Debugs complex runtime issues by instrumenting code with structured execution traces and data checkpoints.
The debugging-with-llm-trace skill empowers Claude to diagnose deep-seated logic errors and performance bottlenecks by providing visibility into actual runtime data rather than relying on guesswork or static analysis. By instrumenting the codebase with spans and data checkpoints, it allows for iterative narrowing of root causes through a structured trace tree, effectively replacing traditional print-statement debugging with a robust, AI-readable observability layer that shows exactly what happens during execution.
Key Features
01Data checkpointing to capture state snapshots before and after transformations
02CLI integration for listing, filtering, and inspecting hierarchical trace trees
03Structured trace and span instrumentation for granular execution flow tracking
04Iterative debugging workflow designed to halve the search space for bugs
051 GitHub stars
06Automatic error and stack trace capture within instrumented spans
Use Cases
01Debugging complex asynchronous flows and external API integration failures
02Diagnosing silent logic failures where functions return incorrect data without crashing
03Identifying performance bottlenecks by analyzing span durations across the execution tree
