@murmurations-ai/cli

Command-line interface for the Murmuration Harness daemon.

Status: Phase 1A — start is the only functional command. Phase 1B adds status, stop (to a running daemon), and init.

Commands

| Command | Description | Status | | -------------------- | ------------------------------------------------------------------------------------------------------------- | ----------- | | murmuration start | Boot the daemon, register agents from the Phase 1A hardcoded registry, fire hello-world wake, wait for SIGINT | ✅ Phase 1A | | murmuration status | Print status of a running daemon | ⏳ Phase 1B | | murmuration stop | Send SIGTERM to a running daemon | ⏳ Phase 1B | | murmuration init | Run /init-murmuration interview skill | ⏳ Phase 6 |

Phase 1A usage

# From the monorepo root
pnpm --filter @murmurations-ai/cli start

# Expected output: JSON-lines log of boot, wake firing, completion
# Press Ctrl+C to shut down cleanly

What the Phase 1A daemon does

1. Constructs a SubprocessExecutor with a resolver that maps the hello-world agent to node examples/hello-world-agent/agent.js
2. Constructs a Daemon with a TimerScheduler and one registered agent
3. Schedules the hello-world agent with a delay-once trigger (2 seconds)
4. Starts the daemon; the wake fires; the subprocess runs and exits; the daemon logs the result
5. Stays in the foreground until SIGINT / SIGTERM

This proves the wake loop end-to-end: scheduler fires → executor spawns → subprocess runs → result logged → daemon idle but alive. It does not prove anything about real LLM calls, identity doc parsing, or GitHub integration — those are Phase 2+ work.