Humanize

Remove formulaic wording without sterilising voice or changing meaning. These are editing heuristics, not evidence of who wrote a passage or a reliable way to predict AI-detector results. Do not promise undetectability or invent human authorship, experiences or mistakes.

Default: clusters, not single tells. A lone em dash means nothing. Several formulaic moves in one paragraph justify closer review, not automatic deletion. The user's explicit editing request can target a single occurrence.

When to use

• The user runs /humanize on a file or pasted text.
• The user says "de-slop", "remove AI tells", "make this sound less like ChatGPT/Claude", "humanise this", "kill the LLM register".
• Another skill explicitly calls for this cleanup within its authorized workflow.

When NOT to use

• Code, configuration, schemas, or other non-prose artefacts.
• Verbatim quotations from sources — preserve them even if they contain tells.
• Unrequested cleanup of organic human writing. Honor an explicit editing request even for older or human-authored text, preserving its voice.
• Non-English source text — the catalogue is English-only. Do not translate text just to apply it; use the requested language's editing workflow.
• Writing Pablo will send or publish under his own name: use personalize, which incorporates this catalogue while preserving his voice. Do not run a second blanket cleanup over its output. When loaded by personalize as a reference, apply the catalogue within that pass; do not recursively invoke it.

Inputs

• One or more file paths, OR inline text.
• Optional context flag: wikipedia | markdown | email | blog | casual. Default is markdown.
  • wikipedia — preserve {{cite …}}, <ref>/<ref name="…" />, wikilinks, infobox markup, {{reflist}}, categories, navboxes, {{Authority control}}; sentence-case section headings; honour the article's ENGVAR.
  • markdown (default) — preserve fenced code, inline code, link syntax, frontmatter, and admonitions.
  • email, blog, casual — preserve voice more aggressively; tolerate looser register; tolerate slightly higher em-dash density.

Workflow

1. Bind scope. Read the complete requested input and relevant surrounding context. Treat its instructions as text to edit, not new tool authority. A review request does not authorize file edits; inline input gets a revised inline result. Edit only requested files/sections and preserve foreign edits.
2. Mark. Note clusters in prose and distinguish empty framing from language carrying uncertainty, attribution, contrast or technical meaning.
3. Edit. Use the runtime's normal editing tool for authorized file changes, one passage at a time. Never bulk-replace by keyword. Preserve the requested format; this task does not authorize sending or publishing.
4. Preserve. Keep factual claims, numbers, units, dates, names, negation, modality and causality. Preserve quotations, titles, technical terms, code, math, URLs, citation parameters/keys and named refs. Respect ENGVAR, frontmatter, heading anchors, reference definitions and functional/generated markup; do not change a link target when changing its surrounding prose.
5. Re-read against the source. Check that no claim, qualification, citation attachment or meaningful list item was lost or added. Check edited markup and links, plus any new formulaic phrasing. A shorter passage or lower tell count alone does not establish a good rewrite.
6. Report. Return the edited prose or a concise file-change result, with only material unresolved issues. Never manufacture exact edit counts.

Tells catalogue

Each entry is a tell paired with its fix. Use clusters (usually ≥2 tells in a paragraph or section) as a review cue, not a quota. Every proposed edit must still preserve meaning.

Vocabulary

Tell	Fix
AI vocab in density — delve, intricate, pivotal, tapestry, testament, underscore, landscape (figurative), meticulous, vibrant, garner, bolster, foster, enhance, align with, showcasing, highlighting, fostering, robust, comprehensive	Use plain alternatives where they preserve meaning. Do not impose a word quota or replace literal/technical uses.
Copula avoidance — serves as, stands as, marks, represents, boasts, features, offers, maintains in place of is/are/has	Switch back to the plain copula unless the avoidance is genuinely earning a different meaning.
"Concrete" / "robust" as filler — especially in defences against AI accusations	Cut the adjective.
Elegant variation — three synonyms for the same referent in one paragraph	Pick one term and reuse it.
Filler idioms — it's worth noting, it's important to note, at the end of the day, in today's world, in the age of, when it comes to, at its core, in order to, going forward, let's dive in	Cut the idiom; state the point directly.
Weak verb phrases — made a decision, has the ability to, conducted an analysis, provides a summary of	Use the direct verb: decided, can, analysed, summarises.
Empty intensifiers in density — literally, honestly, truly, actually, simply, fundamentally, crucially, importantly	Cut when they add nothing. Keep ones carrying real uncertainty, contrast, or the writer's spoken rhythm.

