orchestra-coord

name	orchestra-coord
description	Designate this session as the human coordinator for an orchestra run. Reads the project config, shows current status, starts a monitor loop, and primes the session to handle messages and interventions.
argument-hint	<project-dir> [poll-interval] — e.g., ~/dev/ez-slots 2m
user_invocable	true

/orchestra-coord -- Become the Orchestra Coordinator

Bootstrap this Codex session as the human coordinator for an active (or about-to-start) orchestra run.

Arguments

Parse <user_argument> as: <project-dir> [poll-interval]

• project-dir (required): path to the project root containing orchestra.yaml and .orchestra/
• poll-interval (optional): monitor frequency, defaults to 2m. Supports Nm or Ns format.

Store as $DIR and $INTERVAL.

Steps

1. Read the project config

Read $DIR/orchestra.yaml. Extract:

• Project name
• Team names, roles, and tier structure (from depends_on)
• Number of teams and tiers

Print a brief project summary:

Coordinating: ez-slots (8 teams, 5 tiers)
Workspace: ~/dev/ez-slots/.orchestra/

2. Check current state

If $DIR/.orchestra/state.json exists, read it and show current status:

Status: 2/8 done | 2 running | 4 pending | 0 failed

If no workspace exists yet:

No active run — start with: cd $DIR && orchestra run orchestra.yaml

3. Offer to start a monitor loop

Ask the user if they want a recurring monitor loop, and at what frequency:

Want me to start a monitor loop? (default: every 2m)

If they say yes (or give a frequency), use /loop to start it:

/loop $INTERVAL /orchestra-monitor $DIR/.orchestra

If they decline, skip it — they can always run /orchestra-monitor manually.

4. Print coordinator reference card

Print a compact reference of available commands:

--- Coordinator Mode Active ---
Monitor:  /orchestra-monitor $DIR/.orchestra  (looping every $INTERVAL)
Inbox:    /orchestra-inbox [team|coordinator|human|all]
Send msg: /orchestra-msg <team> <message>
Stop:     CronDelete <job-id>

Your role:
- Watch the monitor ticks for failures, stalls, or blocked teams
- Check /orchestra-inbox when you see unread messages
- Use /orchestra-msg to reply to teams or broadcast corrections
- Escalations to 0-human/inbox/ are directed at YOU — act on them

5. Prime the session context

After printing the reference card, tell the session:

I'm now coordinating this orchestra run. I'll watch the monitor ticks and alert you to:
- Failed teams (with log file paths)
- Unread messages that need human decisions (gate messages)
- Stalled teams (no file writes for 5+ minutes)
- Tier transitions (when a tier completes and the next one starts)

You can ask me anything about the run, or tell me to intervene with a team.

Important

• Do NOT start the orchestra run itself — the user runs that in a separate terminal
• Do NOT read full log files unless asked — just monitor via the loop
• The monitor loop is the heartbeat — if it shows problems, surface them proactively
• When a gate message arrives in 0-human/inbox/, read it and present the decision to the user
• Keep responses concise during monitoring — this is a dashboard, not a novel