code-decay

Run /code-decay with the repository path to rank every tracked file by score = churn × complexity. Churn comes from git log history and complexity from ast-grep branch and nesting analysis, producing a dated Markdown report in docs/code-decay inside that repo.

The score multiplies a file's churn count by its complexity value (branch count plus nesting depth), so files that change often and are structurally complex rank highest. Labels like God Class, Shotgun Surgery, and Dormant Complexity highlight percentile-based outliers.

The ast-grep backend covers Python, JavaScript, TypeScript, TSX, Lua, Bash, C, C++, and Solidity. Any other file type falls back to a regex and indentation-based proxy backend that always returns a complexity estimate without erroring.

No, each run analyzes exactly one repository and never merges history or scores across repos. Run the skill separately per repository; each report is written inside the analyzed repo, not a shared location.

The run continues with a warning rather than stopping. The shallow-clone warning is written into the report body itself, noting that churn numbers reflect only the truncated history and may understate real hotspots.

A missing or broken ast-grep binary fails loudly by design for supported languages. A silent fallback to the rougher proxy backend would disguise a broken installation as an accuracy problem, so the error must be fixed instead.