heterogeneous-agent

Debugging a heterogeneous agent pipeline involves capturing raw CLI traces, comparing them to adapted events, and isolating the failing layer across adapters, IPC, and tool-call chains. This skill provides a structured workflow with reference flows and invariants to diagnose these cross-layer failures efficiently.

Tracing tool-call failures across Claude Code and Codex adapters requires comparing raw traces against adapted events to pinpoint mapping mismatches. The workflow provides specific reference flows for these adapters, subagent handling, and tool state tracking to isolate persistence or hydration faults.

Isolating faults in agent tool persistence and UI hydration requires checking step boundaries and adapter mapping against raw CLI traces. The skill defines well-defined invariants for step boundaries and subagent handling, enabling focused tests and breakpoint strategies to isolate the exact failing layer.

Yes, this workflow is specifically designed for diagnosing issues across external CLI agents, IPC, and tool-call chains. It provides a repro-to-fix workflow with practical guidance for isolating faults, comparing raw and adapted events, and running focused tests across the pipeline.

Agent pipeline failures during trace mapping often stem from broken invariants at step boundaries or subagent handling mismatches. The skill provides reference repository files and structured breakpoint strategies to compare raw and adapted events, isolating whether the fault lies in the adapter, persistence, or UI hydration.