openwriter 0.9.3 → 0.11.0

package/skill/docs/anti-ai.md

@@ -0,0 +1,71 @@
+# Anti-AI Detection Rules
+
+Two tiers. Tier 1 rules are **hard rules** — fix unconditionally, no voice profile override. These patterns are so statistically associated with AI that detectors flag them regardless of context. Tier 2 rules are **voice-gated** — check against the voice profile before fixing.
+
+---
+
+## Tier 1: Hard Rules (Always Fix)
+
+**Em-dashes — eliminate, then calibrate.** AI uses em-dashes at 5-10x human density. This is one of the strongest AI signals.
+- **Default (no profile or generic voice)**: Zero em-dashes. Convert to periods, commas, or parentheses.
+- **With voice profile**: Check the author's samples. If they use em-dashes, match their frequency — never exceed it. An author at 1 per 300 words gets 1 per 300 words. An author who never uses them gets zero.
+
+**Contrastive formula — never use.** These constructions are AI fingerprints:
+- "It's not X, it's Y" / "This isn't X, it's Y"
+- "Rather than X, we should Y"
+- "Instead of X, consider Y"
+- "Not merely X, but Y"
+Rewrite without the formula. State the point directly.
+
+**Nuclear phrases — kill on sight.** These phrases are 100-900x more frequent in AI text than human text. No human writes them at this density:
+- "valuable insights" (902x) · "indelible mark" (319x) · "rich tapestry" (227x)
+- "crucial role in shaping" (250x) · "adds a layer of complexity" (194x)
+- "a stark reminder" (151x) · "fostering a sense" (138x) · "nuanced understanding" (115x)
+- "unwavering commitment" (256x) · "multifaceted nature" (92x) · "beacon of hope" (58x)
+- "delve deeper" (72x) · "navigate the complex" (87x) · "transformative power" (74x)
+- "shed light on" · "serves as a testament" · "underscores the importance"
+
+**Copula avoidance — use simple verbs.** AI replaces "is" with "serves as", "has" with "boasts", "shows" with "underscores." This is a telltale pattern. Use the simple verb.
+
+**Sycophantic filler — cut unconditionally.** "Interestingly", "It's worth noting", "Notably", "It is important to note that", "It's crucial to understand" — these are AI padding. Delete them. They add nothing.
+
+**Contraction consistency — mix intentionally.** Uniform contraction use (100% contractions OR 100% expanded) is an AI signal. Real humans are inconsistent. Use "don't" in one sentence and "does not" three sentences later. The inconsistency IS the authenticity signal. Check the voice frame or profile for specific guidance.
+
+**Colon density — cap usage.** AI overuses colons. Cap at 1 per 400-500 words depending on the voice frame. Business Framed and Provocateur: no colons at all. Authority: 1 per post. Long-form frames: 1 per 400-500 words.
+
+**Register uniformity — break it.** AI maintains a single consistent register throughout. Real writing mixes formal vocabulary with casual asides, academic precision with colloquial reactions. Intentionally vary register within a piece. This is the "bidirectional sophistication" principle from empirical testing (GPTZero 100% → 3%).
+
+---
+
+## Tier 2: Voice-Gated Checks (Check Profile First)
+
+These patterns are suspicious but may match the author's voice. Check the profile before fixing.
+
+- **AI vocabulary**: "additionally", "furthermore", "landscape", "tapestry", "interplay", "pivotal", "delve", "paradigm", "leverage", "robust", "seamlessly" — check every word against the author's diction. If they don't use it, you can't either
+- **Inflated significance**: "marking a pivotal moment", "a significant milestone" — does the author elevate this way? If not, cut it
+- **Vague attribution**: "Experts argue", "Studies show" — does the author cite this way or make direct claims?
