deep-clean

name	deep-clean
description	Measured codebase cleanup: dead code, dependency cycles, weak types, duplicate logic, deprecated paths, test debt, and complexity, using real tools and before/after proof.

Deep Clean

Use this for broad cleanup after big changes, migrations, or obvious codebase drift. It is not a substitute for feature work or security validation.

Measure First

Run the repo's relevant static tools before editing: dead-code scan, dependency graph, typecheck, duplicate scan, lint, tests, and build where available. Record the exact command and baseline counts.

Order

1. Structure: dependency cycles, canonical modules, import direction.
2. Strengthen: weak types, dead code, error paths, missing tests.
3. Polish: stale docs, deprecated paths, formatting, AI slop.
4. Measure again with the same commands.

Do not parallelize across those phases; each phase changes what the next phase sees.

Rules

• Delete unused code when callers are gone.
• Migrate live callers before deleting old paths.
• Extract duplicate logic only when intent is shared, not merely shape.
• Keep public API changes explicit and version-aware.
• Stop and fix any test/build regression you introduce.

Use references/full-reference.md for tool commands, phase details, and reporting templates.

Then consider

• harden for security-sensitive surfaces discovered during cleanup.
• converge if cleanup leaves CI red.
• polish for a narrower final quality pass.