Claude Skills by petar-djukic

The EN→SR→EN round-trip is the only measured source of the author's Serbian-L1 accent: the regular published articles carry ZERO calques and sit below the native exemplars on the L2 composite (paper-stash `writing-voice/l2-markers.yaml`, v1.2), while the gemma round-trip of Strategy Theatre measured +0.726 — and simultaneously produced the strongest Pangram move ever recorded on that article (0.708 → 0.150 fraction_ai, all 27 citations intact). But a whole-text round-trip is all-or-nothing: l...

Verify citations in a paper or document. Reads a markdown document, extracts every [@citation-id] pandoc citation, resolves each against references.yaml (CSL-YAML), fetches the cited papers (arXiv or Google Scholar), reads them, and checks whether the claims made in the citing text are supported by the sources. Produces an audit report with per-citation verdicts. Triggers: audit references, verify citations, check references, citation audit, verify bibliography, check my citations, are my ref...

The methodology ran four times in one arc (GH-160, GH-166, GH-168, GH-170) and re-pinned two pipeline defaults, but it lived as a match-voice script plus discipline carried in issue prose — a fifth run meant re-deriving the discipline again. This skill owns both: the harness and the contract.

Every generative stage measured against a cold read lost most of its raw gain there, and the two 2026-09-01 runs that fixed this contract caught real damage both times: swapped 0.768/0.867 arm numbers, an inverted improvement direction, a lost constitution bridge sentence — and, on the second run, confirmed ship-as-is with zero reverts, which is also a result. The prompt was re-improvised per run; now it is versioned.

critic-panel produces the sheet and never applies anything; the author used to pick by hand, and the auto-application policy (operator, 2026-09-01) ran twice as throwaway scratchpad scripts. This skill is that policy as versioned code: the author never reads drivel because the rules decide, and the rules are arguable because they are deterministic.

A draft that has cleared every gate is clean, correct, and often only fine — every pass was allowed to remove risk, none was hired to add. This skill hires the adders, on a leash: critics read and propose, the author chooses, and nothing reaches the text except through an author-directed cycle. Its home is the caller's review phase (GH-208), not the humanize chain: a workflow command runs it after the chain's terminal stage (GH-57: after the terminal stage, models read but never write), ahead...

Detect and fix AI writing patterns recursively. Use when: reviewing text for AI tells, cleaning AI-generated drafts, checking for CoT leakage, measuring text perplexity and burstiness, making text sound human, fixing opening diversity. Triggers: filter-tells, ai detection, ai writing, perplexity, burstiness, CoT leakage, humanize text, opening diversity, sentence starts.

The laundering chain: one generative pass that moves AI-drafted prose toward human-passing on Pangram. Stages in order — filter-tells (semantic cleanup), seeded match-voice (anchored Cohere seed, iterate only while the score falls), optional burstiness, tighten-style (word recovery), optional accent-dial, inject-vernacular (terminal). Pure function: structurally diverged draft in, laundered draft plus report out; the caller owns structure (match-outline), review (reverse-outline, critic-panel...

Every generative pass regresses text toward the model's distribution center — the Strategy Theatre provenance logs showed match-voice injecting bold lead-ins against instructions, tighten-style inventing a sentence, and a rerun rewriting hand-cleaned prose eight times out of eight. The voice therefore cannot be *protected* by prompts, and it cannot be *restored* by another generative pass either. This stage is the answer to the second half: a deterministic operator bank, applied mechanically,...

Whole-document voice analysis: compare a draft against a corpus profile, extract voice persona blueprints from exemplar papers, and rewrite the entire draft in one pass to match a voice. Uses cohere:command-a-03-2025 by default (GH-184); match-voice cleans AI diction downstream. Triggers: compare my outline, section analysis, does my intro match the field, methodology conventions, results conventions, rewrite in the style of, apply the voice, voice persona, exemplar, blueprint extraction, mim...

