ClawCode

Your creative dev tool — AI coding Swiss Army knife

English | 简体中文

Quick Start • Why ClawCode • Features • Research • Documentation • Contributing

---

ClawCode is an open-source coding-agent CLI for Anthropic, OpenAI, Gemini, DeepSeek, GLM, Kimi, Ollama, Codex, GitHub Models, and 200+ models via OpenAI-compatible APIs. It goes beyond code generation — it's a self-improving engineering partner.

Why ClawCode

Typical AI Coding Tool	ClawCode
Suggestion-only chat	Terminal-native execution
One-shot answers	Self-improving learning loop
Single model, single thread	14-role virtual R&D team
No memory	Persistent sessions + experience capsules
Vendor lock-in	200+ models, fully configurable

Idea → Memory → Plan → Code → Verify → Review → Learned Experience

Features

⚡ Terminal-Native Execution

Analyze, code, verify, and review — all in one surface. No IDE overhead, no context switching.

clawcode                          # Interactive TUI
clawcode -p "Refactor this API"   # Non-interactive
clawcode -p "Summarize changes" -f json  # JSON output

🧠 Self-Improving Learning

ClawCode features ECAP (Experience Capsule) and TECAP (Team Experience Capsule) — a closed-loop learning system that turns every task into reusable knowledge:

• Instinct → Experience → Skill evolution chain
• Automatic write-back from /clawteam --deep_loop
• Portable, feedback-scored, privacy-controlled capsules

🎨 Design Team (/designteam)

Spin up specialist design agents (research, IXD, UI, product, visual) and ship structured design specs — not just "chatty UI suggestions."

👥 Virtual R&D Team (/clawteam)

Orchestrate 14 professional roles in one command:

Role	Focus
Product Manager	Priorities, roadmap
System Architect	Architecture, tech choices
Backend / Frontend / Mobile	Implementation
QA / SRE	Quality, reliability
DevOps / Team Lead	CI/CD, decisions

/clawteam "Build a REST API with auth"           # Auto-assign roles
/clawteam --deep_loop "Design microservice arch" # Convergent iteration

🔬 Research Mode (clawcode research)

Multi-phase investigation workflows with tool-backed evidence collection:

Workflow	Purpose
deepresearch	Template-driven: plan → research → verify → deliver
peerreview	Critical review with verification
lit	Literature survey
audit	Inspect URL/repo/artifact

Research tools: research_web_search (Firecrawl/Tavily/Parallel), research_paper_search (arXiv/Semantic Scholar), research_fetch_url, research_sandbox_exec, research_code_audit.

clawcode research start "Quantum error correction" --workflow deepresearch -o ./outputs/qec
clawcode research list-prompts  # View available templates

🔧 44 Built-in Tools

Category	Examples
File I/O	view, write, edit, patch, grep
Shell	bash, terminal, execute_code
Browser	browser_* (×11 automation tools)
Agent	Subagent spawning with isolation
Integration	MCP, Sourcegraph, Desktop automation
Research	research_* (web, papers, audit)

🔄 Claude Code Compatible

Migration-friendly: supports .claude/agents/, Claude-style tool names, plugin/skill systems, and familiar slash workflows.

Quick Start

1. Install

cd clawcode
python -m venv .venv
.\.venv\Scripts\Activate.ps1   # Windows
pip install -e ".[dev]"

Requirements: Python >=3.12

2. Configure

Create .clawcode.json in your project root:

{
  "providers": {
    "openai": {
      "api_key": "sk-...",
      "disabled": false
    }
  },
  "agents": {
    "coder": {
      "model": "gpt-4o",
      "provider_key": "openai"
    }
  }
}

Or use environment variables:

export CLAWCODE_OPENAI__API_KEY="sk-..."

3. Run

clawcode -c "/path/to/project"     # Interactive TUI
clawcode -p "Refactor this API"    # Non-interactive

Documentation

Topic	Link
Architecture	docs/architecture.md
Agent & Team Orchestration	docs/agent-team-orchestration.md
ECAP/TECAP Learning System	docs/ecap-learning.md
Slash Commands Reference	docs/slash-commands.md
Configuration Guide	docs/clawcode-configuration.md
Performance & Testing	docs/clawcode-performance.md
Research Mode	docs/research_mode.md

---

Research Mode

ClawCode includes an independent research subcommand for multi-phase investigation workflows with tool-backed evidence collection.

Workflows

Workflow	Command	Description
deepresearch	clawcode research start "topic" -w deepresearch	4-phase template: plan → research → verify → deliver
peerreview	clawcode research start "topic" -w peerreview	Critical review: review → verify → deliver
lit	clawcode research start "topic" -w lit	Literature survey
audit	clawcode research audit <url>	Inspect URL/repo/artifact
compare	clawcode research start "topic" -w compare	Side-by-side comparison

Research Tools

• research_web_search — Web search with Firecrawl/Tavily/Parallel (DuckDuckGo fallback)
• research_paper_search — arXiv + optional Semantic Scholar
• research_fetch_url — Fetch and extract page content
• research_sandbox_exec — Shell execution in sandbox
• research_code_audit — Compare claims against repo code

Quick Examples

# Deep research with Markdown template phases
clawcode research start "Quantum error correction" --workflow deepresearch -o ./outputs/qec
Peer review a proposal
clawcode research start "Review: LLM scaling laws" --workflow peerreview -o ./outputs/review
Audit a repository
clawcode research audit https://github.com/example/repo -o ./outputs/audit
List available templates
clawcode research list-prompts
Validate config without calling LLM
clawcode research start "Test" --dry-run

Configuration

Add to .clawcode.json:

{
  "web": {
    "backend": "firecrawl",
    "firecrawl_api_key": "YOUR_KEY",
    "tavily_api_key": "YOUR_KEY",
    "parallel_api_key": ""
  },
  "research": {
    "enabled": true,
    "s2_api_key": "YOUR_SEMANTIC_SCHOLAR_KEY",
    "subagents": { "max_concurrent": 3 }
  }
}

Optional API keys:

• Firecrawl — Enhanced web search/extraction
• Tavily — Research-optimized search
• Parallel — Alternative search backend
• Semantic Scholar — Higher rate limits for papers

Without API keys, research falls back to DuckDuckGo (web) and arXiv (papers, no key required).

---

Test Results

Suite	Tests	Status
Unit + Integration	833	✅
CLI Flags	22	✅
TUI Interactions	27	✅
Real Skills + Plugins	53	✅

Total: 944 items. 935 passed, 9 skipped, 0 failed.

Tiered Onboarding

Level	Time	Steps
Run it	~5 min	Install → clawcode -p "..." → try /clawteam
Close the loop	~30 min	Real task → /clawteam --deep_loop → inspect write-back
Team rollout	Repeatable	Align model → inventory skills → wire ECAP feedback

Contributing

pytest
ruff check .
mypy .

For larger design changes, open an issue first.

Security

AI tooling may run commands and modify files. Use ClawCode in a controlled environment, review outputs, and apply least privilege.

License

GPL-3.0 license.

---

Built by DeepElementLab