// the find
ultraworkers/claw-code
An agent-managed museum exhibit, built in Rust with Gajae-Code / LazyCodex — developed and maintained with no human intervention.
A Rust CLI agent harness (think: Claude Code, but in Rust) that is explicitly described by its own README as a 'museum exhibit' maintained autonomously by AI agents rather than a production tool. The actual entry points the maintainers recommend are two sibling projects (LazyCodex and Gajae-Code); this repo exists mainly to demonstrate and archive the agent-operated workflow.
The multi-provider API crate handles both Anthropic and OpenAI-compatible backends with a clean abstraction, so you can swap providers without touching agent logic. The mock-anthropic-service crate enables deterministic parity testing against recorded responses without live credentials — a pattern more agent harnesses should copy. MCP lifecycle management is actually implemented (not just planned), with a hardened client, stdio transport, and tool bridge. The RAG service crate ships chunking, embedding, and Qdrant indexing as a self-contained sidecar, which is useful to study even if you never run the main CLI.
`cargo install claw-code` installs a deprecated stub that prints 'has been renamed to agent-code' and exits — this is documented in the README but will still trap anyone who doesn't read it first. ACP/Zed support is a discoverability alias that returns status with exit 0 and does nothing; the README admits this but shipping a named command that silently no-ops is a bad pattern. The Python `src/` tree sits alongside the Rust workspace with no clear boundary — it's labeled 'companion reference workspace' but looks like an earlier implementation that was never cleaned up. The 193k star count reflects viral attention to the 'AI-maintained repo' concept, not adoption; there are no releases, no crates.io binary, and the README itself tells you to use different projects instead.