<< All versions
Skill v1.0.0
currentLLM-judged scan90/100alvin-reyes/better-agentic-ide/anti-slop
──Details
PublishedOctober 21, 2026 at 11:43 PM
Content Hashsha256:5f327c3b94b34040...
Git SHA5d73348eeb80
──Files
Files (1 file, 3.0 KB)
SKILL.md3.0 KBactive
SKILL.md · 37 lines · 3.0 KB
version: "1.0.0" name: anti-slop description: Rules against AI slop in code, comments, commit messages and docs. Use whenever you write or edit code, write documentation, a README, a commit message or a PR description, or review changes for quality.
Anti-slop
Slop is output that looks finished but isn't, or that pads the work with words and code nobody asked for. Hold every change to these rules.
Code
- Do what was asked, and only that. No extra features, options, config flags, abstraction layers or "future-proofing" the task doesn't need.
- No stubs. Never leave
TODO,FIXME,pass,throw new Error("not implemented"), "implement this", "your logic here" or "... rest of the code". Write the real code, or say plainly what you didn't do and why. - No debug leftovers. Remove
console.log,printdebugging anddebuggerstatements you added. Real logging goes through the project's logger. - No commented-out code. Delete it; git keeps history.
- Comments explain why, not what. Don't narrate code ("This function adds two numbers", "Increment i"). Keep comments that record a reason, a constraint or a non-obvious decision.
- Match the codebase. Follow the file's naming, structure, error handling and comment density. Reuse existing helpers instead of writing near-duplicates.
- No new files without a reason. Don't add READMEs, example files, summary markdown or test scaffolding unless asked.
- Don't swallow errors. No empty
catch, noexcept: pass, no silent fallbacks that hide a failure. - Tests test behaviour. No tests that only assert a mock was called, no snapshot dumps nobody reads, no tests edited to pass without fixing the cause.
- Finish the loop. Run the project's typecheck, lint and tests for what you touched. Report failures honestly.
Writing (docs, READMEs, commit messages, PR descriptions, replies)
- Lead with the point. First sentence says what changed or what the answer is.
- Plain words. Not: delve, leverage, utilize, robust, seamless, cutting-edge, game-changer, supercharge, elevate, unlock, embark, tapestry, "in today's fast-paced world", "it's worth noting", "a testament to".
- No filler. Cut openers ("Great question!", "Certainly!"), recaps of what the reader just read, and closing offers of more help.
- No hype or hedging stacks. Say what it does. Qualify only when it matters.
- Formatting serves the reader. No emoji in headings, no bold on every other phrase, no bullet lists for things that read better as a sentence.
- Commit messages: imperative subject under ~70 characters, then what and why. No "This commit...".
- Be specific. Name the file, the function, the number. "Improved performance" says nothing; "cut cold start from 2.1 s to 0.8 s" does.
Before you finish
Re-read your diff as a reviewer would. Remove anything on the lists above. If the ADE plugin's slop check reports a finding, fix it unless it's intended, and say so in one line when it is.