What problem does it solve? Debugging a large codebase often means dumping entire files into context, which is slow, expensive, and error-prone. This Skill localizes and diagnoses a bug by retrieving only the minimal sufficient code context, escalating from cheap text search to structural graph traversal only when needed. ## Core Features & Use Cases - Tiered retrieval ladder: Start with agentic grep/ripgrep on stack-trace anchors, escalate to symbol/call-graph traversal and AST-aligned spans, then broad iterative search only if the locus stays unknown. - Structured diagnosis report: Produces ranked suspect file:line locations, a root-cause hypothesis with quoted evidence, a P0–P4 severity rating, and a minimal context bundle for a fixer. - Cache-aware ordering: Orders retrieval so the stable prompt prefix stays cacheable, keeping each iteration fast and cheap. - Use Case: Given a failing test or stack trace, trace the root cause across a repository and hand off an auditable diagnosis report — without loading the whole codebase or writing any fix. ## Quick Start Use the progressive-bug-hunter skill to localize the root cause of this failing test from its stack trace and produce a diagnosis report.