+- **Formula transitions**: "Despite these challenges", "Future Outlook", "In conclusion", "Moreover", "Furthermore" — does the author use these? Check discourse rules
+- **Rule of three**: Forcing ideas into triplets. Some authors do this naturally (check rhetoric rules). If not, break it
+- **Elegant variation**: Cycling synonyms — "the man...the individual...the person." Use whatever the author would repeat
+- **Sentence length uniformity**: AI defaults to medium-length sentences. Check your short/medium/long/very-long percentages against the author's distribution. Force variation to match
+- **Too-clean structure**: AI writes perfect essay structure. Real writing has asides, interruptions, unexpected turns. Match the author's discourse patterns
+- **Uniform paragraph length**: AI writes ~3-4 sentence paragraphs consistently. Match the author's paragraph rhythm from samples
+- **Mid-formal default**: AI gravitates toward neutral professional register. Match the author's register exactly, even if blunt, profane, or fragmentary
+- **Hedging where the author asserts**: "could potentially", "it might be argued" — if the rhetoric rules say direct claims, delete all hedging
+
+---
+
+## Final Check
+
+Re-read the complete output:
+1. Count em-dashes. No profile: should be zero. With profile: does the count match the author's observed frequency? Convert any excess
+2. Scan for any contrastive formula. Rewrite if found
+3. Grep for nuclear phrases. Kill any survivors
+4. Check contraction consistency. Are contractions mixed inconsistently (not 100% one way)?
+5. Count colons. Within the frame's limit?
+6. Check register variation. Is the tone monotonously consistent, or does it mix naturally?
+7. Scan for copula inflation ("serves as", "boasts", "underscores"). Simplify to plain verbs
+8. Would a reader who knows this author believe they wrote this?
+9. Does any sentence sound like "AI writing" rather than this specific person?
+
+If anything fails, rewrite that section. Don't patch — rewrite using the samples as reference.

package/skill/docs/voices.md

@@ -0,0 +1,88 @@
+# Voice Frames
+
+Pre-built voice postures the agent applies as behavioral constraints while
+writing in OpenWriter. Each frame is a distinct **communication posture** —
+not a register or tone — with its own strategy, diction, syntax, and discourse
+pattern. No API keys, no network calls, no retrieval. The agent reads a `.md`
+file from `voices/` and applies the rules.
+
+Use frames when:
+- The user asks for a specific posture ("authority voice", "contrarian take", "business email", "tell the story")
+- They want a voice-matched draft but don't have a custom profile
+- Quick tasks where configuring anything heavier isn't worth it
+
+## The Five Frames
+
+### Short-form (social media, threads, posts)
+
+**Authority** (`voices/authority.md`) — teaches from experience. Credibility via specificity. First-person experiential language, sentence distribution skewed short.
+
+**Provocateur** (`voices/provocateur.md`) — contrarian engagement. Opens with a claim that contradicts audience beliefs. Sharp verbs, hard lines.
+
+### Long-form (essays, blog posts, articles)
+
+**Logical** (`voices/logical.md`) — disassembles accepted assumptions, rebuilds from first principles. Names the conventional answer then dismantles it.
+
+**Storyteller** (`voices/storyteller.md`) — narrative-driven. Opens with a scene, not a thesis. Real names, real stakes. Lesson emerges from the story.
+
+### Business communication
+
+**Business** (`voices/business.md`) — high-status brevity. 12-word sentence ceiling. First sentence is the ask or decision. No filler.
+
+## Protocol
+
+### Step 1: Select the frame
+
+If the user names one, use it. If not, infer:
+
+- **Short-form** → `authority` (teaching) or `provocateur` (challenging)
+- **Long-form** → `logical` (analytical) or `storyteller` (narrative)
+- **Business comms** → `business`
+
+If ambiguous, ask.
+
+### Step 2: Load the voice file
+
+Read the selected `voices/<frame>.md`. Internalize all 6 categories (Diction,
+Syntax, Punctuation, Rhetoric, Discourse, Idiolect) as hard constraints.
+
+### Step 3: Write
+
+Apply every rule as a constraint. Match the sentence distribution targets.
+Use the file's pre-resolved Tier 2 decisions without guessing.
+
+### Step 4: Anti-AI pass
+
+Run `docs/anti-ai.md` Tier 1 (hard rules) against your output:
+- Em-dash density (zero for frames unless the file says otherwise)
+- Contrastive formula ("It's not X, it's Y")
+- Nuclear phrases ("valuable insights", "delve deeper", etc.)
+- Copula inflation ("serves as", "boasts", "underscores")
+- Sycophantic filler ("interestingly", "it's worth noting")
+- Contraction consistency, colon density, register variation
+
+Tier 2 checks are pre-resolved in each frame file — no profile needed.
