health-check

name	health-check
description	Quick operational health check of all trading pipelines. Use when: user says "health check", "is the bot running", "pipeline status", "system check", "any errors", "is it working", or "/health-check". Checks CI freshness, prediction recency, order fills, circuit breaker proximity, and test status. Produces a compact status dashboard in under 60 seconds.

Health Check

Quick operational pulse across all 4 pipelines. Is the bot running? Are orders filling? Is the circuit breaker close? Any errors? This is the "is everything okay right now?" skill.

Prerequisites

• Always git pull first.

Process

1. Pull latest

cd /Users/mrmrnm-max/polymarket-bot
git pull --rebase || git pull

2. Check CI pipeline freshness

git log --oneline -10 --format="%H %ai %s" | head -10

Look for the most recent Auto: commit. Calculate how many minutes ago it was.

• BTC 5m: stale if > 10 minutes since last Auto commit
• BTC 15m: stale if > 20 minutes
• ETH 5m: stale if > 10 minutes
• Kalshi: stale if > 10 minutes

If the last Auto commit is >15 minutes old, check GitHub Actions:

gh run list --limit 5

3. Check prediction freshness per pipeline

-- For each database:
-- data/predictions.db (BTC 5m)
SELECT MAX(predicted_at) as last_prediction, COUNT(*) as total FROM predictions;

-- data/predictions_15m.db (BTC 15m)
SELECT MAX(predicted_at) as last_prediction, COUNT(*) as total FROM predictions;

-- data/predictions_eth.db (ETH 5m)
SELECT MAX(predicted_at) as last_prediction, COUNT(*) as total FROM predictions;

-- data/predictions_kalshi.db (Kalshi)
SELECT MAX(predicted_at) as last_prediction, COUNT(*) as total FROM predictions;

4. Check trade execution (BTC 5m + ETH 5m)

-- data/predictions.db — BTC 5m orders
SELECT
  COUNT(*) as total_orders,
  SUM(CASE WHEN mode='live' THEN 1 ELSE 0 END) as live_orders,
  SUM(CASE WHEN mode='live' AND status='filled' THEN 1 ELSE 0 END) as fills,
  SUM(CASE WHEN mode='live' AND status='pending' THEN 1 ELSE 0 END) as pending
FROM orders;

-- Today's live P&L
SELECT SUM(pnl) as daily_pnl FROM orders
WHERE mode='live' AND date(placed_at) = date('now');


 ()  orders
 mode  status
   placed_at  datetime(, );

Repeat for ETH 5m (data/predictions_eth.db) if it has an orders table.

5. Check circuit breaker and kill switch

# Kill switch file check
test -f data/KILL_SWITCH && echo "KILL SWITCH ACTIVE" || echo "Kill switch: OFF"

Circuit breaker proximity:

• Daily P&L from step 4 vs $300 daily loss limit
• If P&L < -$200: CAUTION
• If P&L < -$250: CRITICAL

6. Check experiments

python3 src/optimization_tracker.py summary

Note: how many active, how many need attention (validated or revert candidate).

7. Run quick test

python3 -m pytest tests/ -v --tb=short 2>&1 | tail -5

8. Produce compact status dashboard

Format the output as:

Pipeline Health — YYYY-MM-DD HH:MM CST
═══════════════════════════════════════

BTC 5m  (PROD):   ✅ OK    Last: HH:MM UTC  |  Orders today: N  |  Fills: X/Y
BTC 15m (paper):  ✅ OK    Last: HH:MM UTC
ETH 5m  (paper):  ✅ OK    Last: HH:MM UTC  |  Orders today: N  |  Fills: X/Y
Kalshi  (paper):  ✅ OK    Last: HH:MM UTC

Circuit Breaker:  $X.XX / $300  (Y%)   ✅ OK
Kill Switch:      OFF
Consecutive Loss: N / 5                 ✅ OK
Tests:            NNN passed            ✅ OK
Experiments:      N active (M need attention)
Last CI Commit:   N min ago             ✅ OK

Status indicators:

• ✅ OK — normal operation
• ⚠️ WARN — stale data (>10 min), approaching limits (P&L < -$200), or pending orders stuck
• 🔴 DOWN — no predictions in expected window, kill switch active, or circuit breaker tripped
• 🧪 ATTENTION — experiment needs validation decision

9. Route to other skills if needed

If the health check finds issues, suggest the appropriate next step:

• Edge problems → "Run /critique-performance for a deep dive"
• Decision thresholds → "Run /review-decisions to check triggers"
• Experiment alerts → "Run /validate-optimization to check status"
• Pipeline down → Check GitHub Actions for errors, check docs/ops/BREAK_FIX_LOG.md

Key Rules

• Speed over depth. This should take < 60 seconds. If something needs investigation, flag it and suggest the deeper skill.
• Always show all 4 pipelines. Even if the user only cares about BTC 5m, show the full picture.
• UTC and CST. Show prediction timestamps in UTC (what the DB stores) and the report header in CST (user's timezone, GMT-6).
• Don't fix things. This skill diagnoses. It does not make changes, commit code, or modify databases.