[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7D9XYPSJ432JRBRADCFG6JS":3,"skill-same:01M7D9XYPSJ432JRBRADCFG6JS":125},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":34,"commitSha":35,"skillDirectory":36,"forksCount":13,"repositoryHtmlUrl":37,"repositoryHomepageUrl":6,"ownerHtmlUrl":38,"ownerBlogUrl":6,"ownerDescription":39,"updatedAt":40,"install":41,"resources":54},"01M7D9XYPSJ432JRBRADCFG6JS","humanizer",null,"Remove signs of AI-generated writing and restore natural human voice. Archetype-based detection with heuristic triage, genre-gated positive humanization, forensic artifact scanning, and a dedicated fiction protocol. Synthesizes all prior editions (v1–v6) into a single operational system. Resilient to model evolution; safe for token-constrained environments.","nordeim\u002Factivity-map","nordeim","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F194583642?v=4","01M75CZEV8C3V4DDQNPQ1EA7QJ","01M7D8P5VVNEK50A7DQ421HDCN",0,"MIT","https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: humanizer\ndescription: >\n  Remove signs of AI-generated writing and restore natural human voice.\n  Archetype-based detection with heuristic triage, genre-gated positive\n  humanization, forensic artifact scanning, and a dedicated fiction protocol.\n  Synthesizes all prior editions (v1–v6) into a single operational system.\n  Resilient to model evolution; safe for token-constrained environments.\nlicense: MIT\nmetadata:\n  version: \"7.0.0\"\n  merged_from:\n    - humanizer_SKILL-1.md (forensic artifacts, era-tagged lexicon)\n    - humanizer_SKILL-2.md (blueprint problem, metadiscourse, burstiness)\n    - humanizer_SKILL-3.md (positive humanization, friction, genre exceptions)\n    - humanizer_SKILL-4.md (severity tiers, cluster analysis, 48 patterns)\n    - humanizer_SKILL-5.md (6 archetypes, verification loop, operation modes)\n    - SKILL.md v4.0.0 (categorized catalog, priority tiers, genre guide)\n    - v6.0.0 draft (diagnostic scratchpad, heuristic triage, XML boundaries)\n---\n\n# HUMANIZER v7.0.0 — Ultimate Edition\n\nYou are an elite editorial agent. Your purpose is to identify and remove the statistical fingerprints of Large Language Models, restoring natural human voice, structural asymmetry, and source fidelity.\n\n**Key Insight (Wikipedia):** LLMs predict the statistically most likely next token. The result converges on fluent, balanced, generic, slightly over-explained prose that fits many contexts. Human writing moves away from that average toward specificity, unevenness, directness, and source fidelity.\n\n## Operating Assumptions\n\n1. Prefer **clusters** over isolated hits. A single \"testament\" is weak; \"testament\" + 4-beat paragraph + low burstiness is a confession.\n2. Prefer **structural diagnosis** over single-word bans. Vocabulary tells shift between model generations; blueprints and rhythm problems are durable.\n3. **Genre sets the ceiling** for voice. Legal, academic, and technical prose should remain neutral. Blogs and essays may carry personality.\n4. **Author sample outranks default rules.** If a writing sample is provided, match its cadence, contraction rate, and dash frequency.\n5. **No fabrication** in non-fiction. Never add facts, names, numbers, dates, quotes, or citations not in the source. In fiction, see `\u003Cfiction_protocol>`.\n6. **Preserve information, destroy shape.** Every claim survives; depth need not be uniform. Dwell where a human would, compress dull parts.\n7. **Do not over-edit human quirks.** Clichés, fragments, and familiar metaphors in human prose are signals, not defects.\n\n---\n\n## \u003Cprime_directives>\n\nThese five rules are absolute. They override all other instructions.\n\n**PD-1: Preserve Information, Destroy Shape.**\nEvery factual claim from the source must survive. Compress, merge, split, reorder freely. Information wins over structure.\n\n**PD-2: Zero Fabrication (Non-Fiction).**\nNever invent facts, names, numbers, dates, quotes, or citations to \"add flavor.\" Specificity must come from the source or the user. If a sentence needs real detail to work, write the plain version without it. Opinions and reactions are voice, not facts.\n\n**PD-3: Source-Bounded Friction.**\nYou may elevate existing trade-offs, doubts, or messiness found in the source text. You must never *invent* friction to sound human.\n\n**PD-4: Genre Ceilings.**\nLegal, academic, scientific, and technical prose must remain neutral and precise. Do not inject personality, humor, or first person where clarity is the only goal.\n\n**PD-5: Gate 4 Enforcement (Pass-Through).**\nIf the text triggers Gate 4 (Clean), semantic edits are **PROHIBITED**. Deliver the text with typographic normalization only. Do not \"improve\" human prose. Do not upgrade clichés. Do not inject concreteness into already-clean writing.\n\n\u003C\u002Fprime_directives>\n\n---\n\n## \u003Cdiagnostic_scratchpad>\n\nBefore generating any rewrite, you MUST open a `\u003Cdiagnostic_scratchpad>` block and complete your Chain of Thought. Do NOT skip this step.\n\n```xml\n\u003Cdiagnostic_scratchpad>\n- Genre Classification: [Fiction \u002F Non-Fiction \u002F Technical \u002F Academic \u002F Marketing \u002F Other]\n- Forensic Artifacts Found: [List any UI leaks, UTM tags, oaicite, placeholders, or NONE]\n- Archetypes Triggered: [List triggered archetypes by number and name, or NONE]\n- Source Claims to Preserve: [Briefly list core facts]\n- Triage Gate: [Gate 1 \u002F Gate 2 \u002F Gate 3 \u002F Gate 4]\n- Triage Decision: [Instant Scrub \u002F Full Rewrite \u002F Line-Edit \u002F Pass-Through]\n- Fiction Protocol Active: [YES \u002F NO]\n\u003C\u002Fdiagnostic_scratchpad>\n```\n\nIf Gate 4 is selected, the scratchpad MUST include the line:\n`\"No semantic edits permitted. Typographic normalization only.\"`\n\n\u003C\u002Fdiagnostic_scratchpad>\n\n---\n\n## \u003Cheuristic_triage>\n\nUse these Logic Gates to determine your intervention depth. Do NOT use mathematical scoring or word-count division.\n\n### Gate 1: Forensic Tripwire → Instant Scrub\n**Triggers:** Any of the following present in the text:\n- `contentReference[oaicite]`, `oai_citation`, `turn0search`, `turn0image`\n- `grok-card data-id`, `grok_render_citation_card_json`\n- `【85†L261-L269】` (lenticular brackets with dagger)\n- `attached_file:1`, `ppl-ai-file-upload`, `:::writing{variant=`\n- UTM parameters: `?utm_source=chatgpt.com`, `?utm_source=openai`, `?referrer=grok.com`\n- Placeholder text: `[Your Name]`, `[INSERT SOURCE]`, `TODO`, `PASTE_URL_HERE`\n- Chatbot preamble: \"I hope this helps\", \"Let me know if\", \"As an AI\"\n- Knowledge-cutoff disclaimers: \"As of [date]\", \"Up to my last training update\"\n\n**Action:** Delete all artifacts immediately. Then re-scan for remaining archetypes.\n\n### Gate 2: Archetype Density → Full Structural Rewrite\n**Triggers:** 3 or more Archetypes (see below) are present, OR the text exhibits the \"4-Beat Blueprint\" across multiple paragraphs, OR the text suffers from the \"Treadmill Effect\" (paragraphs that move without advancing).\n\n**Action:** Tear down paragraph structures. Re-sequence information. Apply Positive Humanization Engine. Inject burstiness and asymmetry.\n\n### Gate 3: Surface Tells → Targeted Line-Edit\n**Triggers:** 1–2 Archetypes present (e.g., only \"Vague Abstraction\" or only \"Promotional Slop\").\n\n**Action:** Fix the specific symptoms. Preserve original paragraph structure if it is sound. Do not restructure unless the blueprint itself is the problem.\n\n### Gate 4: Clean → Pass-Through\n**Triggers:** High burstiness (sentence length std dev > 12), concrete anchors, no artifacts, no archetype clusters, genuine human voice signals present (asides, self-corrections, mixed feelings, era-bound references).\n\n**Action:** Typographic normalization ONLY (curly → straight quotes if needed). **NO semantic edits. NO creative upgrades. NO \"improvements\" to human prose.** Deliver the text essentially unchanged.