edda-conductorlisted

# Edda Conductor: Multi-Phase Plan Orchestration Orchestrate multi-phase AI plans using `edda conduct`. Each phase spawns a Claude Code agent with specific instructions and automated checks. ## When to Use - User asks to run a multi-step plan (e.g., "build the API project") - Heartbeat detects active conductor state - User asks about plan status or progress - A phase fails and needs a decision (retry, skip, abort) - User wants to create a plan.yaml from a description ## 1. Start a Plan ### From an existing plan.yaml ```bash # Interactive (foreground, shows live output) edda conduct run plan.yaml # Background (for unattended execution) # WARNING (decision fleet.lane-launch, edda repo): on Windows, nohup and # Start-Process do NOT escape the session's Windows Job Object — the child # still dies with the session. Launch unattended runs via Task Scheduler # instead (e.g. the tracked `scripts/fleet/lane-launch.ps1`). The nohup form # below is POSIX-only: there is no Job Object on Linux/macOS, so detaching # with nohup is enough there. nohup edda conduct run plan.yaml --quiet > /dev/null 2>&1 & ``` ### Verify it started ```bash edda conduct status --json ``` ### Report to user ``` Plan started: [name] Phases: [N] total Running in background. I'll monitor progress via heartbeat. ``` ### Important flags - `--quiet`: Suppress live agent output (use for background) - `--dry-run`: Preview plan without executing (always use before first run) - `--cwd <path>`: Override worki