System Enhancement: Activity/Handoff Lifecycle Management

Status: Proposed
Owner: Wadsworth 📋
Priority: High
Date: 2026-04-28

Problem

Subagent sessions go stale or cycle without completing handoffs. We have:
- Sessions timing out (Daedalus)
- Sessions running idle (Socrates — "standing by unless @mentioned")
- Handoff docs sitting unread in shared/
- No systematic tracking of dispatched work

This creates silent failures where Matt thinks work is in progress but nothing is happening.

Proposed Solution

Phase 1: Handoff Tracking System

Create memory/dispatches/ directory with structured dispatch records:

memory/dispatches/
  2026-04-28-imap-proxy-socrates.md    # Active dispatch
  2026-04-28-dashboard-css-daedalus.md # Stale dispatch
  DISPATCH-INDEX.md                    # Master list

Each dispatch record:

---
dispatch_id: "disp-20260428-001"
agent: "socrates"
task: "IMAP Proxy Integration"
handoff_doc: "shared/project-docs/handoffs/imap-proxy-phase6.2.md"
dispatched_at: "2026-04-28T20:58:00Z"
status: "pending"  # pending | acknowledged | in_progress | blocked | complete
last_check: "2026-04-28T21:56:00Z"
blocker: null
next_check: "2026-04-28T22:30:00Z"
---

Phase 2: Automated Health Checks

Cron job every 30 minutes:
1. Read DISPATCH-INDEX.md
2. For each active dispatch:
- Check if agent session is alive
- Check if files were modified after handoff
- Check if task keywords appear in agent memory
3. Update status: active | stale | blocked | dead
4. Alert Matt if:
- Dispatch stale > 2 hours
- Session dead with incomplete work
- Agent acknowledges but doesn't start within 1 hour

Phase 3: Session Lifecycle Rules

For Wadsworth (me):
- Before dispatching: check if agent already has active session
- Before accepting "standing by" — require explicit acknowledgment with ETA
- After dispatch: set check-in reminder at T+30min, T+2hr, T+24hr
- On stale detection: escalate to Matt with options (reassign, reprioritize, abort)

For Subagents:
- On receiving handoff: MUST acknowledge within 5 minutes with ETA or blocker
- MUST update dispatch record on status change
- Session timeout = automatic "abandoned" status

Phase 4: Integration with Existing Memory

• Hook into memory/HEARTBEAT.md — add dispatch checks to heartbeat
• Extend memory/YYYY-MM-DD.md with dispatch log section
• Link to shared/project-docs/handoffs/ for durable artifacts

Implementation Plan

1. Now: Create memory/dispatches/ + DISPATCH-INDEX.md
2. Today: Add cron job for dispatch health checks
3. This week: Update AGENTS.md with session lifecycle rules
4. Ongoing: Wadsworth checks dispatches at every heartbeat

Files to Create/Modify

• NEW: memory/dispatches/DISPATCH-INDEX.md
• NEW: memory/dispatches/2026-04-28-imap-proxy-socrates.md
• NEW: memory/dispatches/2026-04-28-dashboard-css-daedalus.md
• MOD: AGENTS.md — add "Inter-Agent Dispatch Protocol" section
• MOD: HEARTBEAT.md — add dispatch health check tasks
• MOD: memory/2026-04-28.md — document dispatch system

Acceptance Criteria

• [ ] Matt can ask "status of dispatches" and get real answer
• [ ] Stale dispatches auto-detected within 2 hours
• [ ] No handoff doc sits unread without escalation
• [ ] Agent session death = immediate alert, not silent failure

Notes

This is a meta-system improvement. It's about making the multi-agent setup actually work, not just theoretically work. The current pattern of "dispatch and hope" is brittle.

Trigger: Realized Socrates was idle for 1+ hour with unread handoff while Matt waited.
Priority: High — affects every future dispatch.