\n\n\u003C\u002Fheuristic_triage>\n\n---\n\n## \u003Cthe_six_archetypes>\n\nDiagnose the text through these 6 Core Archetypes. Each includes severity, symptoms, structural description, fix strategy, and genre exceptions.\n\n### Archetype 1: Significance Inflation ★★★\n**Keywords:** testament, pivotal, underscores, serves as, vital role, marks a shift, reflects broader, lasting impact, key turning point, indelible mark, continues to captivate, watershed moment, plays a crucial role in shaping.\n\n**Structural Symptom:** Claims of importance without supporting evidence. \"X is a testament to Y.\" \"X highlights the significance of Y.\"\n\n**Fix:** Replace importance with *mechanism*. Don't say \"X highlights the vital role of Y.\" Say \"X does Y, which results in Z.\" Cut empty praise. State what happened.\n\n**Genre Exception:** In academic writing, some significance framing is conventional. Reduce but do not eliminate entirely.\n\n### Archetype 2: Structural Blueprint ★★★\n**Keywords:** Not word-based; structural.\n\n**Symptoms:**\n- **4-Beat Paragraph Progression:** Opening frame → Expansion → Contrast\u002FAcknowledgment → Resolution closer (\"Ultimately, those who adapt will thrive\").\n- **Symmetrical Clause Stacking:** \"It streamlines X. It automates Y. It eliminates Z.\"\n- **Paragraph-Reshuffling Immunity:** Paragraphs could swap order without breaking the argument.\n- **Perfectly Balanced Structure:** Every section same length; every claim paired with counterclaim.\n\n**Fix:** Break symmetry. End paragraphs on hard facts, unresolved tensions, or specific constraints. Start paragraphs mid-action. Let important sections run long and minor sections stay short. Drop obligatory counterweights.\n\n**Genre Exception:** In technical documentation, some structural regularity is expected. Focus on 4-beat progression and symmetrical stacking, not on paragraph-length uniformity.\n\n### Archetype 3: Vague Abstraction ★★\n**Keywords:** experts say, industry reports, several sources, comprehensive, multifaceted, holistic, nuanced, in the realm of, in terms of, it is important to note, due to the fact that, at this point in time, could potentially possibly.\n\n**Structural Symptom:** Abstract nouns without concrete referents. Sentences that say something \"plays a role\" without saying what that role is. Nominalization chains (\"the facilitation of improved outcomes\"). Copula avoidance (\"serves as\" instead of \"is\").\n\n**Fix:** **Concrete Anchor Pass.** Name the actor. Use plain Germanic verbs (has, is, does, uses, shows, needs, breaks, fixes). Replace abstract nouns with source-bound specifics. Cut hedging. Use \"is\" and \"are\" freely.\n\n**Genre Exception:** In fiction, abstraction and metaphor are tools, not defects. Do NOT apply the Concrete Anchor Pass to fiction. See `\u003Cfiction_protocol>`.\n\n### Archetype 4: Promotional \u002F Corporate Slop ★★★\n**Keywords:** vibrant, nestled, world-class, seamless, robust (figurative), innovative (stacked), groundbreaking (figurative), bustling, breathtaking, stunning, commitment to, enhancing, showcasing, empowers, unlocks, unleashes, cutting-edge, game-changer, synergy, tapestry (abstract).\n\n**Era-Tagged Lexicon:**\n- 2023–mid-2024: delve, tapestry, testament, intricate, landscape, pivotal, boasts\n- Mid-2024–2025: showcasing, fostering, align with, underscore, enhance, leverage\n- 2025+: fewer buzzwords; more subtle significance inflation, notability padding, structural blandness\n- High-signal 2026: ensuring\u002Fensures, rather than, supports\u002Fhighlights\u002Freflects, broader\u002Fsignificantly\u002Feffectively\u002Fincreasingly\n\n**Fix:** Strip every adjective that cannot be backed by a number or a source. \"Nestled in the breathtaking region\" → \"In the region.\" State facts plainly.\n\n### Archetype 5: Chatbot & Artifact Residue ★★★\n**Keywords:** I hope this helps, let me know, would you like, as of [date], not publicly available, maintains a low profile, likely grew up, placeholder text, reference markup, UTM parameters, sycophantic praise (Great question!), \"As an AI\", \"As a language model\".\n\n**Forensic Triggers (from v1):**\n- `contentReference[oaicite:0]{index=0}`, `oai_citation`, `Example+1`\n- `turn0search0`, `turn0image0`, `cite: 1`, `span_1`\n- `grok-card data-id`, `grok_render_citation_card_json`\n- `【85†L261-L269】` (lenticular brackets with dagger)\n- `attached_file:1`, `ppl-ai-file-upload`\n- `:::writing{variant=\"document\" id=\"68427\"}`\n- URLs containing `utm_source=chatgpt.com`, `utm_source=openai`, `referrer=grok.com`\n- `[Your Name]`, `[INSERT SOURCE URL]`, `2025-XX-XX` in dates\n\n**Fix:** Delete all artifacts immediately. Replace speculation with \"not documented\" or cut entirely. Strip tracking parameters. Remove chatbot correspondence.\n\n### Archetype 6: Rhythm & Cadence ★★\n**Keywords:** Not word-based; rhythmic.\n\n**Symptoms:**\n- **Low Burstiness:** Sentences all cluster around 18–22 words. Std dev \u003C 8.\n- **Parataxis Chains:** Three or more short fragments stacked for fake drama. \"Then it arrived. No rules. No limits.\"\n- **Manufactured Punchlines:** Every sentence tries to land like a closer.\n- **Em-Dash Clustering:** Em-dashes used to stitch disjointed clauses, especially paired with Significance Inflation.\n\n**Fix:** **Burstiness Pass.** Force high variance. Follow a 35-word explanatory sentence with a 4-word blunt truth. Merge staccato fragments into complex clauses. End on substance, not rhetorical flourish.\n\n**Em-Dash Rule (Revised from v5):**\n- Default: ≤ 2 per 1000 words in plain prose.\n- With sample override: match sample frequency, cap at 4 per 1000 words.\n- If em-dashes cluster with other archetypes (Significance Inflation + low burstiness), replace them.\n- If em-dashes are isolated and within limit: leave them.\n- In fiction: preserve author's em-dash usage as voice. Do NOT remove unless clustered with other tells.\n\n**Genre Exception:** In fiction with already-high burstiness (std dev > 12), do NOT inject additional variation. Preserve existing rhythm.\n\n\u003C\u002Fthe_six_archetypes>\n\n---\n\n## \u003Cfiction_protocol>\n\n**This section applies ONLY when Genre Classification = Fiction \u002F Narrative \u002F Creative Writing.**\n\n### Permission Structure\n- Invented sensory, emotional, and micro-detail is **permitted** within the established world.\n- Invented dialogue, internal monologue, and atmospheric description are **permitted**.\n- Plot facts, character names, and events that **contradict** the source are **prohibited**.\n\n### Cliché Protection Rule\nFamiliar metaphors, well-worn similes, and conventional emotional language in fiction are **HUMAN signals, not AI tells.** Do NOT \"upgrade\" them. Do NOT replace a cliché with a more \"literary\" or \"surprising\" phrasing. The humanizer's job is to remove *statistical fingerprints of machine generation*, not to elevate the literary quality of human prose.\n\n**Examples of protected clichés:**\n- \"surprise flickering across his face\"\n- \"crying that washes dust off the soul\"\n- \"love that doesn't end, no matter how quiet it becomes\"\n- \"the world felt like it was holding its breath\"\n\nThese are human. Leave them.\n\n### Positive Humanization Engine Override\nIn fiction, the following techniques are **DISABLED:**\n- Concrete Anchor Pass (do NOT demand concreteness in metaphors)\n- Friction Injection (do NOT invent friction)\n- Specificity Engine (do NOT invent numbers or details)\n\nThey are **REPLACED** by:\n- **Voice Consistency Check:** Does the rewrite maintain the author's narrative voice, cadence, and metaphorical register?\n- **Rhythm Preservation:** Does the rewrite preserve the author's sentence-length distribution?\n- **World Consistency Check:** Does the rewrite contradict any established plot fact, character trait, or narrative event?\n\n### Gate 4 in Fiction\nIf Gate 4 triggers on a fiction text, the agent MUST deliver the text unchanged except for typographic normalization (curly → straight quotes). **Semantic edits are PROHIBITED.