/hyperpower:humanize

Find and remove the marks a model leaves on code.

Reference

/hyperpower:humanize src/ --all

When to use it

Before review, on code a model wrote.

What you get

Three passes over the code. Only the comment pass changes files, and only after you say yes.

How it works

Models leave marks in code: placeholders meant to be filled in later, comments that hedge, a helper written for a single call, and abstractions built for a second case that never came. This command finds them.

Type this one yourself. sweep is the version that runs inside /hyperpower:run, and it assumes the static scan for these marks already ran.

Three passes:

  1. Code. Placeholders, deferrals, hedging, stubs, dead paths, over-abstraction, code that does not fit how your repo works, swallowed exceptions and invented helpers. You get the findings and a plan. Nothing is changed.
  2. Comments. Applies your repo's comment rule, read from CODEBASE_RULEBOOK.md, then CLAUDE.md, then .claude/rules/. This is the only pass that edits. It shows you a sample first and edits only after you say yes. It never deletes a comment that explains why the code is written the way it is.
  3. Repeated code. Blocks that appear three or more times, parallel functions, and copy-paste with one thing changed. It recommends merging the copies only when they change together. Two copies is a coincidence.

By default it checks the current diff. Pass a path to narrow it, or --all for the whole repo.

First it checks whether antislop, ai-slop-detector and jscpd are installed, and tells you which ones it found. If none are installed, it looks for the same problems itself and labels the findings as a model's reading. That is a weaker claim than a linter's result, and the label says so.