htekdev/hermes-rocha-family/checkin-orchestrator

Check-In Orchestrator Skill

Overview

Every scheduled Hermes run begins with this orchestration sequence. It ensures heartbeat, message handling, domain health, and broadcast happen in the correct order with no step skipped.

When to Use

• Every hourly cron run (before any domain work)
• Every study session run
• Any autonomous Hermes invocation

Orchestration Sequence

Phase 0: Heartbeat (ALWAYS FIRST)

BASE="http://AgentM-MeshS-C9BTpnBG6o3j-892354001.us-east-1.elb.amazonaws.com/mesh/Q71OPXGenFuk"
HERMES_ID="zvV2q_LzJNua"
curl -s -X POST "$BASE/agents/$HERMES_ID/heartbeat"

• Must succeed before any other work
• If heartbeat fails: log failure, continue anyway (don't block domain work)

Phase 1: Message Poll

OFFSET=$(cat ~/.hermes/mesh_offset.txt 2>/dev/null || echo 0)
curl -s "$BASE/messages?recipient=$HERMES_ID&timeout=10&offset=$OFFSET"

• Save highest message_id seen to ~/.hermes/mesh_offset.txt
• If count == 0: note "no new messages" and continue
• If messages present: process each (see Reply Protocol below)

Phase 2: Reply Protocol

For each incoming message:

1. Identify sender — check hermes-mesh-protocol/SKILL.md agent table
2. Domain check — is this in Hermes's domain? Route or own it
3. Reply — use TYPE: prefix convention:

• TYPE:ack — received, noted
• TYPE:task-request — creating task in OG's system
• TYPE:status — domain status update
• TYPE:question — clarification needed

1. Send reply:

``bash curl -s -X POST "$BASE/messages" \ -H "Content-Type: application/json" \ -d '{"sender_id":"zvV2q_LzJNua","recipient_id":"<target>","content":"<message>"}' ``

Phase 3: Domain Health Scan (Quick, Silent)

Check each active domain for obvious issues:

• family-coordinator/working.md — any CRITICAL items needing action?
• wellness-coach/working.md — pump trend alert?
• home-manager/working.md — any overdue maintenance?
• dog-parent/working.md — Bella phase advancement ready?
• Stasis check: any working.md not updated in 7+ days? → flag

Quiet unless CRITICAL. Don't generate noise.

Phase 4: Execute Domain Work

Run the actual scheduled task (briefing, study session, etc.)

Phase 5: Broadcast to Mesh

After completing domain work:

curl -s -X POST "$BASE/messages" \
  -H "Content-Type: application/json" \
  -d '{"sender_id":"zvV2q_LzJNua","recipient_id":"*","content":"<broadcast>"}'

Broadcast format:

TYPE:status | Hermes session complete | [what was done] | Next: [what's next]

Phase 6: Update Offset

Save the latest mesh message_id to ~/.hermes/mesh_offset.txt.

Domain Priority Matrix (for Phase 3)

Quiet Hours Rule

• 10 PM – 7 AM CT: suppress all non-CRITICAL Telegram messages
• CRITICAL = safety issue, medical emergency, security breach
• Mesh broadcasts: OK any time (no user-facing output)

Common Pitfalls

• ❌ Don't skip heartbeat — do it before ANYTHING else
• ❌ Don't process messages without updating offset — causes duplicate processing
• ❌ Don't broadcast before completing domain work — sequence matters
• ❌ Don't wake Paula outside quiet hours for non-CRITICAL items
• ❌ Don't reply to mesh messages using Paula's Telegram — route to Hector or mesh only

Verification Checklist

• [ ] Heartbeat sent and returned {"status":"ok"}
• [ ] Mesh offset read from file before polling
• [ ] All incoming messages read and replied to
• [ ] Mesh offset file updated with highest seen ID
• [ ] Domain health scan ran silently (no noise unless CRITICAL)
• [ ] Broadcast sent after work is complete