Convex Agents Debugging

Expose raw LLM requests, responses, and context for Convex Agents debugging.

25|4|Updated Oct 18, 2025
One-click install
npx skills add https://github.com/Sstobo/convex-skills --skill convex-agents-debugging
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: Convex Agents Debugging
Source: https://github.com/Sstobo/convex-skills/tree/main/convex-agents-debugging
Command: npx skills add https://github.com/Sstobo/convex-skills --skill convex-agents-debugging

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Debugging Convex Agents can be time-consuming without visibility into the exact LLM inputs/outputs and the agent's internal state. This skill provides structured visibility into requests, responses, and context to speed diagnosis.

Core Features & Use Cases

  • LLM IO Logging: Capture and inspect raw requests and responses to understand alignment issues.
  • Context Inspection: Reveal what context the agent sees at each step to diagnose missing or incorrect data.
  • Tool Call Auditing: Track tool invocations and results to verify correct orchestration.
  • Database State Checks: Inspect relevant agent tables and messages to trace data flow.

Quick Start

Run the debugging scenario with your local Convex Agent setup and enable raw logging to observe the flow of messages and context.

Frequently Asked Questions about Convex Agents Debugging

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I debug Convex Agents when the LLM outputs are off?

Debug Convex Agents by exposing raw LLM requests and responses to inspect alignment issues. Capturing these inputs and outputs provides the visibility needed to quickly understand and correct why the generated agent outputs are incorrect.

Why does my LLM agent have missing context during tool calls?

Missing context occurs when the agent does not receive the correct data at a specific step. Inspect the context the agent sees at each step to diagnose missing or incorrect data and verify proper message ordering across threads.

What is the best way to audit tool calls and message ordering in LLM agents?

Auditing tool calls requires tracking tool invocations and results to verify correct orchestration. You can trace the flow of messages and tool calls across threads and the database to ensure proper agent execution.

How do I inspect the database state and internal context of Convex Agents?

Inspect the database state of Convex Agents by examining relevant agent tables and messages. Exposing this internal state during development allows you to trace data flow and audit context across threads.

Can I log raw LLM requests and responses locally for Convex Agents?

Yes, you can log raw LLM requests and responses by running the debugging scenario with your local Convex Agent setup. Enabling raw logging allows you to observe the flow of messages and context during development.