Edit AI Draft For Human Voice
Revise an AI-assisted text, from a single paragraph to a full story, essay, email, or post, in English or Italian, so it reads less like default model prose without adding any fact, changing structural choices first and surface style second, then verify the change with a bundled signal measurement and a closed-option structural rubric. Use when a writer owns a draft and wants it to sound like a person, not to conceal required AI disclosure or pass a detector.
Runs on
What it does
Revise any AI-assisted text, from a paragraph to a story, toward a human voice without adding facts.
When it fires
- implementation
- review
How it behaves
Measure surface and narrative-proxy signals in a text or web page, compare a revision with its draft, validate a structural rubric, and write JSON or HTML reports.
It does not promise or test that a text will pass an AI detector, and it never calls the result undetectable.
It does not rewrite text to conceal AI use where the author states, or the context shows, that disclosure is required.
It does not invent names, places, works, quotes, numbers, or anecdotes to make a draft look specific.
It does not change plot, claims, or required wording without the author's authorization.
Configuration
scripts/measure-voice-signals.mjs
- Version
- 1.0.0
- Requires
- node
- Licence
- RiskThis skill edits a draft the author supplies and runs a local helper that reads given files and, on request, fetches one public web page; it does not publish text or claim detector results.
- Low
- ModeHow it works: fixed rules, model instructions, or both.
- Deterministic and prompt
- Behaviour
- Advises, never blocks
- On failure
- Stops and reports
- LatencyThe skill runs a bounded local measurement before and after a revision in sequential steps, with input capped at 512 KiB.
- Bounded
- Data retentionWhether data remains after use and who manages its storage.
- Managed by the runtime
Often used with
All in Writing