slopscore-lint

Version updated for https://github.com/jman4162/slopscore to version v0.8.0.

  • This action is used across all versions by ? repositories.

Action Type

This is a Composite action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

slopscore is an AI-slop linter that measures the density of formulaic, generic, low-specificity, and over-polished writing patterns in text. It reports per-dimension scores and evidence spans to help identify and fix these patterns, nudging writers toward clearer and more specific prose.

What’s Changed

Two new detection axes, both conservative by default.

insight_signaling — a new dimension

Pseudo-profundity tells that announce insight rather than contain it: “load-bearing assumption”, “doing the real work”, “the crux of the issue”, “pressure-test the claim”. Context-gated (literal “load-bearing wall” is not flagged) and mostly low/medium severity — density, not existence, is the tell. Distinct from the WP:AILEGACY puffery in significance_inflation. Rules-only (excluded from the ML FEATURE_ORDER), so no model retrain. Also adds antithesis variants to parallelism (“not merely X but Y”, “less about X, more about Y”, the non-‘it’ em-dash form, “both/and”).

Expanded weasel_attribution — the “fake evidence” axis

Impersonal-passive attribution (“it is widely believed”, “sources say”), unearned-certainty reasoning smells (“Clearly,”, “needless to say”, “it goes without saying”), and hedge+vague-adjective (“somewhat successful”). Scored by default, high-precision.

New --broad flag (also [tool.slopscore] broad)

An opt-in tier of higher-false-positive rules, off by default: rationalist/essayist jargon for insight_signaling (steelman, first principles) and bare quantifiers/hedges/intensifiers for weasel_attribution (many/very/may). Kept off the default score because those bare words have no discriminative power on the ESL/simple-English fairness slices — --broad is a self-editing highlighter, not an accusation. Deliberately excludes the hedges human_writing_signals rewards.

The JSON report gains an insight_signaling key in dimensions; SCHEMA_VERSION is 0.8.0. Fairness gate stays 0% on the plain/non-native slices with the default rules.

Full changelog: https://github.com/jman4162/slopscore/blob/main/CHANGELOG.md