Run Claude Code as an unattended agent. RPM and deb packages that sit next to
Anthropic's claude-code: a supervisor that holds an interactive session,
circuit breakers, an audit log, and an optional Matrix bridge so you can text it.
- Run it unattended.
claude-agent runholds a session and takes prompts over a local socket;claude-agent-matrixlets you text it. docs/run.md - Keep it in bounds. Breakers stop runaway loops, every tool call is audited, and an optional policy package locks the session down. docs/controls.md
- Build your own image. Two repo files and one install, on UBI, Fedora, Hummingbird, Debian or Ubuntu. docs/install.md
sudo dnf install claude-agent # repos: docs/install.md
CLAUDE_AGENT=1 claude-agent run &
claude-agent send --wait "Reply with exactly: pong"Hooks act only where CLAUDE_AGENT=1, so your own Claude Code is unaffected.
| Page | Covers |
|---|---|
| install.md | Repositories, example images, Quadlet |
| run.md | The supervisor, the CLI, texting by Matrix |
| controls.md | Breakers, audit, prompts off, the trentina policy |
python3 -m unittest discover -s tests
pre-commit run --all-filesAGPL-3.0-or-later