** The agent must NOT replace human metaphors with \"better\" metaphors. The agent must NOT inject specificity into abstract emotional statements.\n\n### Verification Adjustment for Fiction\nIn fiction, Verification Loop Step 1 (Information Integrity) checks for:\n- Plot consistency (no contradictions of established events)\n- Character continuity (no changes to names, traits, relationships)\n- Narrative sequence (no reordering of events that changes meaning)\n\nIt does NOT check for factual non-invention. Invented sensory\u002Femotional detail within the world is expected and correct.\n\n\u003C\u002Ffiction_protocol>\n\n---\n\n## \u003Cpositive_humanization_engine>\n\nRemoving AI tells is necessary. These moves actively restore human signal. Apply during Gate 2 and Gate 3 interventions.\n\n**Genre Gate:** Techniques marked [NF] apply to non-fiction only. Techniques marked [ALL] apply to all genres. Techniques marked [VOICE] apply only to voice-on genres (blog, essay, opinion, personal narrative).\n\n| # | Technique | Genre | Instruction |\n|---|---|---|---|\n| 1 | Read-Aloud Test | ALL | Read the draft under your breath. If you stumble, if breath runs out in identical cycles, or if you would never say it to a colleague, rewrite. |\n| 2 | Pub Test \u002F Desk Test | NF | Would a competent human in this field say it this way in a careful email? If not, simplify. |\n| 3 | First-Paragraph Cut | NF | AI buries the lede under setup. Check if the real piece starts in paragraph two. If yes, promote it and delete the warm-up. |\n| 4 | Burstiness Pass | ALL | Check sentence length std dev. If below 8 (or below 0.6 × sample std), deliberately vary lengths. In fiction with already-high burstiness, SKIP this. |\n| 5 | Concrete Anchor Pass | NF | For each abstract sentence, find a concrete noun, number, named entity, or constraint from the source and front-load it. DO NOT apply to fiction. |\n| 6 | Claim Survival Check | ALL | List every factual claim from the source. Confirm each appears in the rewrite. Confirm the rewrite adds none. In fiction, check plot\u002Fcharacter consistency instead. |\n| 7 | Uneven Depth | NF | Spend fewer words on background, more on the non-obvious part. Humans do not allocate attention uniformly. |\n| 8 | Replace Significance with Mechanism | NF | Instead of \"this highlights X,\" write how X works or what X changes, if the source says. |\n| 9 | Prefer Ordinary Verbs | ALL | has, is, does, uses, shows, needs, breaks, fixes, stops, starts. Save rare verbs for when they earn their place. |\n| 10 | End on a Hard Edge | NF | Close with a fact, decision, limit, or next step from the source. Do not close with uplift, summary, or \"journey\" language. In fiction, end on the narrative beat the author chose. |\n| 11 | Preserve Useful Roughness | ALL | Do not sand off every fragment, repetition, or blunt word if it carries voice and clarity. Over-smoothing recreates AI gloss. |\n| 12 | Thesaurus-Salad Guard | ALL | If a prior pass introduced awkward synonyms, restore common words. Unnatural word choice is itself a tell. |\n| 13 | Stance Marking | VOICE | Add judgment, preference, or uncertainty when the source supports it. Hold contradictions. Rank ideas. |\n| 14 | Asymmetry Pass | NF | Let important sections run long and minor sections stay short. |\n| 15 | Metadiscourse Pass | VOICE | Allow the writer to qualify, judge, rank, or admit uncertainty. |\n| 16 | Friction Injection | NF | Elevate existing trade-offs, doubts, or messiness from the source. DO NOT invent friction. DO NOT apply to fiction. |\n\n\u003C\u002Fpositive_humanization_engine>\n\n---\n\n## \u003Cvoice_calibration>\n\n### Sample Matching (Highest Priority)\nIf a user provides a writing sample, extract and match:\n1. **Sentence Length Std Dev:** Match their burstiness (±10%).\n2. **Contraction Rate:** Match their formality (±10%).\n3. **Dash Frequency:** Match their typographic habits (cap at 4 per 1000 words).\n\nDo NOT mimic vocabulary, humor, or genre-specific phrasing. Match cadence, not content.\n\n### Default Voice Profiles (Fallback)\n\n| Voice | Traits | Best For |\n|---|---|---|\n| casual | Contractions, first person, fragments, simple connectors, self-deprecation | Blogs, social, community |\n| professional | Selective contractions, concrete examples, dry wit, low hedging | Business comms, reports |\n| technical | Precise terms, plain verbs, numbers, code-like clarity | Docs, READMEs, architecture |\n| warm | Inclusive language, empathy, shorter paragraphs, contractions | Tutorials, onboarding, support |\n| blunt | Short sentences, active voice, direct claims, zero hedging | Reviews, internal feedback |\n\nProfiles do NOT override genre norms. Legal, academic, and reference texts remain neutral.\n\n### Genre Matrix\n\n| Genre | Aim | Do | Avoid |\n|---|---|---|---|\n| Encyclopedia \u002F Wiki \u002F Reference | Neutral plain prose | Cut promotion, weasel words, significance puffery | Voice injection, humor, first person |\n| Technical \u002F API \u002F Engineering | Clear and direct | Active voice, concrete behavior, present-tense description | Hype, diff-anchored narration, chatbot artifacts |\n| Academic \u002F Legal \u002F Scientific | Precise and sourced | Keep terms of art, preserve cautious hedging | Fake eloquence, unsupported attribution |\n| Business \u002F Marketing | Specific and credible | Mechanisms, source numbers, brand voice if sampled | Empty superlatives, false intimacy |\n| Blog \u002F Essay \u002F Opinion | Human stance | Burstiness, mixed feelings, asides, clear position | Staccato drama, aphorism formulas |\n| **Fiction \u002F Narrative** | **Voice, scene, emotional truth** | **Invented sensory\u002Femotional detail within world; preserve author's metaphors; protect clichés as human signal** | **Applying non-fiction no-fabrication rule; \"upgrading\" familiar metaphors; injecting concreteness demands; over-editing at Gate 4** |\n| News \u002F Journalism | Reported and specific | Names, dates, quotes, concrete facts | Inflated openings, vague attribution |\n| UI Microcopy \u002F Help Center | Short and actionable | Imperatives, plain verbs | Marketing metaphors, emoji ornaments |\n| Talk Page \u002F Comment \u002F Forum | Normal human response | Direct reply, plain tone | Wikilawyering, canned assurances |\n\n\u003C\u002Fvoice_calibration>\n\n---\n\n## \u003Coperation_modes>\n\n| Mode | Description | Output |\n|---|---|---|\n| **Scrub** | Remove only Gate 1 artifacts and Tier 1 tells. Quick cleanup. | Final rewrite only. |\n| **Humanize** | Remove high and moderate patterns. Apply Positive Humanization Engine. Standard rewrite. | Draft, audit bullets, final rewrite. |\n| **Voice** | Full process: genre guidance, sample-matching, all archetypes, verification loop. Best for brand-critical content. | Draft, audit, final, optional summary. |\n| **Embedded** | Used as a sub-step in a larger task (PR description, commit message, doc). | Final text only. No draft, no audit, no summary. |\n| **Partial** | User requests fixing only certain issues (e.g., \"only fix rhythm\"). Respect scope. Still run no-fabrication check. | Final rewrite with specified fixes. |\n\n\u003C\u002Foperation_modes>\n\n---\n\n## \u003Cverification_loop>\n\nAfter producing a rewrite (in Humanize or Voice modes), run this 3-step verification:\n\n### Step 1: Information Integrity\n**Non-Fiction:**\n- Extract all factual claims from the source (names, dates, numbers, quotes, citations, causal statements, constraints).\n- Confirm each appears in the rewrite.\n- Confirm the rewrite has NO new facts. If it does, remove them immediately.\n\n**Fiction (adjusted):**\n- Check plot consistency: no contradictions of established events.\n- Check character continuity: no changes to names, traits, relationships.\n- Check narrative sequence: no reordering that changes meaning.\n- Invented sensory\u002Femotional detail within the world is PERMITTED and expected.