Watchdog — Cron Health Monitor

Monitors all cron jobs for failures and auto-fixes them. Posts to Slack only when issues are found or unfixable errors exist.

CRITICAL: Slack Routing

When sending messages to Slack, you MUST specify channel: "slack" in every message tool call:

message(action: "send", channel: "slack", target: "C0AHYTV5WP7", message: "...")

Without channel: "slack", messages will fail silently.

Schedule

Every 6 hours: 5, 11, 17, 23 CT

Steps

1. cron(action: "list") — get all jobs and their current status
2. For each job, check: lastStatus error? consecutiveErrors > 0? What was lastError?
3. For model not allowed errors: use cron(action: "update", jobId: "...", patch: { payload: { model: "anthropic/claude-sonnet-4-6" } }), then force-run, log change
4. For timeout errors: use cron(action: "update", jobId: "...", patch: { payload: { timeoutSeconds: \x3Ccurrent + 60> } }) — NEVER edit cron JSON files directly
5. For other errors: analyze, attempt fix if possible, or flag as unresolved
6. Post to Slack C0AHYTV5WP7 (#morning-briefs) ONLY if issues were found/fixed or unfixable errors exist
7. If everything is healthy: no Slack message (silent pass)

CRITICAL: Never Edit cron/jobs.json Directly

Always use the cron tool with action="update" to modify job settings. Direct file edits break the cron system.

Slack Alert Format

🐺 Watchdog Report — \x3Ctimestamp>

✅ Fixed: \x3Cjob-name> — \x3Cwhat was fixed>
❌ Unfixable: \x3Cjob-name> — \x3Cerror summary>
⚠️ Flagged: \x3Cjob-name> — \x3Cissue description>

Only send if at least one issue exists.