health

Prompt file imported from SJParthi/tickvault (.claude/commands/health.md). Fill in {{TICKVAULT_MCP_PROFILE}}, {{TICKVAULT_PROM_STATUS}}, {{TICKVAULT_QDB_STATUS}}, {{TICKVAULT_GRAF_STATUS}}, {{TICKVAULT_API_STATUS}} before use. Copyright stays with the author.

Run the full tickvault automation + observability health check and report a single PASS/FAIL table. This is the one command an operator types to answer "is everything OK?" — works in any Claude Code session (fresh or existing, Mac local or AWS remote) with zero manual setup.

What to run (in this order — each gets its own row in the final table)

1. Runtime attach Source the session env so the MCP log tools resolve:

   [ -f .claude/.session-env ] && . .claude/.session-env
   echo "profile={{TICKVAULT_MCP_PROFILE}}  prom={{TICKVAULT_PROM_STATUS}}  qdb={{TICKVAULT_QDB_STATUS}}  graf={{TICKVAULT_GRAF_STATUS}}  api={{TICKVAULT_API_STATUS}}"
   

2. Static guards — scripts/validate-automation.sh 31 mechanical checks (compile gates, observability guards, file-level invariants, source-code invariants). Must be 31/31 PASS.

3. Full-system health — scripts/doctor.sh 7 sections: source invariants, Docker stack, service endpoints, zero-touch artefacts, live error signal, env vars, AWS readiness. On a Mac with make docker-up done, every section must be PASS (AWS readiness legitimately SKIPs when terraform isn't initialized).

4. 100% audit — scripts/100pct-audit.sh 41 P-category dimensions (mechanically provable), 4 R (runtime), 3 L (layered asymptotic), 3 I (impossible-absolute). Report only P + R counts; L + I are advisory per 100pct-audit-tracker.md.

5. Live ERROR tail — MCP tail_errors Call mcp__tickvault-logs__tail_errors with limit: 10. Report the count, plus the top code breakdown (e.g. I-P1-11×2, WS-GAP-03×1). count=0 is the healthy baseline.

6. Live summary — MCP summary_snapshot Call mcp__tickvault-logs__summary_snapshot. Report whether it exists and whether Zero ERROR-level events is in the body.

7. Novel signatures in the last hour — MCP list_novel_signatures Call with since_minutes: 60. Any novel signature = operator attention needed. Report count + top 3 (if any).

8. Metric pulse — CloudWatch metrics (prod) / app /metrics exporter The Prometheus container was removed in #O3 and the prometheus_query MCP tool was retired in #O5 (2026-05-30). The app still emits these counters in Prometheus FORMAT on its /metrics endpoint (port 9091), scraped by the CloudWatch agent in prod. Read them from CloudWatch metrics (or curl the exporter directly in dev); skip with "N/A — metrics endpoint unreachable" if the app is down:

   • tv_ticks_dropped_total — must be 0
   • tv_depth_sequence_holes_total — operator reviews rate
   • tv_instrument_registry_cross_segment_collisions — known collisions only (e.g. 2)
   • tv_websocket_connections_active — 0 off-hours, 5 during 9:00-15:30 IST
   • tv_questdb_spill_bytes — 0 during steady state

9. QuestDB pulse — MCP questdb_sql Run these 3 queries, report row count / first value (skip with "N/A — QuestDB OFFLINE" if TICKVAULT_QDB_STATUS != REACHABLE):

   • SELECT count() FROM ticks WHERE ts > dateadd('h', -1, now()) — hour's tick count
   • SELECT count() FROM historical_candles — total candles
   • SELECT count() FROM derivative_contracts WHERE status='active' — active contracts

10. Active alarms — CloudWatch alarms / run_doctor The Alertmanager container was removed in #O2 and the list_active_alerts MCP tool was retired in #O5 (2026-05-30). Check CloudWatch alarms (prod) or mcp__tickvault-logs__run_doctor for the current health verdict. Any firing alarm = red row.

Output format

Print exactly this markdown table at the end (no filler prose above/below):

| # | Check | Status | Detail |
|---|-------|--------|--------|
| 1 | Runtime attach | PASS | profile=local qdb=REACHABLE api=OFFLINE |
| 2 | Static guards (validate-automation) | PASS | 31/31 |
| 3 | Doctor full-system health | PASS | 6/7 sections green (aws SKIP) |
| 4 | 100% audit | PASS | P=41/41 R=4/4 |
| 5 | ERROR tail (last 10) | PASS | count=0 |
| 6 | Summary snapshot | PASS | zero ERROR events in window |
| 7 | Novel signatures (60min) | PASS | count=0 |
| 8 | Metrics — ticks_dropped | PASS | 0 |
| 8 | Metrics — depth_seq_holes | PASS | 0 |
| 8 | Metrics — id_collisions | PASS | 2 (expected: NIFTY id=13, BANKNIFTY id=25) |
| 8 | Metrics — ws_active | INFO | 0 (off-hours expected) |
| 8 | Metrics — qdb_spill_bytes | PASS | 0 |
| 9 | QuestDB — ticks last 1h | INFO | 0 (off-hours expected) |
| 9 | QuestDB — historical candles | PASS | 220500 |
| 9 | QuestDB — active contracts | PASS | 105846 |
| 10 | Active alarms | PASS | none firing |

Rules

• If anything is FAIL, name the runbook file from docs/runbooks/ that applies. If no runbook exists, say so — do not invent one.
• INFO is the right status for "expected 0 off-hours" — don't flag it as FAIL.
• If TICKVAULT_QDB_STATUS=OFFLINE, mark row 9 as SKIP — Docker not up; run 'make docker-up' to enable.
• If TICKVAULT_PROM_STATUS=OFFLINE, same for row 8.
• Never hallucinate values. If an MCP tool returns nothing, report N/A.
• Do NOT fix anything unless I say so. Just report. If I ask for fixes, apply them one at a time with a ratchet test each.

Performance budget

Total runtime ≤ 60 seconds on a warm system. If a single check exceeds 10 seconds, print its elapsed time next to the status so I can see which one is the bottleneck.