analyze

Invoke the analyze skill with your question, such as why a workflow fails or how a feature is wired. It reads the relevant files, ranks competing explanations by evidence strength, and returns a synthesis with concrete file references and confidence levels.

Evidence is a claim directly supported by repository artifacts like code, tests, or configs with file references. Inference is a reasoned conclusion drawn from that evidence. This skill labels every material claim as evidence, inference, or unknown so nothing is overclaimed.

No. The skill is read-only by contract and never edits files or produces implementation plans. If a next step is needed, it only suggests a discriminating read-only probe that would reduce uncertainty.

Skip it when you explicitly want code edits or fixes, when you need a product plan, or when the answer is a simple one-file fact lookup. In those cases, read the file directly or use an implementation-oriented workflow instead.

It opens bounded parallel exploration lanes, typically one for primary code paths, one for configuration and orchestration, and one for tests and docs. Each lane answers a concrete sub-question, and results are merged into a single ranked synthesis.