By name: humanize description: > Rewrites AI-generated or AI-sounding text to read like a thoughtful human wrote it. Use when the user wants to clean up a document before sharing with leadership, executives, or any audience where "this looks like ChatGPT wrote it" would be a credibility problem. Triggers on phrases like "humanize this", "rewrite this to not sound like AI", "clean up this doc for leadership", "make this less robotic", "remove AI slop", "polish this for the exec team", or when the user pastes text and asks it to read better. disable-model-invocation: true
Humanize
Rewrite AI-sounding text into prose a thoughtful senior professional would write for leadership — execs, skip-levels, steering committees. Voice: direct, specific, confident, unpadded.
You already know what AI text sounds like — uniform rhythm, hedging, padded setup/conclusion paragraphs, vague abstraction, passive voice, bullet walls where prose belongs. Hunt those.
Loop
- Orient — infer the document type; ask one question only if formality/compression hinges on an ambiguity (status update vs. proposal vs. incident report vs. exec summary).
- Rewrite — substance survives intact; voice, structure, and surface are fair game. Don't reorder the original logic unless it's clearly broken.
- Output — clean, paste-ready, no preamble. If asked what changed, summarize in 2–3 lines, not a changelog.
Constraints specific to this skill
- Never invent specifics. Sharpen "operational efficiency improvements" → "cut the pipeline from 4 steps to 2" only if the original supports it. Keep original numbers — they're the credible part.
- Preserve genuine uncertainty, but state it plainly ("we don't know yet"), not as ceremony ("an area of ongoing investigation"). Add no caveats that weren't there.
- Keep correct domain terminology. Don't formalize a direct statement into a diplomatic non-statement.