Rewrites AI-generated copy to remove the most common AI writing tells, overused words, hollow phrasing, em dashes, inflated language, without changing the meaning, length, or intent of the original text.
SaferSkills independently audited humanize (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Rewrites AI-generated copy to remove detectable AI writing patterns. The goal is to make text sound like it was written by a specific human with a clear voice, not by a committee, not by a marketing bot, not by a language model trained to please everyone at once.
Do not change the meaning. Do not change the intent. Do not shorten or lengthen the text beyond what's necessary to remove an AI pattern. Every sentence that comes out must say the same thing as the sentence that went in, just in a way that sounds like a person wrote it.
When the user provides text to humanize:
If no text is provided, ask: "Paste the text you'd like me to humanize and I'll get to work."
Work through the text systematically, applying all of the following fixes. Track every category of change made for the report at the end.
Replace or remove the overused AI words and phrases catalogued in references/ai-vocabulary.md. Load that reference at the start of any humanize pass and apply every replacement that matches text in the input.
The reference is the authoritative vocabulary list. Do not rely on memory; consult it on every pass, since the list may have been updated since the last run.
Apply this principle from the reference: do not simply swap one AI word for another. Rewrite the sentence to avoid the concept being performed in an AI way.
Negative parallelisms, "Not just X, but Y": Rewrite any "Not only... but also...", "It's not just..., it's...", or "Not X. Rather, Y." constructions. These are a strong AI tell. Combine into a direct statement.
Example:
Rule of three overkill: If a sentence lists exactly three adjectives or three parallel phrases in a row purely for rhetorical effect, cut it to two, or collapse into a single precise word.
Em dash elimination: Replace ALL em dashes with commas, colons, parentheses, or sentence breaks. No em dashes should remain in the output. This is a hard rule with no exceptions. Em dashes are one of the strongest AI tells and are to be removed entirely.
Trailing "-ing" phrases of superficial analysis: Cut or rewrite sentences that end with a present participle phrase tacked on to explain significance:
Closing summary sentences that restate the paragraph: Delete or integrate. Humans don't summarize each paragraph they just wrote.
Throat-clearing openers:
Remove or deflate language that overstates importance without evidence:
AI avoids "is" and "are" by substituting more elaborate verbs. Restore plain constructions where they're clearer:
Only restore where it genuinely simplifies. Don't flatten deliberate stylistic verb choices.
Delete any text that sounds like the AI is talking to the user rather than writing for an audience:
Delete phrases like:
If the claim is needed, rewrite it as a plain factual hedge: "Details aren't publicly confirmed" instead of AI boilerplate.
Before outputting the rewritten text:
If yes to all, output the text.
Present the rewritten text cleanly, without preamble.
Then, after the text, provide a Change Report in this format:
What was changed:
List each category of change that was applied, with a brief example of the most notable substitution from that category. Use plain language. Be specific but concise.
Example format:
Then ask:
Any further edits? Or would you like to download this as a file, and if so, which format: Word (.docx), plain text (.txt), or Markdown (.md)?
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.