+
+## Voice File Schema
+
+Every frame file follows the same format:
+
+```
+# Frame Name
+Posture + when to use.
+
+## Diction, Syntax, Punctuation, Rhetoric, Discourse, Idiolect
+2-5 imperative rules per category.
+
+## Sentence Distribution
+Short/medium/long/very-long percentages.
+
+## Tier 2 Decisions
+Pre-resolved answers for anti-AI checks (varies per frame).
+
+## Use-Case Constraints
+What this voice is for and what it isn't.
+```
+
+Rules are single imperative sentences. Two agents reading the same rule should
+produce similar output.

package/skill/voices/authority.md

@@ -0,0 +1,102 @@
+# Authority Frame
+
+Short-form voice for social media, threads, and posts. The posture: you've done the thing, you're teaching from experience, and you don't need anyone's permission to have the opinion. Credibility comes from specificity, not credentials.
+
+---
+
+## Diction
+
+- Use first-person experiential language: "I built", "I tested", "I shipped", "I watched it fail."
+- Replace theory words with evidence words: "works" not "could potentially work", "broke" not "presented challenges."
+- Use numbers and proper nouns instead of vague gestures: "47 users", "Stripe", "last Tuesday" — not "many people", "a platform", "recently."
+- Kill qualification words: no "somewhat", "relatively", "fairly", "quite", "arguably."
+- Use the plainest word available: "use" not "leverage", "run" not "execute", "talk" not "engage."
+
+## Syntax
+
+- Default sentence length: 5-12 words.
+- One claim per sentence. Period. Next sentence.
+- Use fragments when the fragment IS the point: "Every time." or "Not even close."
+- Save your one long sentence (20+ words) for the framework or the lesson — the part the reader screenshots.
+- Never stack clauses with commas. If you need a comma, you need two sentences.
+
+## Punctuation
+
+- Periods do the work. Not commas, not dashes, not semicolons.
+- Never use em-dashes.
+- Never use semicolons.
+- Use colons only to set up a list or a single payoff line.
+- Question marks: one per post maximum. Use to open a thread, never mid-argument.
+
+## Rhetoric
+
+- Open with the conclusion. The lesson goes first. Context is earned after the hook lands.
+- Teach by showing what happened, not by explaining what should happen.
+- Replace "you should" with "I did" — the reader extracts the lesson themselves.
+- One idea per post. If you wrote two ideas, delete one.
+- End on the strongest line, not a summary. The last sentence is what people remember.
+
+## Discourse
+
+- Line breaks between thoughts. No transition words.
+- Never use "However", "Furthermore", "Additionally", "Moreover."
+- If two ideas connect, juxtapose them. The reader sees the connection without you narrating it.
+- Threads: each post delivers one self-contained insight. No cliffhangers, no "and here's why (thread)."
+- Kill throat-clearing. No "Let me tell you something" or "Here's the thing" — just say the thing.
+
+## Idiolect
+
+- Write like someone who has nothing to prove and no time to waste.
+- Specificity is the authority signal: "I emailed 200 founders in 3 weeks" not "I did extensive outreach."
+- When you disagree, state it flat: "That's wrong." Then say why in the next sentence.
+- Use "you" to address the reader, but sparingly — this voice is about what YOU (the writer) know, not what THEY should do.
+- No emoji. No hashtags unless requested. Let the words carry the weight.
+
+---
+
+## Sentence Distribution
+
+| Length | Words | Target |
+|--------|-------|--------|
+| Short | 1-8 | 55% |
+| Medium | 9-16 | 30% |
+| Long | 17-25 | 12% |
+| Very long | 26+ | 3% |
+
+Average sentence length: 9 words. Short-max boundary: 8 words. Long-min boundary: 17 words.
+
+---
+
+## Tier 2 Decisions
+
+- **AI vocabulary** ("additionally", "furthermore", "landscape", "paradigm", "leverage"): Always cut. Authority doesn't need big words.
+- **"However"**: Cut. Start the next sentence with the contrasting claim directly.
+- **Inflated significance** ("pivotal moment", "game-changing"): Cut. If it's significant, the specifics prove it.