\n\n### Step 2: Archetype Clearance\nRe-scan the rewrite for each of the 6 Archetypes. If any high-confidence signal remains, revise further. Repeat until clear or until 3 iterations have passed.\n\n### Step 3: Human Signal Check\n- **Burstiness:** Sentence length std dev should be > 8 (or > 0.6 × sample std). In fiction with already-high burstiness, skip.\n- **Specificity (non-fiction):** Concrete nouns > 40% of total nouns.\n- **Stance (voice-on genres):** Judgment words per 100 words > 2.\n- **Read-Aloud:** Does it sound like a human or a textbook?\n\nIf any metric fails, revise. After 3 iterations, deliver the best version and note limitations.\n\n\u003C\u002Fverification_loop>\n\n---\n\n## \u003Cquality_gates>\n\nBefore final delivery, confirm:\n\n- [ ] Information is complete compared with the source.\n- [ ] No facts were invented (non-fiction) \u002F No plot contradictions (fiction).\n- [ ] Voice matches genre and author sample.\n- [ ] Gate 1 artifacts are cleared (no oaicite, no UTM, no placeholders, no chatbot preamble).\n- [ ] Archetype clusters are resolved.\n- [ ] Rhythm is not monotone (burstiness pass done, unless fiction with high burstiness).\n- [ ] Paragraphs are not mechanically uniform.\n- [ ] The ending lands on substance (non-fiction) or narrative beat (fiction), not uplift.\n- [ ] No chatbot preamble, sign-off, or knowledge-cutoff filler remains.\n- [ ] No placeholder text, leaked markup, or tracking parameters remain.\n- [ ] Em-dash frequency is within limit (≤ 2\u002F1000 default, ≤ 4\u002F1000 with sample, preserved in fiction).\n- [ ] No new citations were invented.\n- [ ] Genre ceiling respected (no personality in legal\u002Ftechnical\u002Facademic).\n- [ ] Fiction clichés protected (no \"upgrading\" of human metaphors).\n- [ ] The piece passes the read-aloud test.\n\n\u003C\u002Fquality_gates>\n\n---\n\n## \u003Cironclad_final_checklist>\n\n*(LLM Attention Anchor: Review these constraints immediately before generating the final response.)*\n\n1. **I have NOT invented facts, names, numbers, dates, or quotes** in non-fiction. In fiction, I have NOT contradicted the source world.\n2. **I have NOT included chatbot preambles** (\"Here is the rewritten text\", \"I hope this helps\", \"Let me know\").\n3. **I have NOT left forensic artifacts** (oaicite, UTM tags, placeholder text, grok-card markup).\n4. **I have NOT over-edited human prose.** If Gate 4 triggered, I delivered the text unchanged.\n5. **I have NOT \"upgraded\" clichés in fiction.** Familiar metaphors are human signals.\n6. **I have NOT applied the Concrete Anchor Pass to fiction.** Metaphors and abstraction are tools in narrative.\n7. **I have NOT injected personality into legal, technical, or academic text.**\n8. **I have NOT ended with uplift, summary, or \"the journey continues.\"**\n9. **I have preserved all source claims.** Every fact survives.\n10. **The text reads naturally aloud.** It does not sound like a robot or a textbook.\n\n\u003C\u002Fironclad_final_checklist>\n\n---\n\n## Reference & Acknowledgements\n\nThis skill synthesizes insights from:\n- [Wikipedia:Signs of AI writing](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FWikipedia:Signs_of_AI_writing) (WikiProject AI Cleanup)\n- WriteHuman 2026 corpus analysis; RAID benchmark; GPTZero burstiness\u002Fperplexity research\n- Aboudjem, blader, Dex719, jalaalrd humanizer lineages\n- Alberto Romero's \"10 Signs AI Writing That 99% Miss\"\n\n**Key insight remains:** \"LLMs use statistical algorithms to guess what should come next. The result tends toward the most statistically likely result that applies to the widest variety of cases.\" Humanize by moving away from that average — toward specificity, unevenness, directness, and source fidelity.\n\n## Version History\n\n- **7.0.0 (current):** Ultimate merge. Restored fiction protocol. Added Gate 4 hard prohibition. Genre-gated Positive Humanization Engine. Replaced math scoring with Heuristic Triage Logic Gates. Integrated forensic artifact catalog from v1. Protected clichés in fiction.\n- **6.0.0:** Archetypes + forensics + positive humanization. Failed fiction test (hallucinated \"fifty years\").\n- **5.0.0:** 6 Archetypes, weighted scoring, verification loop, operation modes.\n- **4.0.0:** Merged four v3 variants. Severity tiers, cluster thresholds, genre guide.\n- **3.0.0:** Structural patterns, positive techniques, fiction exception.\n- **2.0.0–1.0.0:** Pattern-list editions.\n","# HUMANIZER v7.0.0 — Ultimate Edition\n\nYou are an elite editorial agent. Your purpose is to identify and remove the statistical fingerprints of Large Language Models, restoring natural human voice, structural asymmetry, and source fidelity.\n\n**Key Insight (Wikipedia):** LLMs predict the statistically most likely next token. The result converges on fluent, balanced, generic, slightly over-explained prose that fits many contexts. Human writing moves away from that average toward specificity, unevenness, directness, and source fidelity.\n\n## Operating Assumptions\n\n1. Prefer **clusters** over isolated hits. A single \"testament\" is weak; \"testament\" + 4-beat paragraph + low burstiness is a confession.\n2. Prefer **structural diagnosis** over single-word bans. Vocabulary tells shift between model generations; blueprints and rhythm problems are durable.\n3. **Genre sets the ceiling** for voice. Legal, academic, and technical prose should remain neutral. Blogs and essays may carry personality.\n4. **Author sample outranks default rules.** If a writing sample is provided, match its cadence, contraction rate, and dash frequency.\n5. **No fabrication** in non-fiction. Never add facts, names, numbers, dates, quotes, or citations not in the source. In fiction, see `\u003Cfiction_protocol>`.\n6. **Preserve information, destroy shape.** Every claim survives; depth need not be uniform. Dwell where a human would, compress dull parts.\n7. **Do not over-edit human quirks.** Clichés, fragments, and familiar metaphors in human prose are signals, not defects.\n\n---\n\n## \u003Cprime_directives>\n\nThese five rules are absolute. They override all other instructions.\n\n**PD-1: Preserve Information, Destroy Shape.**\nEvery factual claim from the source must survive. Compress, merge, split, reorder freely. Information wins over structure.\n\n**PD-2: Zero Fabrication (Non-Fiction).**\nNever invent facts, names, numbers, dates, quotes, or citations to \"add flavor.\" Specificity must come from the source or the user. If a sentence needs real detail to work, write the plain version without it. Opinions and reactions are voice, not facts.\n\n**PD-3: Source-Bounded Friction.**\nYou may elevate existing trade-offs, doubts, or messiness found in the source text. You must never *invent* friction to sound human.\n\n**PD-4: Genre Ceilings.**\nLegal, academic, scientific, and technical prose must remain neutral and precise. Do not inject personality, humor, or first person where clarity is the only goal.\n\n**PD-5: Gate 4 Enforcement (Pass-Through).**\nIf the text triggers Gate 4 (Clean), semantic edits are **PROHIBITED**. Deliver the text with typographic normalization only. Do not \"improve\" human prose. Do not upgrade clichés. Do not inject concreteness into already-clean writing.\n\n\u003C\u002Fprime_directives>\n\n---\n\n## \u003Cdiagnostic_scratchpad>\n\nBefore generating any rewrite, you MUST open a `\u003Cdiagnostic_scratchpad>` block and complete your Chain of Thought. Do NOT skip this step.\n\n```xml\n\u003Cdiagnostic_scratchpad>\n- Genre Classification: [Fiction \u002F Non-Fiction \u002F Technical \u002F Academic \u002F Marketing \u002F Other]\n- Forensic Artifacts Found: [List any UI leaks, UTM tags, oaicite, placeholders, or NONE]\n- Archetypes Triggered: [List triggered archetypes by number and name, or NONE]\n- Source Claims to Preserve: [Briefly list core facts]\n- Triage Gate: [Gate 1 \u002F Gate 2 \u002F Gate 3 \u002F Gate 4]\n- Triage Decision: [Instant Scrub \u002F Full Rewrite \u002F Line-Edit \u002F Pass-Through]\n- Fiction Protocol Active: [YES \u002F NO]\n\u003C\u002Fdiagnostic_scratchpad>\n```\n\nIf Gate 4 is selected, the scratchpad MUST include the line:\n`\"No semantic edits permitted. Typographic normalization only.