Remove AI Slops

What it does

This skill finds and removes typical AI-generated "slop" across a bounded scope — your branch diff or an explicit file list. The ordering is the whole point:

1. Scope — files changed since git merge-base main HEAD, excluding deleted, binary, generated and vendored paths.
2. Lock behavior — write the narrowest regression test for any uncovered behavior and make it green before deleting a single line. No green baseline means STOP.
3. Deletion ladder — before trimming smells, ask whether the code should exist at all: delete entirely → reuse an existing repo helper → replace with stdlib/platform native → only then simplify in place.
4. Ten slop categories — obvious comments, over-defensive code (except Exception, empty catch {}), excessive complexity (nested ternaries, if/elif variant chains → match/case + assert_never, object annotations), needless abstraction, boundary violations, dead code, duplication, provably-equivalent performance wins, missing tests, and oversized modules (>250 pure LOC, with a mandatory modular split plan). Applied safest-to-riskiest.
5. Quality gates + critical review — tests, lint, typecheck, security scan; failing hunks get reverted rather than patched over. Three failures on one file escalates to the user.

Every category ships explicit KEEP rules, so WHY-comments, trust-boundary validation and reflection-referenced code survive the pass.

Who it's for

• Reviewers and tech leads triaging large AI-authored pull requests
• Teams drowning in restating comments, redundant try/except, and single-use wrappers
• Developers who want automated refactoring with a hard "behavior must not change" guarantee
• Architects enforcing file-size and complexity budgets across a codebase

Examples

• PR cleanup: say "remove AI slop on this branch" and get regression tests added, dead code and over-defensive guards removed, plus a report with per-gate results and net LOC delta.
• Breaking up a monster file: a 420-pure-LOC service.py gets a responsibility-based split plan presented for approval first, then clean modules with re-export-only __init__.py.
• Safe performance fixes: a list scan inside a loop becomes a set lookup (O(n²)→O(n)) when equivalence is obvious, while ambiguous algorithm changes land in the "Skipped" list with a stated reason.

Install the skill I found on Claude Skill Mart.
Copy the packages/shared-skills/skills/remove-ai-slops folder from the GitHub repo code-yeongyu/oh-my-openagent into my ~/.claude/skills/remove-ai-slops/.
When it's done, tell me in one line what this skill can do.