A collective of people making AI agents safer.
Open-source infrastructure for AI-assisted coding — and beyond.

What we're building • Projects • Principles • Contribute

AI coding agents are powerful and unsupervised. They run shell commands, edit files, and ship code — often with safety primitives disabled by default and no memory between sessions.

Liwala builds the small, sharp tools that close that gap. Not frameworks. Not platforms. Plumbing.

• Make the sandbox actually on. Audit, apply, and drift-check the safety configs of every coding agent on your machine.
• End the amnesia. Pull learnings out of every session so your agent stops repeating the same mistakes.
• Move freely between tools. Carry your context from Claude Code to Codex CLI and back without losing the conversation.

Everything we ship is open source, local-first, and designed to compose with the agents you already use.

Your AI coding agents ship with sandboxes. sandshell makes sure they're actually on.

One taxonomy across Claude Code, Codex CLI, and Gemini CLI. Detect installed agents, audit safety findings by severity, apply safe defaults, and surface config drift between sessions.

detect · audit · apply · drift · verify

Your AI agent keeps making the same mistakes. sheal fixes that.

A CLI toolkit that analyzes AI coding sessions to extract learnings, detect failure patterns, and continuously improve agent behavior. Pre-session health checks, retrospectives, token cost dashboards, cross-session search — it compounds.

check · retro · cost · ask · browse

Move AI coding sessions between Claude Code and Codex CLI without losing context.

Local-only JSONL conversion between agents. Start a conversation in one tool, continue it in another. No cloud service. No API keys. Available on npm as @liwala/session-porter.

list · transfer · handoff · fork · continue

1. Small tools, sharp edges. One job per binary. No frameworks.
2. Local-first. Your sessions, your configs, your machine. We don't run a service.
3. Compose, don't capture. Built to work alongside Claude Code, Codex CLI, Gemini CLI — never to replace them.
4. Surface, don't decide. We make the unsafe-by-default visible. You stay in control of what gets applied.
5. Open source, default MIT. Every tool we ship is yours to read, fork, and run.

We're a small collective. Easiest ways in:

• ⭐ Star a project that solves a problem you have. It's the highest-signal feedback we get.
• 🐛 Open an issue with a reproduction. Failure cases are gold.
• 🧪 Write a learning. If you use sheal, share a learning that helped your agent — sheal learn add "..."
• 🤝 Send a PR. Each repo has a CONTRIBUTING.md. We try to review fast.

liwala.dev · All repos · sheal · sandshell · session-porter

_{Made with care by the Liwala collective. MIT licensed unless noted.}