slop

Ask for a review through the single level of abstraction lens. The agent classifies each statement as core, down, up, or cross-cutting, then reports every place where a high-level intent line sits next to low-level mechanics with file:line evidence.

SLAP requires every statement in a function to sit at the same conceptual level, so high-level intent is never interleaved with low-level mechanics. Error handling is treated as a separate concern that lives in a dedicated handler above or below the business logic.

Use it only when you explicitly want this perspective, such as asking whether a function does too much or whether errors are handled at the right layer. General reviews and refactors are intentionally left to other lenses.

No. The agent reports only detections, raw counts, move directions, and exemption flags. Severity grades are produced downstream by a deterministic rubric, anchored examples, or a human reviewer to avoid uncalibrated labeling.

For each foreign concern it proposes a concrete move: push down into a value, parser, or infrastructure helper, or push up into the edge handler that maps failures to status codes or messages. Each move is tied to file:line evidence.

The exemption checklist covers tiny functions, functions whose single job is the foreign concern itself, atomicity or correctness requirements, and edge-translator handlers. Exemptions are recorded as facts and used downstream to cap the grade.