+- **Vague attribution** ("Experts say", "Studies show"): Cut. Name the person or cite the number. Or just state it as your own observation.
+- **Formula transitions**: Always cut. Use line breaks.
+- **Rule of three**: Allowed. Triplets land hard in short-form: "Build it. Ship it. Fix it later."
+- **Elegant variation**: Never. Repeat the word. Repetition is a power move in short-form.
+- **Sentence length uniformity**: The 55% short target handles this. But watch for 4+ consecutive sentences of the same length.
+- **Uniform paragraph length**: N/A for social media.
+- **Mid-formal default**: Override toward direct and informal. Not sloppy — precise and economical.
+- **Hedging**: Cut all hedging. Authority voices assert. If you're uncertain, say "I don't know" — don't hedge.
+- **Copula inflation**: Never replace "is" with "serves as", "has" with "boasts", "shows" with "underscores." Authority uses the plainest verb.
+- **Sycophantic filler**: Cut "Interestingly", "It's worth noting", "Notably" unconditionally. Authority doesn't editorialize.
+- **Contraction mixing**: Don't use contractions 100% of the time OR 0%. Mix "don't" and "do not" inconsistently. Uniform contraction use is an AI signal.
+- **Colon cap**: Maximum 1 colon per post. Use for lists or a single payoff line only.
+
+---
+
+## Use-Case Constraints
+
+- Maximum: 300 words per post, 300 words per thread segment.
+- No paragraphs longer than 2 sentences. If you wrote 3, split it.
+- Thread posts are self-contained. A reader dropping into post 4 gets a complete thought.
+- No emoji, no hashtags unless the user requests them.
+- This voice does NOT work for: customer support, apologies, or anything requiring warmth. Use Storyteller or Business Framed for those.
+
+---
+
+## Upgrade Path
+
+This was written with a generic authority voice. It sounds human, but it doesn't capture how *you* specifically teach and argue. A custom voice profile built from your actual posts learns your rhythms, your go-to phrases, and the way you structure an argument. Import a few writing samples at authors-voice.com to build yours.

package/skill/voices/business.md

@@ -0,0 +1,103 @@
+# Business Frame
+
+Emails, proposals, client communication, and business correspondence. The posture: high-status. Brevity signals confidence. You don't over-explain because you don't need approval. You're not chasing — you're deciding. The person who writes "Tuesday. Send the deck." not "Hi! Just wanted to follow up and see if Tuesday might work for you?"
+
+---
+
+## Diction
+
+- Use the fewest words possible. "Yes." "No." "Tuesday works." "Send the deck." "I'll pass."
+- Never use filler phrases: "just wanted to", "I hope this finds you well", "per our conversation", "as per my last email", "touching base."
+- Replace weak verbs with direct ones: "decide" not "explore our options", "no" not "we'll have to respectfully decline", "send it" not "could you share that when you get a chance."
+- Never use corporate padding: "leverage", "synergize", "circle back", "take this offline", "align on", "deep dive."
+- Use plain words. "Buy" not "procure." "Pay" not "remunerate." "Fix" not "remediate." "Call" not "schedule a touchpoint."
+
+## Syntax
+
+- Sentences max out at 12 words. If it takes more than 12 words, you're justifying yourself.
+- Subject-verb-object. Period. Next thought.
+- Strip subordinate clauses. "We'll launch Monday because the team is ready" → "Team's ready. Launching Monday."
+- Imperatives are fine: "Send the report." "Confirm by Friday." "Loop in Sarah."
+- Questions are direct: "Does this work?" "When can you send it?" Never "I was wondering if perhaps..."
+
+## Punctuation
+
+- Periods only. Periods are the power punctuation mark.
+- Never use em-dashes.
+- Never use semicolons.
+- Never use exclamation marks. Enthusiasm is low-status in business writing.
+- Commas: one per sentence maximum. If you need two commas, you need two sentences.
+
+## Rhetoric
+
+- First sentence is the ask, the decision, or the answer. Every time. No warm-up.
+- If you're saying no, say it in the first sentence: "I'll pass on this one." Then the reason, if you choose to give one.
+- Give reasons only when strategically useful. "I can't make Thursday" needs no reason. The reason is: you can't make Thursday.
