What problem does it solve? AI-generated code often ships with defensive clutter: double-guards, dead fallbacks, redundant re-validation, narrating comments, and speculative abstractions. Manually auditing a diff for these patterns is slow and error-prone, and naive cleanup risks deleting load-bearing validation at trust boundaries. ## Core Features & Use Cases - Ten slop categories: Detects defensive double-guards, dead fallbacks, redundant re-validation, narrating comments, speculative flexibility, premature abstraction, over-broad error swallowing, duplicated shape checks, boilerplate restating defaults, and journal comments. - KEEP rules and trust-boundary proofs: Every category pairs with a rule naming what must survive, and deletions at trust boundaries (user input, API responses, file reads) require constructing the adversarial input case before removal. - Scoped, verifiable workflow: Operates only on the diff or an explicit file list, supports batched delegation to executor agents, and requires running the project's build, lint, and tests after edits. - Use Case: After an LLM generates a feature branch, run this skill on the diff to strip narrating comments and dead fallbacks while preserving the input validation that actually guards external data. ## Quick Start Ask the agent to remove AI slop from the files changed in the current diff and verify the build still passes.
