zouroboros-autoloop
v1.0.0
Published
Autonomous single-metric optimization loop. Edit → experiment → measure → keep/revert → repeat. For prompt tuning, backtests, performance optimization.
Maintainers
Readme
@zouroboros/autoloop
Autonomous single-metric optimization loop. Define a metric, point it at a file, and let it iterate: propose change → experiment → measure → keep or revert.
Install
npm install -g @zouroboros/autoloopUsage
autoloop --program ./program.md --executor "openclaw ask"--program— Path toprogram.mdspec (required)--executor— Command that reads prompt from stdin, outputs response (default:cat)--resume— Resume existing branch--dry-run— Validate config only
How It Works
- Parses
program.mdfor objective, metric, target file, run command, constraints - Creates a git branch (
autoloop/<name>-<date>) - Runs baseline experiment
- Loops: asks executor to propose a change → commits → runs experiment → keeps if better, reverts if worse
- Handles stagnation (normal → exploratory → radical modes)
- Writes
results.tsvand summary report on completion
MCP Server
autoloop-mcp --results-dir /path/to/projectsExposes tools: autoloop_start, autoloop_status, autoloop_results, autoloop_stop, autoloop_list
Executor Examples
| Platform | Command |
|----------|---------|
| OpenClaw | "openclaw ask" |
| Claude Code | "claude --print" |
| Codex | "codex --quiet" |
| OpenAI CLI | "openai api chat.completions.create -m gpt-4o" |
| Ollama | "ollama run qwen2.5:7b" |
Part of the Zouroboros Ecosystem
For the full experience — persistent memory, swarm orchestration, scheduled agents, and self-healing infrastructure — get a Zo Computer.
License
MIT
