Claw Code
github.com/instructkr/claude-code- Category
- AI Agents
- Rank
- No. 89Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- @satyawani_aman
- GitHub
- 195.2k stars
- Date
About
A Python/Rust rewrite of Claude's agent harness system for building AI tool workflows. Created as a clean-room implementation after Claude's code was leaked, focusing on harness engineering and agent orchestration patterns.
What it does
Claw Code is a terminal-based agent harness that accepts one-off prompts or runs interactive sessions. It manages authentication, model access, tools, configuration, and stored session transcripts. A companion Python workspace maps commands and tools, routes prompts, simulates runtime branches, and audits porting parity.
Why it's ranked here
This is most useful as an inspectable engineering artifact, not a production recommendation. The repository exposes practical harness concepts, diagnostics, session behavior, and parity work, but its own maintainers call it a museum exhibit and direct serious workloads to other projects.
What's good
The command-line surface covers prompts, interactive sessions, configuration, status, and health checks. The health check validates API credentials, model access, and tool configuration. Documentation includes explicit PowerShell instructions, binary locations, credential-free smoke checks, container guidance, and warnings about a misleading package name.
Tradeoffs
Installation requires building the repository from source, and the similarly named public package installs a deprecated stub instead. Authentication requires provider API keys rather than a Claude subscription. ACP and Zed integration has no daemon or JSON-RPC entrypoint. Several Python subsystems are archive-backed placeholders, not complete implementations.
How to use it well
Use it to study agent harness structure, command and tool routing, session persistence, parity auditing, or cross-platform command-line setup. It suits engineers comfortable building Rust workspaces and inspecting companion Python models. Do not choose it for production workloads or working ACP and Zed integration.
Technical notes+
README.md identifies rust/ as the canonical Rust workspace and documents a build-from-source claw binary. src/main.py implements the companion Python CLI with inventory listing, prompt routing, bootstrap reports, turn loops, transcript persistence, remote-mode simulations, and parity audits. src/__init__.py exports the Python workspace’s runtime, query engine, session, manifest, command, and tool surfaces. src/QueryEngine.py wraps PortRuntime.route_prompt for Markdown route reports. src/cli/__init__.py, src/vim/__init__.py, src/buddy/__init__.py, src/hooks/__init__.py, src/state/__init__.py, src/types/__init__.py, src/utils/__init__.py, src/voice/__init__.py, src/bridge/__init__.py, src/memdir/__init__.py, and src/remote/__init__.py are metadata-driven placeholders for archived subsystems.
Observed
- Primary implementation
- Rust workspace producing the claw CLI binary
- Companion workspace
- Python reference and audit helpers, not the primary runtime
- Installation
- Build from source with Cargo; the similarly named crates.io package is a deprecated stub
- Interface
- Command-line interface supporting one-off prompts and interactive sessions
- Platform support
- Documented paths for macOS, Linux, and Windows PowerShell
- Editor protocol support
- No ACP/Zed daemon or JSON-RPC entrypoint
- Authentication
- Provider API keys are required; Claude subscription login is unsupported
Read from README.md, src/main.py, src/__init__.py, src/QueryEngine.py, src/cli/__init__.py, src/vim/__init__.py, src/buddy/__init__.py, src/hooks/__init__.py, src/state/__init__.py, src/types/__init__.py, src/utils/__init__.py, src/voice/__init__.py, src/bridge/__init__.py, src/memdir/__init__.py, src/remote/__init__.py.
What it can do
Orchestrate AI agent workflows
AI agent configuration and workflow definitions → Executed agent workflow with coordinated tasks
Build AI tool integration pipelines
Tool specifications and integration requirements → Connected AI tool workflow system
Manage agent harness operations
Agent harness configuration and control parameters → Managed agent execution environment
Execute multi-step AI workflows
Workflow steps and AI model instructions → Completed workflow results and outputs
Coordinate between multiple AI agents
Multiple agent definitions and coordination rules → Synchronized multi-agent execution results
Tags
Featured in Intel
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.