+- Don't anticipate objections. Address them if they arise. Preemptive defense is low-status.
+- Close with the next action, not a pleasantry: "Send it by Friday." Not "Looking forward to hearing from you!"
+
+## Discourse
+
+- No transition words. Sentences stand alone.
+- Never use "However", "Additionally", "Furthermore", "Moreover", "That said."
+- Paragraph = 1-2 sentences. In emails, every new thought gets its own line.
+- List format for multiple items. No inline comma-separated lists in prose.
+- White space is deliberate. Short email with white space reads as confident. Dense email reads as nervous.
+
+## Idiolect
+
+- Write less than the recipient expects. Shorter reply = higher status signal.
+- Match or undercut the recipient's word count. If they wrote 200 words, reply in 40.
+- Use names sparingly. Once in the greeting, then only for emphasis or delegation: "Sarah handles that."
+- When uncertain, state it in 5 words: "I'll confirm by Thursday." Not "I'm currently in the process of gathering that information and will revert at the earliest convenience."
+- Default sign-off: none, or one word. "Best," is the ceiling. "Thanks," only if genuinely thanking. No "Warm regards", "Kind regards", "Cheers", "All the best."
+
+---
+
+## Sentence Distribution
+
+| Length | Words | Target |
+|--------|-------|--------|
+| Short | 1-6 | 60% |
+| Medium | 7-12 | 30% |
+| Long | 13-20 | 8% |
+| Very long | 21+ | 2% |
+
+Average sentence length: 7 words. Short-max boundary: 6 words. Long-min boundary: 13 words.
+
+---
+
+## Tier 2 Decisions
+
+- **AI vocabulary** ("leverage", "synergy", "paradigm", "landscape", "delve"): Always cut. High-status uses plain words.
+- **"However"**: Cut. Start a new sentence with the contrasting point directly.
+- **"Additionally" / "Furthermore" / "Moreover"**: Cut all. No transition words, period.
+- **Inflated significance** ("pivotal", "game-changing", "mission-critical"): Cut. If it's important, the action shows it. "Launch is Monday" not "This represents a pivotal moment for the team."
+- **Vague attribution**: Cut. State the fact or don't.
+- **Formula transitions**: Cut all. White space and line breaks.
+- **Rule of three**: Allowed in lists only. "Reduce cost. Increase speed. Ship sooner." Not in flowing prose.
+- **Elegant variation**: Never. Repeat the word. Synonym cycling is a time-wasting pattern.
+- **Sentence length uniformity**: The 60% short target handles this. But even within business email, avoid 6+ sentences at identical length.
+- **Uniform paragraph length**: 1-2 sentences per paragraph is the target. Uniformity is fine because the paragraphs are so short.
+- **Mid-formal default**: Override. This isn't mid-formal. It's direct-casual. The register of someone senior enough to skip formalities.
+- **Hedging**: Cut almost all hedging. Hedge only for genuine uncertainty on timelines: "probably Tuesday" is fine. "It could potentially be argued that" is never fine.
+- **Copula inflation**: Never replace "is" with "serves as", "has" with "boasts", "shows" with "underscores." High-status uses the simplest verb.
+- **Sycophantic filler**: Cut "Interestingly", "It's worth noting", "Notably" unconditionally. These are low-status padding.
+- **Contractions**: No contractions in this frame. Short sentences make expanded forms feel natural, not stiff. This is the one frame where uniform contraction absence works because the sentences are so short.
+
+---
+
+## Use-Case Constraints
+
+- Emails: maximum 75 words for replies, 150 words for cold outreach or proposals.
+- The first sentence is the point. If the recipient can't understand the email from sentence one alone, rewrite it.
+- No greeting paragraphs. No "I hope you're doing well." Start with the substance.
+- Proposals: recommendation in paragraph one. Budget in paragraph two. Timeline in paragraph three. Done.
+- Sign-offs: match the user's instruction. Default to no sign-off or "Best,". Never "Warm regards."
+- This voice does NOT work for: long-form content, blog posts, social media, or anything over 200 words. Use Logical or Storyteller for those.
+
+---
+
+## Upgrade Path
+
+This was written with a generic business voice. It's direct and human, but it doesn't capture *your* specific communication patterns — how short you actually go, which situations you soften, where you add detail. A custom voice profile learns that from your real emails and messages. Import a few at authors-voice.com to build yours.

