Deslop
Remove AI-generated code slop introduced in a branch so the diff reads like a human wrote it and matches the surrounding file and this repo's conventions.
When to use
- After writing or generating a batch of code, before opening a PR.
- When a diff shows the tells of AI generation: over-commenting, defensive scaffolding, or style that fights the existing file.
- Prose and user-facing markdown are out of scope. Run the
humanizerskill over those instead.
What to remove
- Comments a human would not add: restating what the code already says, or inconsistent with the comment density of the rest of the file.
- Defensive checks and
try/catchblocks abnormal for the area, especially on trusted or already-validated code paths. Thesecurityrule puts validation at trust boundaries, not in internal code. - Casts to
anyused only to bypass a type error. Fix the type instead. - Deep nesting that early returns would flatten.
- Naming, import, or export style inconsistent with the file and the
code-stylerule.
Guardrails
- Keep behavior unchanged unless you are fixing a clear bug.
- Prefer minimal, surgical edits over broad rewrites.
- Never remove a check that guards a real trust boundary or real error handling. When unsure whether a check is slop or load-bearing, leave it.
- Do not weaken types, lint rules, or tests to make the edit pass. Fix the root cause.
- After editing, run
pnpm format && pnpm lint:fixand let the tests pass. - Report a 1-3 sentence summary of what changed.
Related skills
| Skill | Use for |
|---|---|
| humanizer | Removing AI tells from prose and user-facing markdown |