skills.rest
agentydragon
agentydragonCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

debug_agent_traces

Parse LLM conversation traces and tool calls from PostgreSQL agent runs.

4|1|Updated May 19, 2025
Search Tags:#debugging#postgresql#conversations#tool-calls#resurrection#llm-traces
One-click install
npx skills add https://github.com/agentydragon/ducktape --skill debug-agent-traces
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: debug_agent_traces
Source: https://github.com/agentydragon/ducktape/tree/main/.claude/skills/debug_agent_traces
Command: npx skills add https://github.com/agentydragon/ducktape --skill debug-agent-traces

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Debug props agent traces by reading LLM request/response history, parsing tool calls, and resurrecting past agent conversations to ask follow-up questions about its decisions.

Core Features & Use Cases

  • List agent runs to identify debugging targets.
  • Read LLM traces to inspect rounds, models, tokens, and errors.
  • Parse tool calls from traces to understand tool usage and reasoning.
  • Resurrect conversations to ask follow-up questions and validate decisions.

Quick Start

Connect to the evaluation database and run a trace inspection on a chosen agent_run_id.

Frequently Asked Questions about debug_agent_traces

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

FAQPage Schema
How do I debug LLM agent traces from past runs?▼

Debug LLM agent traces by connecting to a PostgreSQL evaluation database, fetching conversation histories via structured queries, and inspecting rounds, models, tokens, and errors to understand model decisions.

Can I resurrect a past AI agent conversation to ask follow-up questions?▼

Resurrect past AI agent conversations by parsing tool calls and reconstructing JSON-like data structures from stored LLM traces, enabling follow-up questions about previous model decisions.

How do I parse tool calls from LLM conversation histories?▼

Parse tool calls from LLM conversation histories by querying the PostgreSQL evaluation database for agent runs, reading structured trace outputs, and extracting tool usage and reasoning steps.

What is the best way to list and inspect multiple agent runs?▼

List and inspect multiple agent runs by applying structured queries to the evaluation database, fetching traces by agent_run_id, and reading rounds to identify debugging targets and analyze outputs.

Do I need PostgreSQL to inspect AI agent traces and model decisions?▼

Yes, PostgreSQL is required to fetch traces, reconstruct conversations, and analyze model decisions, as the process relies on structured queries and JSON-like data structures stored in the database.

Why does parsing agent tool calls fail when trace data is incomplete?▼

Parsing agent tool calls fails when LLM traces lack complete JSON-like data structures, preventing the reconstruction of conversation histories and the ability to resurrect past agent runs for analysis.

Related Skills

Explore More Software Engineering Skills→
omerspacomerspac

manage-conversation-db

Create and load conversations with ownership checks and save messages asynchronously.

Community
Advanced
Undertone0809Undertone0809
280

debug-run-transcript

Reconstruct Rudder agent run execution transcripts from logs and database sources.

Community
Intermediate
VSF-QC-TTSVSF-QC-TTS

common-telemetry

Tracks token usage and execution metadata for workflow cost reports.

Official
Advanced
ScottHMcKeanScottHMcKean

instrumenting-with-mlflow-tracing

Instrument Python and TypeScript code with MLflow Tracing for AI observability.

Community
Intermediate