Unslop Code

AI code often compiles and looks “professional” while quietly violating YAGNI, fail-fast, and local conventions. This skill bans those fingerprints and reworks code until it matches the problem’s real scale — not a tutorial, not an enterprise scaffold. Scope is model defaults (narrating comments, AbstractFactory-for-todo, silent return [], as unknown as), not every classical maintainability smell.

Core Mandate

1. Ban fingerprints first. Do not polish over-abstracted or over-commented code in place — delete the default, then keep only what the requirement needs.
2. Match problem scale. One call path → one function. Abstract at ≥3 real variants that exist now.
3. Fail loud. No silent catch fallbacks ([], {}, null, fake defaults) unless the product explicitly requires soft degradation.
4. Types over escapes. Ban new any / as unknown as / unexplained @ts-ignore. Fix the model or narrow unknown.
5. Prefer delete. Unused params, speculative helpers, unrequested features, and narrating comments go.
6. Pivot rule. If a module hits 3+ banned patterns, redesign that unit — do not rename Managers into Managers2.

Hard Bans (never ship)