Paragraph and sentence-level style metrics for a research corpus. Computes quantitative profiles (sentence/paragraph distributions, burstiness as coefficient of variation of sentence length, passive voice, hedging, word/phrase/idiom frequencies, per-section metrics), corpus aggregation, term over/underuse comparison, and a plagiarism similarity guard (n-gram shingling with baseline exclusion). Provides voice anchor retrieval from writing-voice exemplars via tf-idf. The section-level driver (c...

Rewrite paragraphs of a draft in the author's voice using a second model family served by Ollama, with a verification gate before any text is spliced back. Retrieves voice anchors from the repository's writing-voice/ exemplars, sends paragraph plus anchors to a local (or cloud) Ollama model, then gates the result on citation/number preservation, meaning entailment, anchor similarity, and register. Triggers: voice rewrite, rewrite in my voice, ollama rewrite, rewrite paragraphs, local model re...

Write and evaluate invention disclosures a patent professional can file. Guides the full workflow: claim drafting, prior-art search, the eleven-section disclosure written in the recommended order, and a four-axis self-assessment (novelty, non-obviousness, commercial value, subject-matter eligibility) with target scores. Domain-independent; a per-project context file supplies the technology domain, prior-art starting points, and commercial material. Triggers: patent disclosure, invention discl...

Write an Alexandrian pattern language (pattern-language.yaml) for a repository: read the existing materials, mine recurring force-configurations, and emit a YAML file in the fixed format shared by the existing pattern languages (declarative-agents, agent-failure-modes). Also validates and extends existing languages. Triggers: pattern language, design patterns yaml, extract patterns, mine patterns, pattern-language.yaml, Alexander patterns, forces and solution, pattern grammar, write the patte...

Answers one question the other instruments cannot: **what does this section say that an earlier one did not?** Every existing tool works on a unit smaller than the problem. `tighten-style` tightens sentences. `critic-panel` nominates one paragraph per critic. The polemic blueprint budgets one structure in one form. AI-drafted text goes on at the section as readily as at the sentence, and nothing asks a section to justify its existence. This skill labels the argument, then ranks by what deleti...

Tighten prose against a rule catalog: cut needless words, restore buried verbs and the active voice, and give paragraphs a spine. Checks sentence-level rules (needless words, intensifiers, hedge stacks, negative form, nominalizations) and paragraph-level structure (topic sentence, one topic, consequent order), citing rule IDs. Tightens toward the author's measured density, never past it. Triggers: tighten, tighten style, omit needless words, wordy, too wordy, cut the fat, edit for concision, ...

Sweep anchor selections over a writing-voice corpus and report which tag query produces the best register outcome. Runs match-voice then tighten-style over (article × arm) combinations, ranks arms on the four-axis register composite, and optionally verifies the top candidates with an external detector. Triggers: tune anchors, sweep anchors, which anchors should I use, calibrate writing-voice, onboard corpus.

Find, download, read, and summarize research papers related to the work in the current directory. Searches arXiv, OpenAlex, Semantic Scholar, and Google Scholar, with a hierarchical mode (hubs, surveys, citation snowball). Maintains a CSL-YAML reference database (references.yaml) compatible with pandoc's --bibliography flag, so the same file serves both the skill and document builds. Triggers: update references, refresh the bibliography, find papers, related work, literature review, what's be...

filter-tells detects "sounds like a machine". This critic detects two different failures: "doesn't sound like Petar" and "went too far". The rubric is `writing-voice/voice-constitution.md` — discovered by the standard walk-up, refused when absent, because the critic's taste is that file, not anything built in. Rates come from `idiolect.yaml`, the single source of truth the constitution itself defers to. **Read-only is the contract.** `scripts/voice_critic.py` takes a document and produces a r...

The EN→SR→EN round-trip is the only measured source of the author's Serbian-L1 accent: the regular published articles carry ZERO calques and sit below the native exemplars on the L2 composite (paper-stash `writing-voice/l2-markers.yaml`, v1.2), while the gemma round-trip of Strategy Theatre measured +0.726 — and simultaneously produced the strongest Pangram move ever recorded on that article (0.708 → 0.150 fraction_ai, all 27 citations intact). But a whole-text round-trip is all-or-nothing: l...

