Getting Started

Install superclaw, configure your API key, and run your first agent task.

Prerequisites

requiredGo 1.23 or later. superclaw uses the standard library HTTP stack and requires modern Go generics and range-over-func support. Download Go →
requiredAnthropic API key. superclaw calls Claude via the Anthropic API. Create an account and get a key at console.anthropic.com.
optionalDocker 24+. Required only if you want container-isolated runs. See the Docker guide.

Installation

Install the superclaw binary using go install. This downloads, compiles, and places the binary in $(go env GOPATH)/bin.

bash

go install github.com/akshaymemane/superclaw/cmd/superclaw@latest

Verify the installation:

bash

superclaw --version
# superclaw v0.3.0 (go1.23.4 darwin/arm64)

GOPATH not in PATH?

Add export PATH="$(go env GOPATH)/bin:$PATH" to your shell profile (~/.zshrc or ~/.bashrc).

Set your API key

superclaw reads ANTHROPIC_API_KEY from the environment. The key is never written to disk.

bash

# Add to your shell profile for persistence
export ANTHROPIC_API_KEY=sk-ant-...

# Or pass it inline for a single run
ANTHROPIC_API_KEY=sk-ant-... superclaw "your task here"

Your first run

Navigate to the directory containing your files and pass the task as a quoted string:

bash

cd /path/to/your/project
superclaw "read the README and write a 3-paragraph summary to summary.md"

You should see output like this:

text

superclaw v0.3.0
Loading default config (no superclaw.json found)
Task: read the README and write a 3-paragraph summary to summary.md

Step  1/20  read_file(README.md)               → 2.1 KB
Step  2/20  write_file(summary.md)             → 847 B

Done in 2 steps, 4.3s.
Session saved to .superclaw/runs.jsonl

With a config file

Create a superclaw.json in your project root to set persistent defaults, restrict the tool allowlist, or load a skill:

superclaw.jsonjson

{
  "model": "claude-opus-4-5",
  "max_steps": 10,
  "timeout_seconds": 60,
  "tools": ["read_file", "write_file", "list_files"],
  "skills": ["summarize"]
}

With this config, the agent can only use 3 tools and is limited to 10 steps and 60 seconds. CLI flags override these values — see Configuration for the full precedence rules.

Common CLI flags

Flag	Description	Example
-model	Override the model	-model claude-haiku-4-5
-max-steps	Override max steps	-max-steps 5
-skills	Load specific skills	-skills coding,github
-timeout	Override timeout (seconds)	-timeout 30
-config	Path to superclaw.json	-config ./custom.json
-dry-run	Print plan, no execution	-dry-run
-verbose	Show full tool I/O	-verbose

Next steps

Configuration reference →

Every superclaw.json field and CLI flag explained.

Tools reference →

What each of the 7 tools does and its input contract.

Skills guide →

When to use summarize, research, extract, coding, or github.

Docker deployment →

Run superclaw in a hardened container for production use.