coalesce-labs/god

god — Cross-Project Omniscient Status View

Combines static state (where things are) with the live event log (how they got there and what's happening right now). Default invocation produces a full dashboard; args narrow the scope or switch interaction mode.

Invocation

/god                        — full multi-project status dashboard
/god <project>              — filter to one project (e.g., adva, catalyst)
/god <orch-name>            — deep-dive into one orchestrator
/god restart                — list crashed/interrupted sessions with restart commands
free-form question          — "what PRs are open?", "which workers are stuck?",
                              "what happened in the last hour?"

Data Sources

| Source | Path / Command | What it gives | |---|---|---| | Global orchestrator state | ~/catalyst/state.json | All orchestrators, progress, worker map | | Per-run state | ~/catalyst/runs/<orch-id>/state.json | Wave breakdown, queue, current wave | | Worker signal files | ~/catalyst/runs/<orch-id>/workers/*.json | Phase, PR, heartbeat, needsAttention | | Session database | catalyst-session.sh list --active --json | Active Claude sessions | | Worktree inventory | ls ~/catalyst/wt/ | Projects and their worktrees | | Claude session recency | ls -lt ~/.claude/projects/ | Which sessions were recently active (by mtime) | | Event log | ~/catalyst/events/YYYY-MM.jsonl | Last 30–60 min of activity | | Orchestrator dashboard | ~/catalyst/runs/<orch-id>/DASHBOARD.md | Human-readable orch summary | | Rollup fragments | ~/catalyst/runs/<orch-id>/workers/<TICKET>-rollup.md | Post-completion summaries |

Naming conventions

• ~/catalyst/wt/<project>/ — top-level project directories (e.g., adva/, catalyst-workspace/)
• orch-<slug> — orchestrator leader worktree (no ticket suffix)
• orch-<slug>-<TICKET> — orchestrator worker worktree
• PM / pm — persistent PM session worktree
• <TICKET> (e.g., ADV-454) — standalone oneshot worktree
• ~/catalyst/runs/<orch-id>/ — orchestrator run state dir (matches orch-<slug> above)
• ~/.claude/projects/-Users-ryan-catalyst-wt-<project>-<worktree>/ — Claude Code session data

Procedure

Step 0: Parse invocation mode

Determine mode from the argument:

• No argument → default dashboard
• Known project directory name (matches ls ~/catalyst/wt/) → project filter
• Starts with orch- → orchestrator deep-dive
• Exactly restart → restart mode
• Otherwise → free-form Q&A (treat the full input as a question)

Step 1: Run the data-gather helper

The helper script collects all state in one pass and outputs JSON:

plugins/dev/scripts/god-gather.sh 2>/dev/null

The script returns a JSON object with keys: ts, projects, sessions, recentEvents, global. Parse it with jq for the sections you need.

If the helper is unavailable or errors, fall back to gathering data directly (see Manual fallback commands below).

Step 2: Present the dashboard

Format the output using the Dashboard format section below. Use the gathered JSON.

For project-filter mode, skip all sections not matching the requested project. For orchestrator deep-dive, show the full worker table and the DASHBOARD.md contents. For restart mode, follow the Restart mode section. For free-form questions, answer naturally using the gathered data as context.

---

Dashboard Format

Use Unicode box-drawing characters for section headers. Present in this order:

[as of 2026-05-07T11:44:00Z]

═══ ADVA (2 orchestrators, 1 PM, 1 oneshot) ═══

ORCH  orch-deal-to-opportunity-2026-04-24    Wave 2/7   active
  ├── ADV-462  pr-created   PR #291 ✅ CI passing
  ├── ADV-474  implementing  (no PR yet)
  └── ADV-475  implementing  (no PR yet)

ORCH  orch-industry-packs-2026-04-24         Wave 1/2   active
  ├── ADV-480  pr-created   PR #293 ❌ CI failing
  ├── ADV-481  researching
  ├── ADV-482  queued
  └── ADV-483  queued

