Digital Innovation Agents

Appearance

Humanizer ​

/humanizer is the writing-style enforcer. It removes signs of AI-generated text from any artifact (BA, FEAT spec, ADR, README, commit message, release notes) so the result reads like a human wrote it. The skill is based on Wikipedia's Signs of AI writing guide.

When to use ​

Trigger phrases that route to the skill: "humanize this", "remove AI tells", "make this sound less like ChatGPT", "/humanizer".

The other DIA skills also reference these rules in their writing output, so the worst patterns rarely make it into a final artifact in the first place. /humanizer is the explicit pass when:

  • A draft was generated quickly and reads stilted
  • A team member pasted AI-generated copy into the repo
  • A release note or README needs final polish
  • A /consistency-check Mode B run flagged style drift

What it removes ​

Punctuation ​

  • No em dashes (U+2014), no en dashes (U+2013). Use periods, commas, parentheses, "and", or "but" instead. Em dashes are the most reliable AI tell.
  • No rule-of-three lists where one example will do. "fast, reliable, and scalable" reads like a marketing brochure. Pick the one that matters.

Vocabulary ​

The skill flags and rewrites the following words on sight:

WordWhy it gets flagged
landscapeVague. What landscape?
nuancedHedge. Say what the actual subtlety is.
delveBot tell. Use "look at" or "examine".
leverageUse "use".
crucialEmpty intensifier.
robustAlmost always meaningless.
seamlessMarketing word.
holisticMeans nothing on its own.
fosterUse "build" or "create".
ensuringUse "so that" or split the sentence.
highlighting / underscoringBot transitions. Just say it.

Sentence patterns ​

  • No negative parallelisms. "Not X but Y" is an AI tell. Say "Y" directly, or contrast both with concrete reasons.
  • No inflated symbolism. "This represents a paradigm shift" almost always means "this is new". Say what is new.
  • No vague attributions. "Critics argue" without naming the critic. Either name them or drop the sentence.
  • No superficial -ing analyses. "highlighting the importance of..." is bot transition glue. Cut it.
  • No filler phrases. "It is important to note that", "It goes without saying", "In today's fast-paced world".
  • No meta-signposting. "In this section we will explore..." Just explore.

Voice and tone ​

  • Active voice. "The agent updates the backlog" beats "The backlog is updated by the agent".
  • Sentence case in headings. "How the guide works", not "How The Guide Works".
  • First person where it fits. "I keep coming back to..." signals a real person thinking. Avoid in objective reference docs; encourage in tutorials and PR descriptions.
  • Vary sentence length. Short. Then longer ones that take their time. Algorithms produce uniform rhythm.

Umlauts ​

In German artifacts:

  • Correct umlauts (ä, ö, ü, ß) when the project file is Unicode-clean.
  • ASCII fallback (ae, oe, ue, ss) only when the existing project file uses ASCII consistently. Do not mix.

Voice calibration (optional) ​

If the user provides a writing sample (their own previous writing), the skill matches the rewrite to that voice instead of the default. Useful for personal blog posts, internal team docs, or any artifact where the author has a recognisable style.

Inline: "Humanize this text. Here's a sample of my writing for voice matching: ..."

File: "Humanize this text. Use my writing style from ./samples/ as a reference."

What it does not do ​

  • It does not "improve" technical accuracy. If the underlying claim is wrong, fix the claim first, then run the humanizer.
  • It does not enforce length. A 300-word section may shrink to 200 words after the AI tells are stripped, but the skill does not aggressively cut.
  • It does not change the language of the artifact. German stays German, English stays English.

Integration with other skills ​

Every other DIA skill follows the same writing rules in its output. The /humanizer skill is the explicit pass when something slipped through, or when older content needs cleanup.

The rules also apply to:

  • Commit messages (every commit, every phase)
  • Release notes (the Closing Handoff)
  • README and documentation
  • Skill prose itself (this very file)

If you read a DIA artifact and want to know whether it follows the rules, run a quick grep:

bash
grep -nP "[\x{2014}\x{2013}]|landscape|nuanced|delve|leverage|crucial|robust|seamless|holistic|foster|ensuring|highlighting|underscoring" {file}

Zero hits is the bar.

Read the skill file ​

skills/humanizer/SKILL.md on GitHub.

See also ​