Identity Audit Workflow

Use this skill when checking whether function, file, or folder names still match their contents.

The CLI scans files, folders, and graph-level declaration symbols. It does not treat every local variable, parameter, or field as an independent naming target; those can still influence the containing file summary. File names are evaluated with parent scope, and declaration-heavy files should normally be read as evidence-thin until follow-up inspection proves an actual rename or split.

Pipeline

1. Run the mechanical scan:

``bash indexion identity audit . ``

For a machine-readable queue:

``bash indexion identity audit --format=json --output=.indexion/cache/identity/report.json . ``

1. Treat each row as a review candidate, not proof. Compare:

- name: the scoped name inferred by the identity package - expectedsummary: what the path/name/scope predicts - actualsummary: graph-derived declarations, docs, module notes, and path terms - assessment: whether this is actual drift, overbroad content, or insufficient content - recommendation: first operation to verify

1. Verify before editing:

``bash indexion doc graph --format=text <path> indexion grep --semantic=name:<symbol> . rg "<name-or-term>" <path> ``

1. Choose the smallest confirmed operation:

- Treat insufficient-content as an evidence problem, not a naming-drift proof. Inspect whether the file is intentionally declarative/thin, unsupported by the graph extractor, empty, or missing doc/declaration material before renaming. - Rename a symbol when its implementation is cohesive but the name is stale. - Rename a file when its declarations are cohesive but the file name is stale. - Rename a folder when contained files share a clearer parent concept. - Move a file when it fits an existing folder better than its current one. - Split a file when candidates show multiple dominant responsibilities.

1. After changes:

``bash moon info && moon fmt moon test indexion identity audit . ``

The audit is designed to surface review work. Do not maximize the score mechanically; verify the actual code ownership and references first.