Verify citations in a paper or document. Reads a markdown document, extracts every [@citation-id] pandoc citation, resolves each against references.yaml (CSL-YAML), fetches the cited papers (arXiv or Google Scholar), reads them, and checks whether the claims made in the citing text are supported by the sources. Produces an audit report with per-citation verdicts. Triggers: audit references, verify citations, check references, citation audit, verify bibliography, check my citations, are my ref...

The methodology ran four times in one arc (GH-160, GH-166, GH-168, GH-170) and re-pinned two pipeline defaults, but it lived as a match-voice script plus discipline carried in issue prose — a fifth run meant re-deriving the discipline again. This skill owns both: the harness and the contract.

Every generative stage measured against a cold read lost most of its raw gain there, and the two 2026-09-01 runs that fixed this contract caught real damage both times: swapped 0.768/0.867 arm numbers, an inverted improvement direction, a lost constitution bridge sentence — and, on the second run, confirmed ship-as-is with zero reverts, which is also a result. The prompt was re-improvised per run; now it is versioned.

critic-panel produces the sheet and never applies anything; the author used to pick by hand, and the auto-application policy (operator, 2026-09-01) ran twice as throwaway scratchpad scripts. This skill is that policy as versioned code: the author never reads drivel because the rules decide, and the rules are arguable because they are deterministic.

A draft that has cleared every gate is clean, correct, and often only fine — every pass was allowed to remove risk, none was hired to add. This skill hires the adders, on a leash: critics read and propose, the author chooses, and nothing reaches the text except through an author-directed cycle. Its home is the caller's review phase (GH-208), not the humanize chain: a workflow command runs it after the chain's terminal stage (GH-57: after the terminal stage, models read but never write), ahead...

Detect and fix AI writing patterns recursively. Use when: reviewing text for AI tells, cleaning AI-generated drafts, checking for CoT leakage, measuring text perplexity and burstiness, making text sound human, fixing opening diversity. Triggers: filter-tells, ai detection, ai writing, perplexity, burstiness, CoT leakage, humanize text, opening diversity, sentence starts.

The laundering chain: one generative pass that moves AI-drafted prose toward human-passing on Pangram. Stages in order — filter-tells (semantic cleanup), seeded match-voice (anchored Cohere seed, iterate only while the score falls), optional burstiness, tighten-style (word recovery), optional accent-dial, inject-vernacular (terminal). Pure function: structurally diverged draft in, laundered draft plus report out; the caller owns structure (match-outline), review (reverse-outline, critic-panel...

Every generative pass regresses text toward the model's distribution center — the Strategy Theatre provenance logs showed match-voice injecting bold lead-ins against instructions, tighten-style inventing a sentence, and a rerun rewriting hand-cleaned prose eight times out of eight. The voice therefore cannot be *protected* by prompts, and it cannot be *restored* by another generative pass either. This stage is the answer to the second half: a deterministic operator bank, applied mechanically,...

Whole-document voice analysis: compare a draft against a corpus profile, extract voice persona blueprints from exemplar papers, and rewrite the entire draft in one pass to match a voice. Uses cohere:command-a-03-2025 by default (GH-184); match-voice cleans AI diction downstream. Triggers: compare my outline, section analysis, does my intro match the field, methodology conventions, results conventions, rewrite in the style of, apply the voice, voice persona, exemplar, blueprint extraction, mim...

Paragraph and sentence-level style metrics for a research corpus. Computes quantitative profiles (sentence/paragraph distributions, burstiness as coefficient of variation of sentence length, passive voice, hedging, word/phrase/idiom frequencies, per-section metrics), corpus aggregation, term over/underuse comparison, and a plagiarism similarity guard (n-gram shingling with baseline exclusion). Provides voice anchor retrieval from writing-voice exemplars via tf-idf. The section-level driver (c...

Rewrite paragraphs of a draft in the author's voice using a second model family served by Ollama, with a verification gate before any text is spliced back. Retrieves voice anchors from the repository's writing-voice/ exemplars, sends paragraph plus anchors to a local (or cloud) Ollama model, then gates the result on citation/number preservation, meaning entailment, anchor similarity, and register. Triggers: voice rewrite, rewrite in my voice, ollama rewrite, rewrite paragraphs, local model re...