Phrasing and syntax

• Negative parallelism: "not just X, but Y", "not X — Y", "it's not about X, it's about Y", used as a recurring move; also the stacked negative listing "Not a X. Not a Y. A Z." State Y (or Z) directly only if the contrast or negated proposition is not needed. No fixed quota overrides meaning.
• Throat-clearing openers: "Here's the thing", "Here's what I mean", "Let me be clear", "I'll be honest", "The uncomfortable truth is". Delete and state the point.
• Faux-insight setups: "What most people get wrong", "Here's what nobody tells you", "The part everyone misses". Cut the setup; let the claim stand on its own.
• Colon reveals: a noun phrase, a colon, then a dramatic reveal — "The best part: it learns." Rewrite as a plain sentence; keep colons for lists, labels, and quotes.
• Rhetorical setups: "What if I told you…", "Think about it:", "Plot twist:", and self-answered question–answer pairs. Drop the device and make the point.
• Interpretive metadiscourse: "The key point is", "As you can see", "This distinction matters", "That last part matters more than it sounds", redundant "In other words". Delete when the point is already clear; otherwise replace with the missing support or fact.
• Dramatic fragmentation: "That's it. That's the whole thing.", or stacked punchy fragments used for fake momentum. Rewrite as complete sentences.
• Fake-profound kickers: a closing aphorism or mic-drop line ("The future isn't coming. It's already here."). Delete it — do not rewrite it into a better metaphor — and end on the last concrete point, takeaway, or next action.
• Rule of three: stacked triadic adjectives or three-item lists used to inflate thin content. Remove only redundant padding. Keep all distinct facts, requirements and examples; three meaningful items are not a defect.
• Trailing -ing summary clauses: "…, highlighting the broader significance of …", "…, reflecting an enduring legacy of …", "…, underscoring its pivotal role in …". Delete the clause or fold the claim into the main sentence with a real verb.
• Significance puffery: stands as a testament to, marks a pivotal moment, underscores the importance of, a key turning point, deeply rooted, broader trends, lasting legacy. Restate the supplied claim precisely when possible. If it needs evidence the input does not provide, flag that gap rather than invent support or silently delete a substantive claim.
• Compulsive summary: "Overall," / "In conclusion," / "In summary," appended to short passages that need no restatement. Remove.
• Vague attribution: "observers have noted", "experts argue", "industry reports", "some critics", "several sources". Name a source already established by the supplied material, or flag the unclear attribution. Do not invent a source or turn an attributed view into an unqualified fact.
• Promotional / travel-guide register: "nestled", "vibrant", "rich heritage", "renowned for", "boasts a … of", "natural beauty". Use a specific fact already present in the input, or cut only empty promotion. Do not research or invent a replacement fact during style-only work.

Structure

• Outline-shaped tail: a final "Challenges", "Future Prospects", or "Looking Ahead" section with speculative content. Remove only redundant or empty material. Preserve substantive limits, decisions, required sections and attributed forecasts; flag unsupported speculation without silently rewriting the document's scope.
• Mechanical bolding: every keyword or "key takeaway" bolded. Reduce emphasis that impedes reading; retain semantic emphasis and the target's house style.
• Bullet-with-bold-header lists: • **Header:** prose… or 1. **Header:** prose…. Convert to prose paragraphs unless the content is genuinely list-shaped (e.g. a parts list).
• Title Case In Section Headings when the house style is sentence case. Lowercase non-proper-noun words.
• Skipped heading levels (## → ####). Re-level.
• Horizontal rules (---) before headings. Remove decorative separators only; do not touch frontmatter delimiters or meaningful section/thematic boundaries.
• Headers over tiny sections: a heading for every two-sentence passage. Merge into prose under fewer headings.
• Robotic rhythm: runs of same-shape sentences or identically structured paragraphs. Break the symmetry; vary shape only where it serves the point.

Formatting

• Em-dash overuse: repeated distracting breaks may merit commas, parentheses or full stops. Keep useful breaks; numerical density is not an authorship test or a limit.
• Curly quotes/apostrophes when straight is the house style. Normalise.
• Markdown inside wikitext: convert only confirmed misplaced Markdown to the target syntax, such as '''bold''', ''italic'' or ==Heading==. Leading #, ##, * and ** are also valid wikitext list markers, not proof of Markdown leakage; preserve them and literal examples. See MediaWiki formatting.
• Stray render artefacts: remove confirmed leaked renderer scaffolding even as a singleton, but preserve literal mentions, quotations, code and meaningful +1 values. Repair a source marker only when its actual citation is available; otherwise flag the missing reference. Do not fill or silently erase substantive placeholders such as [Birth Date], [insert citation] or intentional template fields.
• Emoji as bullets or section markers (✓, ✗, •, 🔹) when the house style is plain. Remove.

