Score AI Slop

Score AI Slop

Give the author a quality-pass read: how much slop is here, by category, and is it ready to publish. This is the aggregate view over the same engine that detect-ai-slop and remove-ai-slop use. It does not rewrite, and it does not return a single 0-10 number; a headline score becomes a thing people optimize against, so the output stays a breakdown plus a plain-language readiness call.

Rules, categories, severity, and the override-merge order live in ../../shared/spec.md. Read it before scoring.

How to score

Deterministic tier (preferred). Run the bundled script and use its summary (the per-category counts) plus its findings for the breakdown:

python3 shared/slop_count.py --file <path> \
  [--allow-list ~/.claude/config/remove-ai-slop/user-allow-list.md] \
  [--allow-list .claude/skills/remove-ai-slop/overrides/user-allow-list.md] \
  [--extra-vocab ~/.claude/config/remove-ai-slop/user-vocabulary.md] \
  [--extra-vocab .claude/skills/remove-ai-slop/overrides/user-vocabulary.md]

Pipe text on stdin instead of --file when the user pasted it. Pass whichever override files exist. The script is the sole override parser; consume its output, do not re-parse the override tables.

Judged tier (you). The script cannot see meaning. Add a short note per meaning-dependent category from the spec (padding-and-filler, elegant-variation, and contextual cases). Mark these judged.

Fallback — when python3 is absent

If python3 is unavailable, do the read yourself from the spec and references, read the allow-list files directly, and tag the result unstable: say "No python3 found; this is an LLM-only pass and the counts are not byte-stable."

Output

A per-category breakdown, then a one-line readiness call. No 0-10 number.

Slop breakdown (deterministic + judged):

| Category                 | Count | Severity | Notes                         |
|--------------------------|-------|----------|-------------------------------|
| overused-vocabulary      | 9     | low      | clustered in para 1           |
| formulaic-constructions  | 1     | high     | "not just X, it's Y" (para 2) |
| formatting-tells         | 2     | med      | em-dash run in para 2         |
| padding-and-filler       | 1     | med      | judged: hedging in para 3     |

Readiness: heavy slop — run remove-ai-slop for a rewrite pass.

Readiness is plain language tied to the spec threshold, not a number:

• No findings → Clean — reads as written by a human, publish as-is.
• A few isolated tells, none clustered → Light — a couple of edits, no rewrite needed.
• Any paragraph over the co-occurrence threshold → Heavy — run remove-ai-slop.

Rules:

• Never emit a single 0-10 number. The per-category counts are the signal.
• If the script ran, note the counts are stable. On fallback, tag the read unstable.
• Respect the allow-list: an allow-listed word is never counted.
• If there are no findings, say so plainly and stop.

Scope

Aggregate read only. To see exactly where each tell is, that is detect-ai-slop. To rewrite the text, that is remove-ai-slop. Do not rewrite from this skill.