By nagisanzenin
Generate social media posts that read as distinct human voices, with built-in AI-detection auditing and rewriting to avoid stylometric tells. Create and manage voice profiles, audit text for AI fingerprints, and mass-produce platform-native content with consistent persona.
Blind semantic judge of whether text reads AI-generated — the layer regex can't see. MUST BE USED for /idiolect:audit and the campaign blind pass. Deliberately receives ONLY the text (plus platform); no authorship context, no scanner results, no hints. Returns strict JSON.
Builds a complete new voice profile from a measured fingerprint (corpus mode) or a gap seed (invention mode) for /idiolect:synthesize and /idiolect:self. Owns the validate loop; returns only when the profile passes spec and roster distance.
Voice-isolated post writer for /idiolect:campaign fan-out. Receives ONE voice profile, one brief, one platform, one structure assignment — and nothing about any other voice or post, which is what makes N posts genuinely different. Self-verifies with the scanner before returning.
Two-layer AI-tell audit of any text — deterministic scanner (lexicon, constructions, stylometry with line numbers) plus a blind semantic judge. Use when the user wants to know whether text reads AI-generated, what specifically gives it away, and how to fix it.
One brief, N posts, N genuinely different people — voice-isolated writers, cross-post overlap detection, per-post verification receipts. Use for launching/promoting something across multiple posts or platforms without the posts sharing a fingerprint.
Rewrite existing text so it reads human — either scrubbed-and-souled in its own register, or fully re-voiced as one of the roster's people. Use when the user has a draft that "sounds like AI" or wants existing copy converted into a specific voice.
Front door and router for the idiolect voice engine. Use when the user mentions idiolect, asks what voices are available or who should write something, wants writing that "doesn't sound like AI", says text "sounds like ChatGPT/AI slop", or asks for human-sounding posts and no more specific idiolect skill has already matched. Also the first-run onboarding.
Build and maintain the user's OWN voice profile from continuously captured writing plus explicit samples — status, build/refresh, add samples, capture on/off. Use when the user wants posts in their own voice, wants to see what's been learned about their style, or wants to manage capture.
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Humanizers remove the AI. idiolect installs a human.
62 fully-specified voice profiles — real psychologies, systematic flaws, platform habitats, few-shot exemplars — plus a deterministic AI-tell scanner, a blind semantic auditor, and a closed verify-revise loop. Write, humanize, audit, and mass-produce short-form posts that read as distinct, specific people, with receipts instead of vibes. Runs on Claude Code, OpenAI Codex, and opencode out of the box.
idiolect (n.) — the variety of a language unique to one individual. The thing this plugin manufactures, measures, and defends.
The same request — "announce a kiln drop: the dark-blue glaze finally came out right, a couple pieces cracked, shop update Thursday" — scored by the same deterministic scanner (idiolect scan; 0 = clean, higher = more AI tells).
Without idiolect — default assistant output · scan 45 / 100 — slop
✨ Big news, friends! ✨ After weeks of anticipation, our stunning dark blue glaze collection is finally here — and we couldn't be more thrilled to share it with you! 🌊
Each piece is lovingly handcrafted and fired to perfection, embodying the beautiful imperfection of the ceramic journey. Yes, a few pieces didn't quite survive the kiln — but that only makes the survivors more special. 💙
Whether you're a seasoned collector or just beginning your pottery journey, these one-of-a-kind treasures are sure to elevate your space and bring a touch of handmade magic into your everyday moments.
Don't miss out — the shop update drops this Thursday! 🛍️ Link in bio.
Scanner flags: em-dash density, "elevate your," "journey" ×2, "whether you're a…", a six-hashtag pile, emoji density. Human texture: none — no specifics, no time anchor, no first-person action.
With idiolect — voice Freja, ceramicist, Copenhagen · scan 0 / 100 — clean · conform PASS, in-range
33 went in. 27 came out whole.
I opened 114 at 7, before anyone else was in.
The dark blue behaved. People have waited for it since firing 112 and I could not promise it, the kiln decides and not me. This morning the color came right on every blue piece, even the two that cracked. I wrote a full line in the log for it. Most firings get only numbers.
The kiln took its share too. 2 to the bin, a mug and the tall bottle, cracked at the foot, photo 4. You cannot both blow and have flour in the mouth, as we say at home.
4 are seconds. A little glaze pooling on the inside, nothing that leaks. They go 40 percent off with the flaw in the photo, not hidden.
Shop update Thursday 19:00 CET. Link in bio. 8 mugs, 5 bowls, 3 bottles in the dark blue. Mugs 260 DKK, bowls 340, bottles 520. Shipping inside EU is 89, the carrier's number.
Stoneware, fired to 1240 degrees. I look forward to see them go Thursday.
Written by the engine, then verified: the new competence layer kept it in clay (no borrowed jargon), soul is on (the blue glaze earns the one long line in the log; the loss is stated flat), and the L1-Danish idiom is her systematic error profile, not decoration. The scanner credits exact quantities, a time anchor, and first-person actions as human texture. Nothing here "sounds human" — it scans clean and conforms to a declared fingerprint.
Five voices, one brief, all conform-checked →
The popular humanizer skills are subtractive: a pattern list that removes "delve," em dashes, and rule-of-three. Useful — and structurally insufficient. Everyone running the same subtraction converges on a new detectable house style (uniformly short sentences, aggressive casualness, zero em dashes), because removal can't produce identity. Meanwhile RLHF models homogenize by design — co-writing with them measurably flattens stylistic diversity (Padmakumar & He 2024), hardest on non-Western writers (CHI 2025).
idiolect's claim, from the research up (see docs/THEORY.md, 40+ sources, corrections included):
Human-reading text = Scrubbed × Voiced × Crafted × Varied — multiplicative, a zero anywhere zeroes the product. So the plugin ships all four layers: a tell scanner (scrub), a voice roster (voice), a craft floor distilled from a century of copy and rhetoric research (craft), and structure-rotation + overlap + distance machinery (variance).
Same brief — "promote a $15/mo booking tool" — through five voices (full posts + receipts):
npx claudepluginhub nagisanzenin/idiolectEvidence-based learning engine: first-principles curricula, generation-first Socratic tutoring, free-recall verification with receipts, FSRS-scheduled memory, and interactive explorable artifacts. Learn anything; keep it.
Enhances Claude Code from producing raw code into delivering production-ready systems. 14 specialized agents handle architecture, tested code, security audit, CI/CD, and documentation. Use for building apps/websites/services, adding features, hardening, deployment, testing, review, or architecture design.
Calibrated per-subagent reasoning effort: classify each subtask, dispatch it to the cheapest tier that still passes.
Voice profile system for consistent, authentic writing. Apply, create, blend, and analyze voices. Includes 4 built-in profiles: technical-authority, friendly-explainer, executive-brief, and casual-conversational.
Apply Sam Dumont's personal writing voice and style when drafting any written content. Works in English and French.
Write in a voice you choose, with less AI tone. One skill, 5 commands (/cadence write, learn, recast, deslop, voices) and a deterministic de-slop detector that scores prose and names every AI tell.
Research-backed writing toolkit: anti-AI prose principles, iterative review loop, author voice capture, and voice-matched content generation
Audit & rewrite content to remove AI writing patterns ("AI-isms"). Supports detect-only and edit-in-place modes, voice profiles, and iterate-to-convergence.
Turns findings into publishable writing in your own voice: drafts from your raw material, runs every draft through a dual review gate that catches AI tells and voice drift, and sharpens its rules from your hand edits. Part of the Fieldwork investigative suite.