// the find
ruvnet/open-claude-code
Nightly Claude Code CLI Decompile — Reverse Engineered & Rebuilt
A reverse-engineered reimplementation of Anthropic's Claude Code CLI, built by decompiling the published npm package and rebuilding from scratch. It mirrors the real CLI's architecture — async agent loop, 25 tools, MCP support, 6 permission modes — and ships as an npm package you can run with npx. Aimed at developers who want to study how Claude Code works internally or run a modifiable version.
The nightly CI pipeline is genuinely well-thought-out: it detects upstream releases, runs 900+ tests, and only publishes if all gates pass — so you're not chasing a stale fork. Multi-provider support (OpenAI, Gemini, Bedrock, Vertex) is a real differentiator over the official tool, which is Anthropic-only. The clean-room legal framing is careful and explicit, citing specific DMCA and EU directive provisions rather than just asserting 'it's fine'. The tool surface — 25 tools matching the real CLI, 4 MCP transports, settings chain with 4 priority layers — is closer to the real thing than most imitations get.
The claim of '100% functional' is marketing; there's no way to verify behavioral parity with the closed-source original, and the nightly sync means it's always one Anthropic refactor away from silent divergence. The decompilation basis means the architecture is inferred, not documented — bugs in the real tool are likely reproduced here, and new bugs introduced in reimplementation have no ground truth to diff against. The README buries the fact that this is JavaScript while the real Claude Code is TypeScript, which matters if you're trying to contribute or debug. 'Clean-room' is legally contested territory regardless of the cited statutes — production use in an enterprise context carries real risk that the repo glosses over.