tap-in

name	tap-in
description	Initialize session with context briefing
context	fork
allowed-tools	Bash, Read, Glob, Write

Tap In - Session Initialization

Read recent activity and current state to get oriented at session start.

Workflow

1. Get Current Date

date +%Y-%m-%d

---

2. Check if Daily Housekeeping Needed

head -1 /root/projects/tmux/DaysActivity.md 2>/dev/null

• If date doesn't match today: Run /daily-housekeeping first
• STOP here until housekeeping completes
• Then resume tap-in

---

3. Read Recent Activity

Last 2-3 entries from DaysActivity.md:

• Note open work items
• Note recent state/issues
• Identify continuity threads

head -80 /root/projects/tmux/DaysActivity.md

---

4. Read CurrentStatus.md

cat /root/projects/tmux/CurrentStatus.md

Get operational context.

---

5. Check Open Beads

bd list --status open

---

6. Check CM Search Substrate

Query CM health to distinguish three failure modes that all look the same to a caller: (a) server down, (b) server up but empty index, (c) server up but stale index.

# Port comes from factory.env CM_PORT; Zgent = 3002.
CM_PORT=$(grep -E '^CM_PORT=' /root/projects/tmux/factory/factory.env 2>/dev/null | cut -d'"' -f2)
CM_PORT=${CM_PORT:-3002}
curl -s --max-time 3 "http://localhost:${CM_PORT}/api/v1/health" 2>/dev/null

Then classify:

• No response / non-2xx → CM server down. Flag: "memory search unavailable — start with systemctl status claude-monitor."
• conversationCount missing from response → server is pre-patch (old schema). Flag: "CM health endpoint predates co-s5e patch; apply factory/patches/claude-monitor/001-health-endpoint-db-stats.patch for richer signals."
• conversationCount == 0 → DB is empty. Flag: "CM index is empty — run cd /root/projects/claude-monitor && bun run backfill to seed."
• lastEntryTimestamp > 24h old → backfill stale. Flag: "CM index hasn't absorbed new content in >24h. Check /tmp/cm-backfill.log and the crontab."
• Otherwise → CM is healthy; no warning needed.

Include any warning in the Current State section of the briefing.

---

7. Output Session Briefing

Write to: /root/projects/tmux/session-briefing.md

## Session Briefing - YYYY-MM-DD HH:MM

---

### Recent Activity

**Last Session**: [timestamp] - [brief summary from most recent handoff]

**Open Work (carried forward)**:
- [item 1]
- [item 2]

---

### Current State

[Summary from CurrentStatus.md — version, rigs, attention items]

---

### Open Beads (active/recent)

| Bead | Title | Type |
|------|-------|------|
| id | title | type |

---

### Resumption Guidance

**Carried forward from last session**:
1. [specific next step]
2. [specific next step]

---

### Ready Status

[Ready to proceed | Issues require attention]

---

Pairs With

• /handoff - Session end
• /daily-housekeeping - Runs before tap-in if date changed

Re-run Anytime

This skill can be invoked mid-session to refresh context:

/tap-in

Removed, deliberately

Agent-teams detection and the Team State Analysis step. The first step used to run claude config get experimental.agentTeams. That call hangs on Claude Code 2.1.216 — it never returns, even with stdin closed (verified: 20s timeout, rc=124) — so it burned the Bash timeout on every session start. Its only consumer was a Team State Analysis step that ran jq against .beads/issues.jsonl, a file that does not exist because the beads store is Dolt-backed. Nothing ever set the flag the team-aware output was gated on, so that output was unreachable too. A hang feeding a no-op. Removed 2026-07-21 (co-kavq). /handoff carried the same call — check there before reintroducing anything like it.