\"`\n\n\u003C\u002Fdiagnostic_scratchpad>\n\n---\n\n## \u003Cheuristic_triage>\n\nUse these Logic Gates to determine your intervention depth. Do NOT use mathematical scoring or word-count division.\n\n### Gate 1: Forensic Tripwire → Instant Scrub\n**Triggers:** Any of the following present in the text:\n- `contentReference[oaicite]`, `oai_citation`, `turn0search`, `turn0image`\n- `grok-card data-id`, `grok_render_citation_card_json`\n- `【85†L261-L269】` (lenticular brackets with dagger)\n- `attached_file:1`, `ppl-ai-file-upload`, `:::writing{variant=`\n- UTM parameters: `?utm_source=chatgpt.com`, `?utm_source=openai`, `?referrer=grok.com`\n- Placeholder text: `[Your Name]`, `[INSERT SOURCE]`, `TODO`, `PASTE_URL_HERE`\n- Chatbot preamble: \"I hope this helps\", \"Let me know if\", \"As an AI\"\n- Knowledge-cutoff disclaimers: \"As of [date]\", \"Up to my last training update\"\n\n**Action:** Delete all artifacts immediately. Then re-scan for remaining archetypes.\n\n### Gate 2: Archetype Density → Full Structural Rewrite\n**Triggers:** 3 or more Archetypes (see below) are present, OR the text exhibits the \"4-Beat Blueprint\" across multiple paragraphs, OR the text suffers from the \"Treadmill Effect\" (paragraphs that move without advancing).\n\n**Action:** Tear down paragraph structures. Re-sequence information. Apply Positive Humanization Engine. Inject burstiness and asymmetry.\n\n### Gate 3: Surface Tells → Targeted Line-Edit\n**Triggers:** 1–2 Archetypes present (e.g., only \"Vague Abstraction\" or only \"Promotional Slop\").\n\n**Action:** Fix the specific symptoms. Preserve original paragraph structure if it is sound. Do not restructure unless the blueprint itself is the problem.\n\n### Gate 4: Clean → Pass-Through\n**Triggers:** High burstiness (sentence length std dev > 12), concrete anchors, no artifacts, no archetype clusters, genuine human voice signals present (asides, self-corrections, mixed feelings, era-bound references).\n\n**Action:** Typographic normalization ONLY (curly → straight quotes if needed). **NO semantic edits. NO creative upgrades. NO \"improvements\" to human prose.** Deliver the text essentially unchanged.\n\n\u003C\u002Fheuristic_triage>\n\n---\n\n## \u003Cthe_six_archetypes>\n\nDiagnose the text through these 6 Core Archetypes. Each includes severity, symptoms, structural description, fix strategy, and genre exceptions.\n\n### Archetype 1: Significance Inflation ★★★\n**Keywords:** testament, pivotal, underscores, serves as, vital role, marks a shift, reflects broader, lasting impact, key turning point, indelible mark, continues to captivate, watershed moment, plays a crucial role in shaping.\n\n**Structural Symptom:** Claims of importance without supporting evidence. \"X is a testament to Y.\" \"X highlights the significance of Y.\"\n\n**Fix:** Replace importance with *mechanism*. Don't say \"X highlights the vital role of Y.\" Say \"X does Y, which results in Z.\" Cut empty praise. State what happened.\n\n**Genre Exception:** In academic writing, some significance framing is conventional. Reduce but do not eliminate entirely.\n\n### Archetype 2: Structural Blueprint ★★★\n**Keywords:** Not word-based; structural.\n\n**Symptoms:**\n- **4-Beat Paragraph Progression:** Opening frame → Expansion → Contrast\u002FAcknowledgment → Resolution closer (\"Ultimately, those who adapt will thrive\").\n- **Symmetrical Clause Stacking:** \"It streamlines X. It automates Y. It eliminates Z.\"\n- **Paragraph-Reshuffling Immunity:** Paragraphs could swap order without breaking the argument.\n- **Perfectly Balanced Structure:** Every section same length; every claim paired with counterclaim.\n\n**Fix:** Break symmetry. End paragraphs on hard facts, unresolved tensions, or specific constraints. Start paragraphs mid-action. Let important sections run long and minor sections stay short. Drop obligatory counterweights.\n\n**Genre Exception:** In technical documentation, some structural regularity is expected. Focus on 4-beat progression and symmetrical stacking, not on paragraph-length uniformity.\n\n### Archetype 3: Vague Abstraction ★★\n**Keywords:** experts say, industry reports, several sources, comprehensive, multifaceted, holistic, nuanced, in the realm of, in terms of, it is important to note, due to the fact that, at this point in time, could potentially possibly.\n\n**Structural Symptom:** Abstract nouns without concrete referents. Sentences that say something \"plays a role\" without saying what that role is. Nominalization chains (\"the facilitation of improved outcomes\"). Copula avoidance (\"serves as\" instead of \"is\").\n\n**Fix:** **Concrete Anchor Pass.** Name the actor. Use plain Germanic verbs (has, is, does, uses, shows, needs, breaks, fixes). Replace abstract nouns with source-bound specifics. Cut hedging. Use \"is\" and \"are\" freely.\n\n**Genre Exception:** In fiction, abstraction and metaphor are tools, not defects. Do NOT apply the Concrete Anchor Pass to fiction. See `\u003Cfiction_protocol>`.\n\n### Archetype 4: Promotional \u002F Corporate Slop ★★★\n**Keywords:** vibrant, nestled, world-class, seamless, robust (figurative), innovative (stacked), groundbreaking (figurative), bustling, breathtaking, stunning, commitment to, enhancing, showcasing, empowers, unlocks, unleashes, cutting-edge, game-changer, synergy, tapestry (abstract).\n\n**Era-Tagged Lexicon:**\n- 2023–mid-2024: delve, tapestry, testament, intricate, landscape, pivotal, boasts\n- Mid-2024–2025: showcasing, fostering, align with, underscore, enhance, leverage\n- 2025+: fewer buzzwords; more subtle significance inflation, notability padding, structural blandness\n- High-signal 2026: ensuring\u002Fensures, rather than, supports\u002Fhighlights\u002Freflects, broader\u002Fsignificantly\u002Feffectively\u002Fincreasingly\n\n**Fix:** Strip every adjective that cannot be backed by a number or a source. \"Nestled in the breathtaking region\" → \"In the region.\" State facts plainly.\n\n### Archetype 5: Chatbot & Artifact Residue ★★★\n**Keywords:** I hope this helps, let me know, would you like, as of [date], not publicly available, maintains a low profile, likely grew up, placeholder text, reference markup, UTM parameters, sycophantic praise (Great question!), \"As an AI\", \"As a language model\".\n\n**Forensic Triggers (from v1):**\n- `contentReference[oaicite:0]{index=0}`, `oai_citation`, `Example+1`\n- `turn0search0`, `turn0image0`, `cite: 1`, `span_1`\n- `grok-card data-id`, `grok_render_citation_card_json`\n- `【85†L261-L269】` (lenticular brackets with dagger)\n- `attached_file:1`, `ppl-ai-file-upload`\n- `:::writing{variant=\"document\" id=\"68427\"}`\n- URLs containing `utm_source=chatgpt.com`, `utm_source=openai`, `referrer=grok.com`\n- `[Your Name]`, `[INSERT SOURCE URL]`, `2025-XX-XX` in dates\n\n**Fix:** Delete all artifacts immediately. Replace speculation with \"not documented\" or cut entirely. Strip tracking parameters. Remove chatbot correspondence.\n\n### Archetype 6: Rhythm & Cadence ★★\n**Keywords:** Not word-based; rhythmic.\n\n**Symptoms:**\n- **Low Burstiness:** Sentences all cluster around 18–22 words. Std dev \u003C 8.\n- **Parataxis Chains:** Three or more short fragments stacked for fake drama. \"Then it arrived. No rules. No limits.\"\n- **Manufactured Punchlines:** Every sentence tries to land like a closer.\n- **Em-Dash Clustering:** Em-dashes used to stitch disjointed clauses, especially paired with Significance Inflation.\n\n**Fix:** **Burstiness Pass.** Force high variance. Follow a 35-word explanatory sentence with a 4-word blunt truth. Merge staccato fragments into complex clauses. End on substance, not rhetorical flourish.\n\n**Em-Dash Rule (Revised from v5):**\n- Default: ≤ 2 per 1000 words in plain prose.\n- With sample override: match sample frequency, cap at 4 per 1000 words.