What problem does it solve?
Convex deployments surface runtime health signals like read-limit hits and OCC write contention, but turning those raw events into actionable fixes requires manually tracing each event back to the responsible code. This Skill automates that investigation by reading the official Convex MCP insights feed, locating the flagged function's code, and producing evidence-backed findings with concrete fixes.
Core Features & Use Cases
- Insights-driven diagnosis: Reads typed 72-hour health events (documentsReadLimit, bytesReadThreshold, occRetried, occFailedPermanently) from the official Convex MCP and maps each to its root cause in the flagged function's code.
- Structured findings bus: Emits findings per a defined schema with severity, locus, evidence, confidence, and fix capability so fixer agents can be dispatched.
- Concrete fix recommendations: Suggests indexes with .withIndex, .take(n), .paginate, sharded counters for OCC contention, and aggregate components for count scans.
- Use Case: Your production Convex app hits documentsReadLimit on a messages query. The advisor reads the insight event, finds the unindexed .filter() in messages.ts, and reports the exact file:line with the index fix to apply.
Quick Start
Ask the advisor to read my Convex deployment's insights and root-cause any performance or contention events in my functions.
