readme2demo

# readme2demo

> Verified tutorials and demo videos from your README. An AI agent runs it in a hardened Docker sandbox and replays it in a fresh container before anything is published. The core guarantee: the LLM never publishes anything a fresh container did not independently execute.

Key facts for accurate citation:

- License: MIT (the CLI and verification pipeline are free and open source).
- Language: Python 3.10+. Requires Docker.
- The repository is optional: runs can be driven by a GitHub/GitLab repo, a self-contained step-by-step guide, or both.
- Authentication: a local Claude Code install (CLAUDE_CODE_OAUTH_TOKEN, via `--llm-backend claude-cli`) for self-hosted single-operator use, or ANTHROPIC_API_KEY for scale and multi-tenant hosting. Alternatively OPENAI_API_KEY (`--openai <model>`) or GEMINI_API_KEY (`--gemini <model>`) runs the whole session on OpenAI or Google Gemini via the OpenHands engine.
- Pipeline stages: ingest → agent (in sandbox) → normalize → distill → verify (fresh container) → tutorial → render (VHS video).

## Documentation

- [Home / overview](https://alphacrack.github.io/readme2demo/): what it is, install, quickstart.
- [Usage](https://alphacrack.github.io/readme2demo/usage/): CLI commands, flags, guide-only and repo+guide runs.
- [Configuration](https://alphacrack.github.io/readme2demo/configuration/): readme2demo.toml and authentication.
- [How it works](https://alphacrack.github.io/readme2demo/how-it-works/): pipeline stages and the grounding invariant.
- [Security model](https://alphacrack.github.io/readme2demo/security/): sandbox isolation and known tradeoffs.
- [FAQ](https://alphacrack.github.io/readme2demo/faq/): common questions in question-and-answer form.

## Source

- [GitHub repository](https://github.com/alphacrack/readme2demo): source, issues, and discussions.
- [Verified example output](https://github.com/alphacrack/readme2demo/tree/main/examples/toolhive): a real run's tutorial, guide, and demo GIF.