humanizer
One markdown file with 33 named patterns of AI writing, each with a before and an after, taken from Wikipedia's "Signs of AI writing". Two things make it more than a word blacklist: it has an **embedded mode**, where another agent runs it as the last step of writing a commit, a PR or a doc, and a **voice calibration** step, where a sample of your own writing overrides the skill's own rules. It also carries a list of what NOT to flag, which is the part most "anti-AI" prompts are missing.
By Siqi Chen (blader) · MIT
Not ours, and we don't ship a copy of it — that way updates and security questions stay with the author.
npx skills add blader/humanizer --global
- Proof
- “The final rewrite contains no em dashes (—) or en dashes (–).” — SKILL.md § 14. MIT licence and 32,720 stars verified at the source on 2 August 2026.
- Can't do
- It ships no checker. The repo is a prompt plus a packaging validator, so nothing verifies the result except the model that just wrote it — and the em dash ban is a style rule, not a fact about your text. Run a counter over the output if you want a number you can check.
Playbook — how to actually use it
A skill without instructions is a download, not a capability. This is the part we wrote: what the commands do, which one to reach for, and how to phrase the request so it lands.
Or hand it straight to your agent
Drops the same guide next to the skill as PLAYBOOK.md — plain markdown, so the agent reads it along with the skill.
curl -sL https://stackscout.app/playbooks/humanizer.md -o humanizer/PLAYBOOK.md
The full playbookRead it ↓Collapse ↑
✦ or: the whole library at once
Skip the downloads — one MCP connector, every skill right inside your agent, always current.
See the MCP