superclaw documentation
superclaw is a compact Go agent runtime built for clarity and control. It exposes a small number of well-defined tools, a curated set of skills, and a bounded execution loop that terminates predictably.
This documentation covers everything from installation to advanced Docker deployment. If you want to get something running immediately, start with Getting Started.
Core design principles
- →Not crawler-first. The architecture is plan → act → observe → stop. Fetching is one bounded tool.
- →Limited tool surface. 7 tools, each with a typed contract. No arbitrary tool execution.
- →Explicit skill set. 5 skills. No dynamic skill discovery, no plugin ecosystem.
- →Container-first isolation. Docker deployment is a first-class target, not an afterthought.
- →Auditable by default. Every run is logged to JSONL. Nothing is hidden.
Getting Started
Install superclaw, set your API key, and run your first agent task. From zero to running in under two minutes.
Configuration
All superclaw.json fields, CLI flag overrides, and configuration precedence. Customize model, step limits, tool allowlist, and more.
Tools
The 7 built-in tools: fetch_url, web_search, read_file, write_file, patch_file, list_files, run_bash. Input/output contracts and examples.
Skills
The 5 curated skills: summarize, research, extract, coding, github. How they work, when to use them, and how to activate them.
Docker
Hardened Docker setup with read-only rootfs, non-root user, capability drops, and workspace bind mount. Production-ready by default.
Session & History
How runs are persisted to .superclaw/runs.jsonl, the session schema, querying history, and retention policy.
Quick reference
7 Tools
- fetch_url
- web_search
- read_file
- write_file
- patch_file
- list_files
- run_bash
5 Skills
- summarize
- research
- extract
- coding
- github
Key defaults
- max_steps: 20
- timeout: 120s
- max_fetch: 5
- max_tokens: 4096
- model: claude-opus-4-5