← ClaudeAtlas

unsloppifylisted

Remove AI writing tells from text. Use when a text artifact is the deliverable or a major part of one, such as documentation, PR or commit text, release notes, reports, marketing copy, code comments, or when asked to de-slop, humanize, or clean up text. Not for chat replies, status updates, or completion summaries around a non-text task.
woerndl/unsloppify · ★ 14 · AI & Automation · score 71
Install: claude install-skill woerndl/unsloppify
# unsloppify Slop is text that performs instead of informs. The fix is never decoration. State the plain version. This applies to any draft regardless of author, whether a person or a model wrote it. Judge the writing on its own terms, not by who produced it. ## When to load Load this skill when a text artifact is the deliverable or a major part of one: a document, PR or commit text, release notes, a report, marketing copy, a substantial set of code comments. Words produced around other work do not qualify: chat replies, status updates, tool commentary, completion summaries. Those follow the writing baseline. If the skill is loaded and the task does not qualify, skip the procedure and continue the task. ## Procedure Outside the mechanical house rules (em dashes, semicolons, and UTF middle dots, removed on sight), every flag below is a prompt to check whether the flagged form is genuinely necessary, the exact word for the exact case that no plain phrase covers. The form stays available, but only with that reason. The goal is caught tropes rewritten, not a draft too timid to say anything. Drafting: 1. Name the context (prose, docs, PR or commit text, code comments, academic). Apply its rules below while writing. Load at most one secondary context. 2. Immediately before producing the final text, re-read the six failure modes. Rules read early in a long session fade. This step exists for that reason. 3. Self-scan against the six tests. One instance may stand. Repetiti