No-Slop Review
Slop is a list. This skill runs that list as a review pass.
Run it against code you just wrote or edited before you call the work done.
How to run it
- Identify every file you created or changed in this task.
- For each one, walk the checklist top to bottom.
- For every hit: fix it, or write one line on why it’s a deliberate exception.
- Only then is the work eligible for the review gate.
Do not report the work as done while any checklist item is an unexplained hit.
The checklist (summary)
Full version with examples: no-slop-checklist.md.
- Dead code — unused vars, imports, functions, branches. Delete it.
- Unhandled errors — every fallible call has a real failure path, not a swallowed exception.
- Duplication — copy-pasted blocks. Three strikes → extract.
- Vague names —
data,handle,temp,doStuff. Name the thing. - Untested edges — empty, null, max, concurrent, the unhappy path.
- Restating comments —
// increment i. Delete. Keep only comments that explain why. - Inconsistent with the file — match the surrounding code’s idioms, not your defaults.
- Scope creep — anything not in the brief. Flag it; don’t smuggle it.
- Fake done —
TODO, stubbed returns, hardcoded values pretending to be real. - Unverified claims — “this works” / “deployed” with no proof in the same turn.
Why this exists
Agentic engineering is a multiplier. Without a no-slop pass, it multiplies inattention. Run it every time.
Customizing
The checklist is a baseline. Add the slop patterns specific to your stack and the ones you personally keep shipping.