PM    ~/catalyst/wt/adva/PM                  last active: 2 min ago
SHOT  ADV-454 (design tokens in docs)        last active: 8 min ago

═══ CATALYST (1 orchestrator, 1 PM) ═══

ORCH  orch-ctl-275-2026-05-07                Wave 1/2   active
  ├── CTL-280  done         PR #465 ✅ merged
  └── CTL-193  implementing  (no PR yet)

PM    ~/catalyst/wt/catalyst-workspace/PM    last active: 3 min ago

─── Recent activity (last 30 min) ─────────────────────────────

  ADV-480: implementing → pr-created [PR #293, CI pending]
  CTL-280: pr-created → done [PR #465 merged]
  github: 3× check_suite completed (2 passing, 1 failing)
  Linear: ADV-474 state → In Review

Status icons

| CI state | Icon | |---|---| | passing / merged | ✅ | | pending / in-progress | ⏳ | | failing | ❌ | | no PR yet / unknown | (omit icon) |

Worker status values

Map signal-file status to display text:

• dispatched → queued
• researching → researching
• planning → planning
• implementing → implementing
• validating → validating
• shipping → shipping
• pr-created → pr-created (append PR info)
• done → done
• failed → ❌ failed
• stalled → ⚠️ stalled
• deploy-failed → ❌ deploy-failed

Workers with needsAttention: true get a ⚠️ prefix regardless of status.

---

Orchestrator Deep-Dive (/god <orch-name>)

When the arg matches an orchestrator name, show:

1. Header with orch ID, project, status, start time, elapsed
2. Wave breakdown table from per-run state.json:

   Wave 1/2  ✅ done   (CTL-275, CTL-276, CTL-277, CTL-278, CTL-279)
   Wave 2/2  🔵 active (CTL-280, CTL-193)
   

3. Full worker table with phase, PR, CI status, last heartbeat
4. Attention items if any
5. Contents of ~/catalyst/runs/<orch-id>/DASHBOARD.md (verbatim, truncated to 60 lines)
6. Recent events for this orchestrator (last 60 min, filtered by .orchestrator == "<orch-id>")

---

Restart Mode (/god restart)

Show sessions that appear interrupted (were active but have no running process):

Detection

# Find worker signal files with non-terminal status but stale heartbeat
find ~/catalyst/runs -name "*.json" -path "*/workers/*.json" 2>/dev/null | while read f; do
  jq -r --arg f "$f" \
    'select(.status != "done" and .status != "failed" and .status != null) |
     "\(.status) \(.lastHeartbeat // "unknown") \(.pid // "?") \(.ticket // .workerName) \($f)"' \
    "$f" 2>/dev/null
done

For each candidate:

1. Check if the PID is still running: kill -0 <pid> 2>/dev/null && echo "running" || echo "dead"
2. Calculate time since last heartbeat
3. Mark as crashed if: PID dead AND heartbeat > 15 min ago

For each crashed session, show the restart command:

⚠️  CTL-193 (implementing) — last heartbeat 42 min ago, PID 81220 dead
    Worktree: ~/catalyst/wt/catalyst-workspace/orch-ctl-275-2026-05-07-CTL-193
    Resume:   cd ~/catalyst/wt/catalyst-workspace/orch-ctl-275-2026-05-07-CTL-193
              claude --resume

⚠️  ADV-474 (planning) — last heartbeat 23 min ago, PID 79441 dead
    Resume:   cd ~/catalyst/wt/adva/orch-deal-to-opportunity-2026-04-24-ADV-474
              claude --resume

If no crashed sessions: print ✅ All active sessions appear healthy.

---

Free-Form Q&A

When the input is a question, gather the relevant data and answer naturally:

| Question type | Data to gather | |---|---| | "what PRs are open?" | gh pr list --repo <each-repo> or parse pr field from worker signal files | | "which workers are stuck?" | Worker signal files where needsAttention: true or status == "stalled" | | "what happened in the last hour?" | Event log tail, last 60 min | | "which orchestrators are done?" | state.json orchestrators where status == "completed" | | "how much did the last orch cost?" | state.json .orchestrators[<orch-id>].usage.costUSD | | "what's the CI status of PR #N?" | gh api repos/<repo>/pulls/<N> |

Always gather the specific data first, then answer. Do not speculate from memory.