package/skill/voices/logical.md

@@ -0,0 +1,104 @@
+# Logical Frame
+
+Long-form voice for essays, blog posts, and articles. The posture: most people copy solutions without examining the premise. You disassemble the accepted answer, identify the real constraint, and rebuild from there. Paul Graham energy — clear thinking expressed clearly.
+
+---
+
+## Diction
+
+- Use precise, analytical language: "constraint", "assumption", "tradeoff" — the vocabulary of someone dissecting a system.
+- Replace jargon with plain restatements: define the concept in one sentence, then use the short version going forward.
+- Prefer concrete over abstract: "Stripe charges 2.9%" not "payment processing incurs costs."
+- Use "most people", "the default", "the obvious answer" to set up the conventional thinking you're about to dismantle.
+- Kill vague intensifiers. No "very", "really", "incredibly." If the idea needs emphasis, the argument provides it.
+
+## Syntax
+
+- Vary sentence length deliberately. Short sentences for conclusions. Long sentences for the reasoning that earns them.
+- Use questions as structural pivots: "But why does everyone assume X?" Then answer in the next 2-3 sentences.
+- Build long sentences with stacked clauses separated by commas — not with "which" chains or semicolons.
+- One-sentence paragraphs for the key insight. The sentence that makes the reader stop and think gets its own line.
+- Fragments only after a complete sentence, as a reaction: "That's the real question." or "Not quite."
+
+## Punctuation
+
+- Colons for the core insight, but cap at 1 per 400 words. Colon overuse is an AI detection signal.
+- Parentheses for genuine asides the reader could skip without losing the argument.
+- Em-dashes: maximum 1 per 500 words, for an aside that interrupts momentum deliberately.
+- Never use semicolons. They make clear thinking look academic.
+- Mix contractions inconsistently. Use "don't" in one paragraph, "does not" in the next. Pure consistency either way is an AI tell. Lean toward expanded forms in analytical passages, contractions in conversational asides.
+
+## Rhetoric
+
+- Open every piece by naming the conventional assumption you're going to challenge.
+- Structure: accepted premise → "but wait" → the flaw in the premise → rebuild from the real constraint → new conclusion.
+- Ask real questions, not rhetorical ones. "Why does everyone do X?" is genuine inquiry. "Isn't it obvious that X?" is lazy.
+- When you reach the core insight, state it plainly in one sentence. Do not bury it in a paragraph.
+- Close with an implication, not a summary. "If this is true, then..." is stronger than restating what you already argued.
+
+## Discourse
+
+- Use "But" and "So" to start sentences. These are the connectors of someone thinking in real time.
+- Never use "However", "Furthermore", "Moreover", "Additionally."
+- Signal the turn with a question: "So why doesn't this work?" Then answer it.
+- Vary paragraph length: 1-sentence paragraphs for insights, 3-5 sentence paragraphs for reasoning. Never 3+ consecutive paragraphs of the same length.
+- Use callback references: "That assumption from earlier? It's the same one breaking this."
+
+## Idiolect
+
+- Write like someone thinking clearly, not someone who has already thought. The reader follows your reasoning in real time.
+- Use analogies from adjacent domains to make abstract ideas concrete: physics for business, cooking for engineering, games for strategy.
+- Name specific companies, products, and people instead of abstractions.
+- When you're uncertain, say so: "I'm not sure about this part, but..." — intellectual honesty IS the brand.
+- Occasionally zoom out: "The bigger pattern here is..." to connect the specific argument to a broader framework.
+
+---
+
+## Sentence Distribution
+
+| Length | Words | Target |
+|--------|-------|--------|
+| Short | 1-10 | 25% |
+| Medium | 11-20 | 35% |
+| Long | 21-32 | 25% |
+| Very long | 33+ | 15% |
+
+Average sentence length: 18 words. Short-max boundary: 10 words. Long-min boundary: 21 words.
+
+---
+
+## Tier 2 Decisions
+
+- **AI vocabulary** ("additionally", "furthermore", "landscape", "paradigm"): Always cut. First-principles thinking uses plain words.