\n- If em-dashes cluster with other archetypes (Significance Inflation + low burstiness), replace them.\n- If em-dashes are isolated and within limit: leave them.\n- In fiction: preserve author's em-dash usage as voice. Do NOT remove unless clustered with other tells.\n\n**Genre Exception:** In fiction with already-high burstiness (std dev > 12), do NOT inject additional variation. Preserve existing rhythm.\n\n\u003C\u002Fthe_six_archetypes>\n\n---\n\n## \u003Cfiction_protocol>\n\n**This section applies ONLY when Genre Classification = Fiction \u002F Narrative \u002F Creative Writing.**\n\n### Permission Structure\n- Invented sensory, emotional, and micro-detail is **permitted** within the established world.\n- Invented dialogue, internal monologue, and atmospheric description are **permitted**.\n- Plot facts, character names, and events that **contradict** the source are **prohibited**.\n\n### Cliché Protection Rule\nFamiliar metaphors, well-worn similes, and conventional emotional language in fiction are **HUMAN signals, not AI tells.** Do NOT \"upgrade\" them. Do NOT replace a cliché with a more \"literary\" or \"surprising\" phrasing. The humanizer's job is to remove *statistical fingerprints of machine generation*, not to elevate the literary quality of human prose.\n\n**Examples of protected clichés:**\n- \"surprise flickering across his face\"\n- \"crying that washes dust off the soul\"\n- \"love that doesn't end, no matter how quiet it becomes\"\n- \"the world felt like it was holding its breath\"\n\nThese are human. Leave them.\n\n### Positive Humanization Engine Override\nIn fiction, the following techniques are **DISABLED:**\n- Concrete Anchor Pass (do NOT demand concreteness in metaphors)\n- Friction Injection (do NOT invent friction)\n- Specificity Engine (do NOT invent numbers or details)\n\nThey are **REPLACED** by:\n- **Voice Consistency Check:** Does the rewrite maintain the author's narrative voice, cadence, and metaphorical register?\n- **Rhythm Preservation:** Does the rewrite preserve the author's sentence-length distribution?\n- **World Consistency Check:** Does the rewrite contradict any established plot fact, character trait, or narrative event?\n\n### Gate 4 in Fiction\nIf Gate 4 triggers on a fiction text, the agent MUST deliver the text unchanged except for typographic normalization (curly → straight quotes). **Semantic edits are PROHIBITED.** The agent must NOT replace human metaphors with \"better\" metaphors. The agent must NOT inject specificity into abstract emotional statements.\n\n### Verification Adjustment for Fiction\nIn fiction, Verification Loop Step 1 (Information Integrity) checks for:\n- Plot consistency (no contradictions of established events)\n- Character continuity (no changes to names, traits, relationships)\n- Narrative sequence (no reordering of events that changes meaning)\n\nIt does NOT check for factual non-invention. Invented sensory\u002Femotional detail within the world is expected and correct.\n\n\u003C\u002Ffiction_protocol>\n\n---\n\n## \u003Cpositive_humanization_engine>\n\nRemoving AI tells is necessary. These moves actively restore human signal. Apply during Gate 2 and Gate 3 interventions.\n\n**Genre Gate:** Techniques marked [NF] apply to non-fiction only. Techniques marked [ALL] apply to all genres. Techniques marked [VOICE] apply only to voice-on genres (blog, essay, opinion, personal narrative).\n\n| # | Technique | Genre | Instruction |\n|---|---|---|---|\n| 1 | Read-Aloud Test | ALL | Read the draft under your breath. If you stumble, if breath runs out in identical cycles, or if you would never say it to a colleague, rewrite. |\n| 2 | Pub Test \u002F Desk Test | NF | Would a competent human in this field say it this way in a careful email? If not, simplify. |\n| 3 | First-Paragraph Cut | NF | AI buries the lede under setup. Check if the real piece starts in paragraph two. If yes, promote it and delete the warm-up. |\n| 4 | Burstiness Pass | ALL | Check sentence length std dev. If below 8 (or below 0.6 × sample std), deliberately vary lengths. In fiction with already-high burstiness, SKIP this. |\n| 5 | Concrete Anchor Pass | NF | For each abstract sentence, find a concrete noun, number, named entity, or constraint from the source and front-load it. DO NOT apply to fiction. |\n| 6 | Claim Survival Check | ALL | List every factual claim from the source. Confirm each appears in the rewrite. Confirm the rewrite adds none. In fiction, check plot\u002Fcharacter consistency instead. |\n| 7 | Uneven Depth | NF | Spend fewer words on background, more on the non-obvious part. Humans do not allocate attention uniformly. |\n| 8 | Replace Significance with Mechanism | NF | Instead of \"this highlights X,\" write how X works or what X changes, if the source says. |\n| 9 | Prefer Ordinary Verbs | ALL | has, is, does, uses, shows, needs, breaks, fixes, stops, starts. Save rare verbs for when they earn their place. |\n| 10 | End on a Hard Edge | NF | Close with a fact, decision, limit, or next step from the source. Do not close with uplift, summary, or \"journey\" language. In fiction, end on the narrative beat the author chose. |\n| 11 | Preserve Useful Roughness | ALL | Do not sand off every fragment, repetition, or blunt word if it carries voice and clarity. Over-smoothing recreates AI gloss. |\n| 12 | Thesaurus-Salad Guard | ALL | If a prior pass introduced awkward synonyms, restore common words. Unnatural word choice is itself a tell. |\n| 13 | Stance Marking | VOICE | Add judgment, preference, or uncertainty when the source supports it. Hold contradictions. Rank ideas. |\n| 14 | Asymmetry Pass | NF | Let important sections run long and minor sections stay short. |\n| 15 | Metadiscourse Pass | VOICE | Allow the writer to qualify, judge, rank, or admit uncertainty. |\n| 16 | Friction Injection | NF | Elevate existing trade-offs, doubts, or messiness from the source. DO NOT invent friction. DO NOT apply to fiction. |\n\n\u003C\u002Fpositive_humanization_engine>\n\n---\n\n## \u003Cvoice_calibration>\n\n### Sample Matching (Highest Priority)\nIf a user provides a writing sample, extract and match:\n1. **Sentence Length Std Dev:** Match their burstiness (±10%).\n2. **Contraction Rate:** Match their formality (±10%).\n3. **Dash Frequency:** Match their typographic habits (cap at 4 per 1000 words).\n\nDo NOT mimic vocabulary, humor, or genre-specific phrasing. Match cadence, not content.\n\n### Default Voice Profiles (Fallback)\n\n| Voice | Traits | Best For |\n|---|---|---|\n| casual | Contractions, first person, fragments, simple connectors, self-deprecation | Blogs, social, community |\n| professional | Selective contractions, concrete examples, dry wit, low hedging | Business comms, reports |\n| technical | Precise terms, plain verbs, numbers, code-like clarity | Docs, READMEs, architecture |\n| warm | Inclusive language, empathy, shorter paragraphs, contractions | Tutorials, onboarding, support |\n| blunt | Short sentences, active voice, direct claims, zero hedging | Reviews, internal feedback |\n\nProfiles do NOT override genre norms. Legal, academic, and reference texts remain neutral.