---

Manual Fallback Commands

If god-gather.sh is unavailable:

# Global orchestrator state
cat ~/catalyst/state.json 2>/dev/null | jq '.orchestrators | to_entries[] | {
  id: .key,
  status: .value.status,
  projectKey: .value.projectKey,
  progress: .value.progress,
  worktreeDir: .value.worktreeDir
}' 2>/dev/null

# Worktree inventory
for project in $(ls ~/catalyst/wt/ 2>/dev/null); do
  echo "=== $project ==="; ls ~/catalyst/wt/"$project"/ 2>/dev/null
done

# Active sessions (if catalyst-session.sh is available)
SESS=$(ls ~/.claude/plugins/cache/catalyst/catalyst-dev/*/scripts/catalyst-session.sh 2>/dev/null | head -1)
[ -x "$SESS" ] && "$SESS" list --active --json 2>/dev/null | jq '.' || echo "session tracking unavailable"

# Recent Claude session activity (mtime of project dirs)
ls -lt ~/.claude/projects/ 2>/dev/null | grep "catalyst-wt" | head -15

# Recent events (last 30 min)
EVENTS_FILE=~/catalyst/events/$(date +%Y-%m).jsonl
if [ -f "$EVENTS_FILE" ]; then
  TOTAL=$(wc -l < "$EVENTS_FILE" | tr -d ' ')
  SINCE=$(( TOTAL > 2000 ? TOTAL - 2000 : 0 ))
  CUTOFF=$(date -u -v-30M +%Y-%m-%dT%H:%M:%SZ 2>/dev/null || \
           date -u --date='30 minutes ago' +%Y-%m-%dT%H:%M:%SZ 2>/dev/null)
  tail -n +"$((SINCE + 1))" "$EVENTS_FILE" 2>/dev/null | jq -c \
    --arg cutoff "$CUTOFF" '
    select(.ts >= $cutoff) | select(
      (.event | startswith("worker-")) or
      (.event | startswith("filter.wake")) or
      .event == "github.pr.merged" or
      .event == "github.check_suite.completed" or
      .event == "attention-raised" or
      .event == "linear.issue.state_changed" or
      (.event == "comms.message.posted" and (.detail.type // "") == "attention")
    )' 2>/dev/null
fi

# Per-run state (waves and worker list)
for STATE in $(find ~/catalyst/runs -name "state.json" -maxdepth 2 2>/dev/null); do
  jq '{run: .orchestrator, currentWave: .currentWave, totalWaves: .totalWaves, waves: [.waves[] | {wave: .wave, status: .status, tickets: .tickets}]}' "$STATE" 2>/dev/null
done

# Worker signal files for active orchestrators
find ~/catalyst/runs -name "*.json" -path "*/workers/*.json" -not -name "*-rollup.md" 2>/dev/null | \
  xargs jq -c '{ticket: .ticket, status: .status, phase: .phase, pr: .pr, needsAttention: .needsAttention, lastHeartbeat: .lastHeartbeat, pid: .pid}' 2>/dev/null

---

Related

• orchestrate — launches multi-ticket orchestrators whose state /god reads
• oneshot — launches standalone workers whose state /god tracks
• teardown — archives completed orchestrators (moves them out of ~/catalyst/runs/)
• [[monitor-events]] — event-log patterns and filter cookbook
• [[catalyst-filter]] — Groq-backed semantic event router (filter daemon)
• [[catalyst-comms]] — agent-to-agent pub/sub; attention posts surface in /god restart
• CTL-282 — HUD-panel / 30-minute briefing variant of this skill (child ticket)
• CTL-192 — session state tracking and crash-resilient restart (related)