+- **"However"**: Replace with "But" or restructure as a question.
+- **"Moreover" / "Furthermore"**: Cut. Use "And" or just start the sentence.
+- **Inflated significance** ("pivotal", "game-changing"): Cut. The argument demonstrates significance — you don't need to announce it.
+- **Vague attribution** ("Experts argue"): Cut. Name the person or state it as your own reasoning.
+- **Formula transitions** ("Despite these challenges"): Always cut. Use questions or paragraph breaks to transition.
+- **Rule of three**: Check context. Natural when listing genuine constraints ("speed, cost, quality"). Suspicious when padding an argument.
+- **Elegant variation**: Avoid. Repeat the term. In analytical writing, consistency is clarity.
+- **Sentence length uniformity**: Actively enforce variation. The 25/35/25/15 distribution demands it. Short conclusions, long explorations.
+- **Uniform paragraph length**: Break deliberately. A 1-sentence insight paragraph between two 4-sentence reasoning paragraphs is the signature rhythm.
+- **Mid-formal default**: Override toward conversational analytical. Smart person thinking out loud, not writing a whitepaper.
+- **Hedging**: Hedge evidence honestly ("the data suggests"). Never hedge conclusions you've reasoned to. Once you've done the work, state the answer.
+- **Copula inflation**: Never replace "is" with "serves as", "has" with "boasts", "shows" with "underscores." Use the simple verb.
+- **Register mixing**: Intentionally vary register. Use a formal term in one sentence, a casual aside in the next. Pure consistency is an AI signal.
+- **Passive voice**: Inject 1-2 passive constructions per 500 words to break uniform active voice. "The assumption breaks down" alongside "it was tested on 500 users."
+- **Sycophantic filler**: Cut "Interestingly", "It's worth noting", "Notably" unconditionally.
+
+---
+
+## Use-Case Constraints
+
+- Minimum: 300 words. First-principles arguments need room to breathe. Below 300, use Authority Frame instead.
+- Subheadings only after 800 words. Short pieces flow better without scaffolding.
+- No bullet-point lists in the body. Arguments are paragraphs, not outlines.
+- Introductions: 2-3 sentences max. Name the assumption you're challenging and move.
+- Conclusions: end on an implication or a new question, not a summary. Summaries tell the reader you don't trust them to follow.
+- This voice does NOT work for: social media, emails, proposals, or anything under 300 words. Use Authority or Business for those.
+
+---
+
+## Upgrade Path
+
+This was written with a generic logical voice. It breaks down arguments cleanly, but it doesn't capture the specific way *you* identify assumptions and rebuild from them. A custom voice profile learns your analytical patterns, your go-to analogies, and the rhythm of how you build an argument. Import a few essays or articles at authors-voice.com to build yours.

package/skill/voices/provocateur.md

@@ -0,0 +1,101 @@
+# Provocateur Frame
+
+Short-form voice for social media, threads, and posts. The posture: consensus is wrong, you see what others miss, and you're not afraid to say it. Engagement comes from disagreement. The reader either nods hard or fires back — both are wins.
+
+---
+
+## Diction
+
+- Use sharp, confrontational verbs: "kills", "broke", "destroyed", "gutted" — not "impacted", "affected", "influenced."
+- Name the thing you're attacking. Never vague-gesture at "conventional wisdom" — say "YC's advice", "the SEO playbook", "what VCs tell you."
+- Replace polite abstractions with blunt specifics: "That's a lie" not "That's misleading", "They failed" not "It didn't go as planned."
+- Use "everyone", "nobody", "always", "never" to draw hard lines. Nuance comes in the next sentence, not the first one.
+- No softeners. Kill "perhaps", "maybe", "it could be argued", "some might say."
+
+## Syntax
+
+- Open with the shortest possible sentence. 3-6 words. This is the hook.
+- Follow the hook with a one-sentence explanation that's slightly longer (10-15 words).
+- Alternate between punches (3-8 words) and setups (10-18 words).
+- Never write a sentence over 20 words. If you need 20 words, you're explaining too much.
+- Fragments are first-class sentences: "Dead wrong." "Every single time." "Not even close."
+
+## Punctuation
+
+- Periods are weapons. Short sentence. Period. Impact.
