prod-watchlisted

# Prod Watch Watch production for new errors using connected MCP tools, cluster them by cause, and surface the ones that need attention. ## Steps 1. Pull Vercel runtime logs for `<project>` since the last run timestamp. 2. Pull Supabase error logs for the same window. 3. Deduplicate and cluster new errors by root cause (not by message string — group similar causes together). 4. Per cluster: give likely cause + file:line or route + request path. 5. Rank clusters by volume (highest first). ## Constraints - Triage only — never deploy, never modify data, never change config. - If you see a critical error (data loss risk, auth bypass, billing), escalate immediately instead of waiting for the next run. ## Report format (each run) ``` New error clusters: <count> Top 3: 1. <cause> — <count> occurrences — <file:line or route> 2. ... 3. ... Escalation needed: yes | no ``` Quiet → print "quiet" and end. ## Loop usage ``` /loop 30m watch prod errors on <project>. ``` Requires: Vercel MCP + Supabase MCP connected.