Tool-call debugging
How to Debug AI Agent Tool Calls
Inspect tool arguments and results, reproduce the failure, test one controlled change, and verify the real fix.
Read the guidePractical debugging
AI agent debugging guides
Learn how to move from a failed run to inspectable evidence, a controlled experiment, a verified production change, and a regression that protects the fix.
Replay and tracing
AI Agent Replay and Tracing Guide
Understand what tracing records, what replay can prove, and how to keep diagnostic evidence separate from the production fix.
Read the guide01 · Capture
Preserve the failed execution
Record agent, tool, MCP, and model boundaries in the existing Python runtime, with complete payloads stored locally.
02 · Investigate
Test a bounded hypothesis
Hold the baseline still, change one captured boundary, and distinguish recorded, edited, and live evidence.
03 · Verify
Prove the actual change
Run the original scenario with exploratory controls removed, evaluate the approved outcome, and save the passing case as a regression.
Open instrumentation
Start with the public Python SDK and IDE skills
Install the alpha SDK with pip install debrix. Cursor, Claude Code, and Codex can then inspect the local debugging evidence and guide the workflow over MCP. Debrix does not require a custom agent runtime.
Bring the next failure to Debrix.
Join the waitlist for controlled early access to the complete local Failure → Fix loop.