pipeline-debugger

Debug multi-agent pipeline failures by following a strict evidence-first workflow that gathers logs and stack traces, reproduces the issue, isolates the root cause to specific architectural layers, and proposes a minimal verified fix.

The best way to isolate root cause is using pipeline-aware investigation that analyzes prior phase outputs and revision loop feedback to narrow failures down to a specific file, function, and architectural layer before proposing targeted fixes.

Fix behavioral mismatches by diagnosing the root cause and proposing a minimal fix that respects hexagonal architecture, ensuring corrections address architectural violations directly rather than working around them, followed by test verification.

Yes, evidence-first debugging applies to performance profiling and performance regressions by turning unclear performance observations into reproducible evidence, isolated root causes, and targeted architecture-respecting changes.

Pipelines fail after exhausting revision loops because prior guesses were not systematically narrowed. A disciplined debugger uses prior phase outputs to isolate the fault to a specific location and architectural layer before applying a minimal fix.