de-slop

# De-slop An editorial gate that runs **after** a first draft exists and **before** the content leaves the building. It does two things: 1. **Style:** detects and removes the characteristic signs of AI writing — slop vocabulary, formulaic structure, inflated significance, formatting tics — while preserving the draft's meaning, facts, and voice. 2. **Substance:** blocks hollow drafts. Slop-free text that says nothing is still not shippable; the editor can spike the piece and send it back with a diagnosis. **Philosophy:** original writers (human or agent) are not burdened with style rules while drafting. Slop removal is a separate editorial pass, the way a copy editor works after the reporter files. Drafting skills should not inline these rules; they hand finished drafts to this gate. **Scope:** external-audience artifacts — email, documents, presentations, posts, web copy. Conversational chat replies are dialogue, not drafts, and are not gated. ## Cold-read rule The author of a draft is the worst judge of its slop — same model, same context, full commitment bias. When this skill runs as a wired gate, the calling skill MUST invoke it as a **fresh-context review** (a subagent or forked context) that receives only: the draft (file path or text), the medium, and the intended audience. It must NOT receive the drafting conversation. Judgments are made from the draft alone, as a first-time reader would encounter it. A manual `/de-slop` invocation by the user may run inline — bu...