Agent Debug — Diagnose AI Agent Failures

Version updated for https://github.com/Viktorsdb/agent-debug to version v0.3.0.

  • This action is used across all versions by ? repositories.

Action Type

This is a Composite action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

Summary:
The agent-debug GitHub Action is a diagnostic tool for AI agents, providing root cause analysis and actionable fixes for agent failures. It automates the identification of issues such as incorrect tool usage or hallucinations, explains why failures occurred, and suggests concrete system prompt or tool definition adjustments. The action supports multiple AI providers and SDKs, integrates seamlessly into GitHub workflows to comment on pull requests with debugging insights, and offers automated trace capture for streamlined troubleshooting.

What’s Changed

What’s new in v0.3.0

🔧 CodeFixer — AI reads your exact file content and fixes it

No more text-matching failures. The AI now reads your actual Python file, identifies the exact lines causing test failures, and applies precise fixes. Zero guesswork.

4-step fix flow

  1. Analyze trace — root cause analysis
  2. Run tests — pre-flight to identify all failures
  3. Fix code — CodeFixer rewrites broken lines
  4. Apply prompt fixes — system prompt + tool description improvements

Batch test validation

All fixes applied first, tests run once at the end. If tests still fail, revert-all option.

Multi-provider support

Pass --system-prompt and --code files — AI generates fixes from actual file content, not guesses.

Install

pip install agent-debug==0.3.0

Usage

agent-debug fix trace.json \
  --code agent.py \
  --system-prompt prompts/system.txt \
  --test-cmd "pytest tests/"