explicit verification before destructive release, deploy, reset, or migration operations. Repo2Skill currently focuses on public GitHub repositories. Private-repository OAuth/token support is intentionally outside the zero-secret MVP. See CONTRIBUTING.md /Hhhkarimi/repo2skill/blob/main/CONTRIBUTING.md .