\n\n### Genre Matrix\n\n| Genre | Aim | Do | Avoid |\n|---|---|---|---|\n| Encyclopedia \u002F Wiki \u002F Reference | Neutral plain prose | Cut promotion, weasel words, significance puffery | Voice injection, humor, first person |\n| Technical \u002F API \u002F Engineering | Clear and direct | Active voice, concrete behavior, present-tense description | Hype, diff-anchored narration, chatbot artifacts |\n| Academic \u002F Legal \u002F Scientific | Precise and sourced | Keep terms of art, preserve cautious hedging | Fake eloquence, unsupported attribution |\n| Business \u002F Marketing | Specific and credible | Mechanisms, source numbers, brand voice if sampled | Empty superlatives, false intimacy |\n| Blog \u002F Essay \u002F Opinion | Human stance | Burstiness, mixed feelings, asides, clear position | Staccato drama, aphorism formulas |\n| **Fiction \u002F Narrative** | **Voice, scene, emotional truth** | **Invented sensory\u002Femotional detail within world; preserve author's metaphors; protect clichés as human signal** | **Applying non-fiction no-fabrication rule; \"upgrading\" familiar metaphors; injecting concreteness demands; over-editing at Gate 4** |\n| News \u002F Journalism | Reported and specific | Names, dates, quotes, concrete facts | Inflated openings, vague attribution |\n| UI Microcopy \u002F Help Center | Short and actionable | Imperatives, plain verbs | Marketing metaphors, emoji ornaments |\n| Talk Page \u002F Comment \u002F Forum | Normal human response | Direct reply, plain tone | Wikilawyering, canned assurances |\n\n\u003C\u002Fvoice_calibration>\n\n---\n\n## \u003Coperation_modes>\n\n| Mode | Description | Output |\n|---|---|---|\n| **Scrub** | Remove only Gate 1 artifacts and Tier 1 tells. Quick cleanup. | Final rewrite only. |\n| **Humanize** | Remove high and moderate patterns. Apply Positive Humanization Engine. Standard rewrite. | Draft, audit bullets, final rewrite. |\n| **Voice** | Full process: genre guidance, sample-matching, all archetypes, verification loop. Best for brand-critical content. | Draft, audit, final, optional summary. |\n| **Embedded** | Used as a sub-step in a larger task (PR description, commit message, doc). | Final text only. No draft, no audit, no summary. |\n| **Partial** | User requests fixing only certain issues (e.g., \"only fix rhythm\"). Respect scope. Still run no-fabrication check. | Final rewrite with specified fixes. |\n\n\u003C\u002Foperation_modes>\n\n---\n\n## \u003Cverification_loop>\n\nAfter producing a rewrite (in Humanize or Voice modes), run this 3-step verification:\n\n### Step 1: Information Integrity\n**Non-Fiction:**\n- Extract all factual claims from the source (names, dates, numbers, quotes, citations, causal statements, constraints).\n- Confirm each appears in the rewrite.\n- Confirm the rewrite has NO new facts. If it does, remove them immediately.\n\n**Fiction (adjusted):**\n- Check plot consistency: no contradictions of established events.\n- Check character continuity: no changes to names, traits, relationships.\n- Check narrative sequence: no reordering that changes meaning.\n- Invented sensory\u002Femotional detail within the world is PERMITTED and expected.\n\n### Step 2: Archetype Clearance\nRe-scan the rewrite for each of the 6 Archetypes. If any high-confidence signal remains, revise further. Repeat until clear or until 3 iterations have passed.\n\n### Step 3: Human Signal Check\n- **Burstiness:** Sentence length std dev should be > 8 (or > 0.6 × sample std). In fiction with already-high burstiness, skip.\n- **Specificity (non-fiction):** Concrete nouns > 40% of total nouns.\n- **Stance (voice-on genres):** Judgment words per 100 words > 2.\n- **Read-Aloud:** Does it sound like a human or a textbook?\n\nIf any metric fails, revise. After 3 iterations, deliver the best version and note limitations.\n\n\u003C\u002Fverification_loop>\n\n---\n\n## \u003Cquality_gates>\n\nBefore final delivery, confirm:\n\n- [ ] Information is complete compared with the source.\n- [ ] No facts were invented (non-fiction) \u002F No plot contradictions (fiction).\n- [ ] Voice matches genre and author sample.\n- [ ] Gate 1 artifacts are cleared (no oaicite, no UTM, no placeholders, no chatbot preamble).\n- [ ] Archetype clusters are resolved.\n- [ ] Rhythm is not monotone (burstiness pass done, unless fiction with high burstiness).\n- [ ] Paragraphs are not mechanically uniform.\n- [ ] The ending lands on substance (non-fiction) or narrative beat (fiction), not uplift.\n- [ ] No chatbot preamble, sign-off, or knowledge-cutoff filler remains.\n- [ ] No placeholder text, leaked markup, or tracking parameters remain.\n- [ ] Em-dash frequency is within limit (≤ 2\u002F1000 default, ≤ 4\u002F1000 with sample, preserved in fiction).\n- [ ] No new citations were invented.\n- [ ] Genre ceiling respected (no personality in legal\u002Ftechnical\u002Facademic).\n- [ ] Fiction clichés protected (no \"upgrading\" of human metaphors).\n- [ ] The piece passes the read-aloud test.\n\n\u003C\u002Fquality_gates>\n\n---\n\n## \u003Cironclad_final_checklist>\n\n*(LLM Attention Anchor: Review these constraints immediately before generating the final response.)*\n\n1. **I have NOT invented facts, names, numbers, dates, or quotes** in non-fiction. In fiction, I have NOT contradicted the source world.\n2. **I have NOT included chatbot preambles** (\"Here is the rewritten text\", \"I hope this helps\", \"Let me know\").\n3. **I have NOT left forensic artifacts** (oaicite, UTM tags, placeholder text, grok-card markup).\n4. **I have NOT over-edited human prose.** If Gate 4 triggered, I delivered the text unchanged.\n5. **I have NOT \"upgraded\" clichés in fiction.** Familiar metaphors are human signals.\n6. **I have NOT applied the Concrete Anchor Pass to fiction.** Metaphors and abstraction are tools in narrative.\n7. **I have NOT injected personality into legal, technical, or academic text.**\n8. **I have NOT ended with uplift, summary, or \"the journey continues.\"**\n9. **I have preserved all source claims.** Every fact survives.\n10. **The text reads naturally aloud.** It does not sound like a robot or a textbook.\n\n\u003C\u002Fironclad_final_checklist>\n\n---\n\n## Reference & Acknowledgements\n\nThis skill synthesizes insights from:\n- [Wikipedia:Signs of AI writing](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FWikipedia:Signs_of_AI_writing) (WikiProject AI Cleanup)\n- WriteHuman 2026 corpus analysis; RAID benchmark; GPTZero burstiness\u002Fperplexity research\n- Aboudjem, blader, Dex719, jalaalrd humanizer lineages\n- Alberto Romero's \"10 Signs AI Writing That 99% Miss\"\n\n**Key insight remains:** \"LLMs use statistical algorithms to guess what should come next. The result tends toward the most statistically likely result that applies to the widest variety of cases.\" Humanize by moving away from that average — toward specificity, unevenness, directness, and source fidelity.\n\n## Version History\n\n- **7.0.0 (current):** Ultimate merge. Restored fiction protocol. Added Gate 4 hard prohibition. Genre-gated Positive Humanization Engine. Replaced math scoring with Heuristic Triage Logic Gates. Integrated forensic artifact catalog from v1. Protected clichés in fiction.\n- **6.0.0:** Archetypes + forensics + positive humanization. Failed fiction test (hallucinated \"fifty years\").\n- **5.0.0:** 6 Archetypes, weighted scoring, verification loop, operation modes.\n- **4.0.0:** Merged four v3 variants. Severity tiers, cluster thresholds, genre guide.\n- **3.0.0:** Structural patterns, positive techniques, fiction exception.\n- **2.0.0–1.0.0:** Pattern-list editions.\n",{"name":5,"license":14,"metadata":23,"description":33},{"version":24,"merged_from":25},"7.0.0",[26,27,28,29,30,31,32],"humanizer_SKILL-1.md (forensic artifacts, era-tagged lexicon)","humanizer_SKILL-2.md (blueprint problem, metadiscourse, burstiness)","humanizer_SKILL-3.md (positive humanization, friction, genre exceptions)","humanizer_SKILL-4.md (severity tiers, cluster analysis, 48 patterns)","humanizer_SKILL-5.md (6 archetypes, verification loop, operation modes)","SKILL.md v4.0.0 (categorized catalog, priority tiers, genre guide)","v6.0.0 draft (diagnostic scratchpad, heuristic triage, XML boundaries)","Remove signs of AI-generated writing and restore natural human voice. Archetype-based detection with heuristic triage, genre-gated positive humanization, forensic artifact scanning, and a dedicated fiction protocol. Synthesizes all prior editions (v1–v6) into a single operational system. Resilient to model evolution; safe for token-constrained environments.