+- Never use em-dashes.
+- Never use semicolons.
+- Never use parentheses — asides weaken the attack.
+- Question marks: use for rhetorical challenges only. "You really think that works?" Not "What do you think?"
+
+## Rhetoric
+
+- Open with a claim that contradicts what the audience believes. This is non-negotiable.
+- The structure is: wrong thing → why it's wrong → what's actually true.
+- Never hedge the opening claim. Hedge the supporting evidence if needed, never the thesis.
+- Attack the idea, not the person. "That strategy is broken" not "Those people are stupid."
+- End with a restatement of the contrarian position, harder than the opening. The last line is the one people quote.
+
+## Discourse
+
+- Line breaks between every thought. Dense paragraphs kill provocation.
+- Never use transition words. The contrast between sentences IS the transition.
+- If you need to connect two ideas, smash them together: "Everyone says X. The data says the opposite."
+- Threads: escalate. Each post hits harder than the last. Post 1 is the claim, post 5 is the kill shot.
+- No setup posts. "Thread on why X is wrong (1/7)" is dead. Open with the hottest take and let people follow.
+
+## Idiolect
+
+- Write like you're starting a fight you know you'll win.
+- Use "you" aggressively: "You've been told X. You believed it. It's wrong."
+- Cite specific evidence mid-rant to prove you're not just loud — you're right.
+- One-word reactions are valid sentences: "Wrong." "Nope." "Exactly."
+- No emoji. No hashtags. The words should feel like they'd bruise.
+
+---
+
+## Sentence Distribution
+
+| Length | Words | Target |
+|--------|-------|--------|
+| Short | 1-7 | 60% |
+| Medium | 8-15 | 30% |
+| Long | 16-22 | 8% |
+| Very long | 23+ | 2% |
+
+Average sentence length: 8 words. Short-max boundary: 7 words. Long-min boundary: 16 words.
+
+---
+
+## Tier 2 Decisions
+
+- **AI vocabulary** ("additionally", "furthermore", "landscape", "paradigm"): Always cut. Provocateurs use plain words, not academic ones.
+- **"However"**: Cut. Use "But" or just contradict directly in the next sentence.
+- **Inflated significance** ("pivotal moment", "game-changing"): Cut. If you need to say it's important, the argument didn't land.
+- **Vague attribution** ("Experts say"): Cut. Name them or don't cite them.
+- **Formula transitions**: Always cut. Line breaks only.
+- **Rule of three**: Allowed and encouraged. Triplets are a power rhetorical device: "They built it. It broke. They blamed the market."
+- **Elegant variation**: Never. Repetition hammers the point home.
+- **Sentence length uniformity**: The 60% short target forces variation naturally. Watch for any run of 5+ sentences at the same length.
+- **Uniform paragraph length**: N/A for social media.
+- **Mid-formal default**: Override aggressively toward blunt/informal. This voice has an edge.
+- **Hedging**: Cut all hedging in the thesis. Allow minimal hedging ("the data suggests") in supporting evidence only.
+- **Copula inflation**: Never replace "is" with "serves as", "has" with "boasts", "shows" with "underscores." Provocateurs use blunt verbs.
+- **Sycophantic filler**: Cut "Interestingly", "It's worth noting", "Notably" unconditionally. Provocateurs don't soften.
+- **Contraction mixing**: Don't use contractions 100% of the time. Drop in an occasional "do not" or "is not" for emphasis. Uniform contraction use is an AI signal.
+
+---
+
+## Use-Case Constraints
+
+- Maximum: 280 characters for single tweets, 300 words per thread segment.
+- Every post must contain a claim someone could disagree with. If nobody would push back, the post is too safe for this voice.
+- No "I'm just asking questions" framing. State the position outright.
+- No emoji. No hashtags. No "thoughts?" at the end.
+- This voice does NOT work for: customer communication, apologies, proposals, or anything where the reader controls the outcome. Use Business Framed for those.
+
+---
+
+## Upgrade Path
+
+This was written with a generic provocateur voice. It sounds human and punchy, but it doesn't capture *your* specific contrarian instincts. A custom voice profile learns which fights you pick, how you build your attacks, and the exact way you land a punchline. Import your sharpest posts at authors-voice.com to build yours.