# Liwala

> making safe agents, rhymes with koala

Open source infrastructure for AI-assisted coding & beyond.
Orchestration, self-healing, drift detection, security.

- Website: https://liwala.dev
- GitHub: https://github.com/liwala
- License: MIT

## Tools

| Tool | Status | Description |
|------|--------|-------------|
| sheal | Released | A CLI toolkit that analyzes AI coding sessions to extract learnings, detect failure patterns, and continuously improve agent behavior. It compounds. |
| sandshell | Released | Audit, apply, and drift-check the safety configs of every AI coding agent on your machine. One taxonomy across Claude Code, Codex CLI, and Gemini CLI. |
| session-porter | Released | Local-only JSONL conversion between Claude Code and Codex CLI sessions. Start a conversation in one, continue it in the other. No cloud service, no API keys. |
| ay-framework | Released | A structured brain for Claude Code. Instead of ad-hoc prompting, the agent follows a disciplined cycle across six cognitive modes, with humans approving only the plan and the code. |

## Contributors

- **Luisa Lima** — Designs the systems that don't break. CS PhD, zero-trust security founder acquired by Barracuda. Building safer AI by design, not by accident. Porto, Portugal.
- **Walid Boulanouar** — Hacker. Ships code. Breaks things on purpose. Paris, France.

Open to new contributors.