As AI text becomes ubiquitous, cleaning stylistic "AI tells" without inventing facts has turned into a common editorial task for teams using generative models. Humanizer treats this as an agent-level skill: instead of a standalone editor, it's a portable instruction artifact that an agent can call to detect and rewrite dozens of recurring AI patterns while keeping the author's facts and intent.
What Sets It Apart
- Pattern-driven audit + rewrite: encodes a catalog of 33 observable AI-writing patterns (with before/after examples) and applies a two-pass workflow—an audit to surface remaining AI tells, then a second rewrite to catch what the first pass missed. So what: yields more robust cleanups than one-shot paraphrasing.
- No‑fabrication policy and voice calibration: rewrites never add names, dates, or facts not present in the input, and can adapt sentence rhythm and word choices to a short sample of a user's writing. So what: preserves factual integrity and maintains author voice instead of producing generic copy.
- Agent-native portability: provided as plain SKILL.md and packaged for common skill/harness ecosystems (Skills CLI, plugin entrypoints), letting many agent runtimes invoke it the same way. So what: minimal integration effort for teams that orchestrate actions via agent skills.
Who it's for and tradeoffs
Great fit if you operate agents that produce user-facing prose and need automated hygiene (docs, marketing drafts, community posts) without factual augmentation. It works best when the source text contains sufficient specifics for a faithful rewrite and when an agent runtime supports skill-style instructions.
Look elsewhere if you need the tool to invent missing specifics, perform fact-checking or add substantive domain knowledge—the skill explicitly refuses to fabricate. Also, its effectiveness depends on the agent harness’s ability to pass contextual prompts and files; it is not a standalone grammar checker or detector of sophisticated adversarially polished AI text.