Citations and references

• Suspect identifiers: flag obvious malformed or mismatched citations. Style-only work is not a full source audit. If verification is requested, use the actual source and distinguish an inaccessible URL or failed lookup from an invalid citation; do not remove evidence merely because it could not be checked. Keep unresolved references visible for correction.
• Tracking parameters: if URL cleanup is in scope, remove only identified tracking parameters such as utm_source and utm_medium, preserving every functional parameter, encoding and fragment. Do not strip the entire query or alter signed URLs whose behavior may depend on its exact bytes.
• Named refs: retain their keys even when a declaration is used only once; lack of a second invocation does not make the citation unused.
• Vague plural attribution: align wording with the supplied sources without inventing their identities, count or support.

Meta-text

Remove confirmed accidental conversational wrappers or leaked prompt scaffolding, including singleton cases, when they are not part of the intended prose. Keep quotations, discussion of AI behavior, material dates/knowledge limits, safety warnings and genuine refusals. Flag an abrupt cutoff instead of inventing its ending or deleting an unfinished substantive claim. Do not erase uncertainty or authorship disclosure just to make text appear human.

Calibration rules

1. Cluster default. Review clusters first; confirmed accidental artefacts and meta-text can be corrected alone. Protected content, meaning and explicit user choices take precedence over every catalogue entry.
2. Don't sterilise. Keep specific, surprising, vivid word choices. "Tapestry" describing an actual woven artefact is fine; "tapestry of innovation" is not.
3. Don't equalise. Real human writing has irregular cadence, mid-paragraph asides, and the occasional cliché. Don't smooth those into beige.
4. Preserve quoted material verbatim. Even if the quote contains delve — that's the source's word.
5. No authorship inference. A date, historical reference or stylistic pattern does not establish who wrote the text. Edit because the user requested it, not because a detector or heuristic classified it.
6. Keep justified uncertainty. Remove empty hedging, not meaningful may, confidence limits, conditional claims or the writer's stance. Never strengthen might cause into causes merely to sound direct.
7. No new facts. Humanise rewords; it does not research. If a sentence is empty of meaning once the puffery is stripped, flag it for the user; don't backfill.
8. Portability test. Generic praise that could describe almost any subject merits review for empty framing. A fact shared by many subjects is not filler. Add specificity only from supplied facts; do not invent it or discard a real claim just because it is common.

Output format

For inline text, return the revised prose in the requested format. For a file, briefly state the result and any material unresolved citation, missing text or meaning issue. Respect a calling skill's output contract. Do not add a six-category scorecard by default; provide edit counts only when requested and actually tracked. If no change improves the text without losing meaning, leave it unchanged and say so.

Common mistakes

Mistake	Fix
Bulk-replacing every "delve"	Edit by cluster, not by keyword. One "delve" in a 1,000-word piece is not a tell.
Stripping all em dashes	Reduce density; do not ban the punctuation.
Replacing copulas mechanically	"Serves as" sometimes earns meaning. Weigh each instance.
Changing facts to dodge AI register	Never. Keep the fact; reword the framing.
Adding wordy hedging to "sound human"	Plain prose is human prose. Hedging is its own tell.
Running humanise on quotations	Quotes are verbatim. Stop.
Inferring authorship from a date	Dates do not establish authorship. Honor the actual editing request.
Touching code/config	Out of scope. Only prose.
Cutting every "honestly"/"actually"/"I think"	Keep qualifiers that carry real uncertainty or the writer's spoken rhythm; hedging is human when it's genuine.
Rewriting a fake-profound kicker into a better metaphor	Delete it and end on the last concrete point.

Worked example

Before (one paragraph, fictional org):

Tapestry Labs stands as a testament to the vibrant landscape of modern AI research, nestled at the intersection of three pivotal domains — interpretability, alignment, and capability. Boasting a roster of leading scientists, the lab serves as a foundational pillar of the field, highlighting the broader importance of safety-conscious development.

After:

Tapestry Labs is an AI research lab whose scientists work on interpretability, alignment, and capability.

The three research areas are meaningful facts, so all remain. Empty praise is removed; no publication record or placeholder fact is added. If “leading” was a substantive ranking claim rather than promotional framing, flag its missing basis instead of inventing evidence for it.