Claw Code

start with the real crab-powered harnesses

github.com/code-yeongyu/lazycodex
github.com/Yeachan-Heo/gajae-code

Join the Discords: ultraworkers discord · gajae-code discord

Important

Claw Code is not the serious production project here. This repository is closer to a museum exhibit than a product pitch, a crustacean-run artifact kept alive by clawed gajaes, swept and labeled by agents, and automatically maintained according to the harnesses above.

As already described in the project philosophy, this is not meant to be hand-operated like a normal product repo. It is an agent-managed exhibit: the harnesses plan, execute, verify, label, and preserve the artifact while the crabs keep the tank running.

If you want to actually run work, start with LazyCodex or Gajae-Code. If you want to inspect the strange little fossil of the Claw Code moment, continue below.

For the longer public explanation behind this philosophy, see here.

ultraworkers/claw-code · Usage · Roadmap · Contributing · Security · UltraWorkers Discord

Claw Code is a command-line AI coding assistant (the clawcli CLI). It's written in Rust.

You need one tool to build it: Rust. Get it from https://rustup.rs/.

---

Install

In the project folder, run:

python3 install/install.py

That's it. The installer works the same on macOS, Linux, and Windows. It builds clawcli and puts it on your PATH.

Need Rust first? Install it from https://rustup.rs/, then open a new terminal and run the install command above.

Windows? Click here.

Use PowerShell:

python install\install.py

The binary is called clawcli.exe. The installer adds it to your PATH automatically.

---

Start it

After install, open a new terminal (so your PATH updates), then run:

clawcli

That opens the interactive assistant. To send one prompt and exit:

clawcli prompt "say hello"

Always-on Caveman responses

clawcli includes Caveman's default full communication style in its built-in system prompt. It keeps technical content intact while dropping filler, articles, and verbosity across interactive, one-shot, resumed, and model-switched sessions, so no /caveman command or skill installation is needed. Say normal mode or stop caveman to use normal prose.

Provider-bound user text, prior assistant text, thinking blocks, and plain tool results receive the same loss-aware compression before each request. Original session content stays intact. Code, commands, paths, URLs, identifiers, quoted text, structured payloads, tool arguments, signatures, and exact technical values stay unchanged. Signed provider thinking blocks also stay exact when their signature covers the text. Set CLAW_CAVEMAN=0 for an uncompressed compatibility run. Compare the three sample probes with ./scripts/caveman-fidelity.sh. The script reports a repeatable chars/4 token estimate and required-term answer fidelity; provider billing tokenizers may report different totals.

clawcli also includes an always-on Superpowers-style development workflow. For non-trivial changes it automatically applies focused discovery, brainstorming, planning, TDD, systematic debugging, review, and verification. Clear requests and small edits stay proportionate; no plugin installation or explicit skill invocation is required.

First run: connect an API

If no credentials are detected, clawcli opens a setup prompt. Choose an OpenAI-compatible or Anthropic-compatible connection, enter its base URL and API key, then enter the model name. The OpenAI-compatible option is selected by default and starts with https://api.openai.com/v1, so custom gateways, local servers, and hosted endpoints use the same flow.

The API key is entered without echo, saved to ~/.claw/.env (or $CLAW_CONFIG_HOME/.env) with user-only permissions, and the selected model is saved to settings.json. To configure it non-interactively, set the matching environment variables yourself:

# Any OpenAI-compatible service; the URL is optional for api.openai.com.
export OPENAI_BASE_URL="https://your-gateway.example/v1"
export OPENAI_API_KEY="..."

Or an Anthropic-compatible service:

export ANTHROPIC_API_KEY="sk-ant-..."

(Windows PowerShell: use $env:OPENAI_API_KEY = "..." instead of export.)

Then check that everything is wired up:

clawcli doctor

---

What the installer does

1. Builds the single clawcli binary from the Rust source in rust/.
2. Copies it into a bin folder on your PATH (~/.cargo/bin on mac/linux, the equivalent on Windows).
3. Runs clawcli --version to confirm it works.

Advanced options

python3 install/install.py --release        # optimized build (slower to compile)
python3 install/install.py --no-verify      # skip the version check
python3 install/install.py --install-dir X  # install to a specific folder
python3 install/install.py --no-path-update # don't change your PATH

Under the hood, install.py is a dispatcher that calls a per-OS backend in install/backends/ (macos.sh, linux.sh, windows.ps1). Each is also runnable on its own.

---

Troubleshooting

• command not found: clawcli — open a new terminal so the PATH change takes effect.
• cargo was not found — install Rust from https://rustup.rs/ first.
• Build is slow — that's normal for a first build (a few minutes). Add --release only if you want faster runtime.

---

Learn more

• USAGE.md — full command reference, auth, sessions, config
• rust/README.md — Rust workspace and crate details
• PARITY.md — port status
• docs/planning/ROADMAP.md — roadmap
• THIRD_PARTY_NOTICES.md — embedded dependency notices

Ownership / affiliation disclaimer

• This repository does not claim ownership of the original Claude Code source material.
• This repository is not affiliated with, endorsed by, or maintained by Anthropic.