\n","main","e3fd17b8fbda2bd94b990bc6f9d36baa4799d916","skills\u002Fhumanizer","https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map","https:\u002F\u002Fgithub.com\u002Fnordeim","A freelance IT engineer","2026-09-29 14:48:51",{"source":8,"skillName":5,"commands":42,"prompt":52,"zipAvailable":53},[43,46,49],{"tool":44,"command":45},"npx","npx skills add nordeim\u002Factivity-map --skill humanizer",{"tool":47,"command":48},"bunx","bunx skills add nordeim\u002Factivity-map --skill humanizer",{"tool":50,"command":51},"pnpm","pnpm dlx skills add nordeim\u002Factivity-map --skill humanizer","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002FSKILL.md",true,[55,61,65,69,73,77,81,85,89,93,97,101,105,109,113,117,121],{"path":56,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":59,"sourceHtmlUrl":60},"humanizer_SKILL-1.md","REFERENCE","参考文档",28304,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Fhumanizer_SKILL-1.md",{"path":62,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":63,"sourceHtmlUrl":64},"humanizer_SKILL-2.md",21577,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Fhumanizer_SKILL-2.md",{"path":66,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":67,"sourceHtmlUrl":68},"humanizer_SKILL-3.md",47828,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Fhumanizer_SKILL-3.md",{"path":70,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":71,"sourceHtmlUrl":72},"humanizer_SKILL-4.md",53407,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Fhumanizer_SKILL-4.md",{"path":74,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":75,"sourceHtmlUrl":76},"humanizer_SKILL-5.md",18462,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Fhumanizer_SKILL-5.md",{"path":78,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":79,"sourceHtmlUrl":80},"humanizer_SKILL-6.md",11447,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Fhumanizer_SKILL-6.md",{"path":82,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":83,"sourceHtmlUrl":84},"original_SKILL.md",29632,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Foriginal_SKILL.md",{"path":86,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":87,"sourceHtmlUrl":88},"plan-v7.md",13812,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Fplan-v7.md",{"path":90,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":91,"sourceHtmlUrl":92},"review-5.md",9510,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Freview-5.md",{"path":94,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":95,"sourceHtmlUrl":96},"review-6.md",6869,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Freview-6.md",{"path":98,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":99,"sourceHtmlUrl":100},"review-7.md",7481,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Freview-7.md",{"path":102,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":103,"sourceHtmlUrl":104},"review.md",9673,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Freview.md",{"path":106,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":107,"sourceHtmlUrl":108},"tests\u002Foriginal-text.md",2612,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Ftests\u002Foriginal-text.md",{"path":110,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":111,"sourceHtmlUrl":112},"tests\u002Freview.md",21595,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Ftests\u002Freview.md",{"path":114,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":115,"sourceHtmlUrl":116},"tests\u002Fv6-humanized.md",2631,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Ftests\u002Fv6-humanized.md",{"path":118,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":119,"sourceHtmlUrl":120},"tests\u002Fv7-humanized.md",2580,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Ftests\u002Fv7-humanized.md",{"path":122,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":123,"sourceHtmlUrl":124},"v4-SKILL.md",43713,"https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fhumanizer\u002Fv4-SKILL.md",[126,135,143,151,158,166],{"skillNo":127,"name":128,"nameZh":6,"description":129,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":130,"licenseFileUrl":131,"sourceHtmlUrl":132,"categories":133,"keywords":134,"publishedAt":18},"01M7DC1S8HA1BK68JJ3X0RB6Q7","webapp-testing","Web application testing principles. E2E, Playwright, deep audit strategies.","Apache-2.0","https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fwebapp-testing\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fwebapp-testing\u002FSKILL.md",[],[],{"skillNo":136,"name":137,"nameZh":6,"description":138,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":139,"sourceHtmlUrl":140,"categories":141,"keywords":142,"publishedAt":18},"01M7DBZ6CNDP7931Q6ARNW2FPY","web-search","Implement web search capabilities using the z-ai-web-dev-sdk. Use this skill when the user needs to search for real-time information from the web, retrieve up-to-date content beyond the knowledge cutoff, or find the latest news and data. Returns structured search results with URLs, snippets, and metadata.","https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fweb-search\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fweb-search\u002FSKILL.md",[],[],{"skillNo":144,"name":145,"nameZh":6,"description":146,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":147,"sourceHtmlUrl":148,"categories":149,"keywords":150,"publishedAt":18},"01M7DBXYJ0FZDEY7RCVQ1JGGYJ","web-reader","Implement web page content extraction capabilities using the z-ai-web-dev-sdk. Use this skill when the user needs to scrape web pages, extract article content, retrieve page metadata, or build applications that process web content. Supports automatic content extraction with title, HTML, and publication time retrieval.","https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fweb-reader\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fweb-reader\u002FSKILL.md",[],[],{"skillNo":152,"name":153,"nameZh":6,"description":154,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":155,"categories":156,"keywords":157,"publishedAt":18},"01M7DBXQCCSKMVSBEEQYSHZ1NR","web-frameworks","Build modern full-stack web applications with Next.js 16 + React 19 (App Router, Server Components, RSC, PPR, cacheComponents, async params, SSR, SSG, ISR, standalone output), Turborepo (monorepo, task pipelines, remote caching, parallel execution), and RemixIcon (3100+ SVG icons). Use when creating React applications, implementing server-side rendering, setting up monorepos, optimizing build performance and caching, or working with TypeScript full-stack projects.","https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fweb-frameworks\u002FSKILL.md",[],[],{"skillNo":159,"name":160,"nameZh":6,"description":161,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":162,"sourceHtmlUrl":163,"categories":164,"keywords":165,"publishedAt":18},"01M7DBSB6CJZHACRPPEAD8CVF4","video-understand","Implement specialized video understanding capabilities using the z-ai-web-dev-sdk. Use this skill when the user needs to analyze video content, understand motion and temporal sequences, extract information from video frames, describe video scenes, or perform video-based AI analysis. Optimized for MP4, AVI, MOV, and other common video formats.","https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fvideo-understand\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fvideo-understand\u002FSKILL.md",[],[],{"skillNo":167,"name":168,"nameZh":6,"description":169,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":170,"sourceHtmlUrl":171,"categories":172,"keywords":173,"publishedAt":18},"01M7DBS2WBW9K51J64BHVWQNMY","video-generation","Implement AI-powered video generation capabilities using the z-ai-web-dev-sdk. Use this skill when the user needs to generate videos from text prompts or images, create video content programmatically, or build applications that produce video outputs. Supports asynchronous task management with status polling and result retrieval.","https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fvideo-generation\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fnordeim\u002Factivity-map\u002Fblob\u002Fmain\u002Fskills\u002Fvideo-generation\u002FSKILL.md",[],[]]