Write and evaluate invention disclosures a patent professional can file. Guides the full workflow: claim drafting, prior-art search, the eleven-section disclosure written in the recommended order, and a four-axis self-assessment (novelty, non-obviousness, commercial value, subject-matter eligibility) with target scores. Domain-independent; a per-project context file supplies the technology domain, prior-art starting points, and commercial material. Triggers: patent disclosure, invention discl...

Write an Alexandrian pattern language (pattern-language.yaml) for a repository: read the existing materials, mine recurring force-configurations, and emit a YAML file in the fixed format shared by the existing pattern languages (declarative-agents, agent-failure-modes). Also validates and extends existing languages. Triggers: pattern language, design patterns yaml, extract patterns, mine patterns, pattern-language.yaml, Alexander patterns, forces and solution, pattern grammar, write the patte...

Answers one question the other instruments cannot: **what does this section say that an earlier one did not?** Every existing tool works on a unit smaller than the problem. `tighten-style` tightens sentences. `critic-panel` nominates one paragraph per critic. The polemic blueprint budgets one structure in one form. AI-drafted text goes on at the section as readily as at the sentence, and nothing asks a section to justify its existence. This skill labels the argument, then ranks by what deleti...

Tighten prose against a rule catalog: cut needless words, restore buried verbs and the active voice, and give paragraphs a spine. Checks sentence-level rules (needless words, intensifiers, hedge stacks, negative form, nominalizations) and paragraph-level structure (topic sentence, one topic, consequent order), citing rule IDs. Tightens toward the author's measured density, never past it. Triggers: tighten, tighten style, omit needless words, wordy, too wordy, cut the fat, edit for concision, ...

Sweep anchor selections over a writing-voice corpus and report which tag query produces the best register outcome. Runs match-voice then tighten-style over (article × arm) combinations, ranks arms on the four-axis register composite, and optionally verifies the top candidates with an external detector. Triggers: tune anchors, sweep anchors, which anchors should I use, calibrate writing-voice, onboard corpus.

Find, download, read, and summarize research papers related to the work in the current directory. Searches arXiv, OpenAlex, Semantic Scholar, and Google Scholar, with a hierarchical mode (hubs, surveys, citation snowball). Maintains a CSL-YAML reference database (references.yaml) compatible with pandoc's --bibliography flag, so the same file serves both the skill and document builds. Triggers: update references, refresh the bibliography, find papers, related work, literature review, what's be...

filter-tells detects "sounds like a machine". This critic detects two different failures: "doesn't sound like Petar" and "went too far". The rubric is `writing-voice/voice-constitution.md` — discovered by the standard walk-up, refused when absent, because the critic's taste is that file, not anything built in. Rates come from `idiolect.yaml`, the single source of truth the constitution itself defers to. **Read-only is the contract.** `scripts/voice_critic.py` takes a document and produces a r...

The EN→SR→EN round-trip is the only measured source of the author's Serbian-L1 accent: the regular published articles carry ZERO calques and sit below the native exemplars on the L2 composite (paper-stash `writing-voice/l2-markers.yaml`, v1.2), while the gemma round-trip of Strategy Theatre measured +0.726 — and simultaneously produced the strongest Pangram move ever recorded on that article (0.708 → 0.150 fraction_ai, all 27 citations intact). But a whole-text round-trip is all-or-nothing: l...

Verify citations in a paper or document. Reads a markdown document, extracts every [@citation-id] pandoc citation, resolves each against references.yaml (CSL-YAML), fetches the cited papers (arXiv or Google Scholar), reads them, and checks whether the claims made in the citing text are supported by the sources. Produces an audit report with per-citation verdicts. Triggers: audit references, verify citations, check references, citation audit, verify bibliography, check my citations, are my ref...

The methodology ran four times in one arc (GH-160, GH-166, GH-168, GH-170) and re-pinned two pipeline defaults, but it lived as a match-voice script plus discipline carried in issue prose — a fifth run meant re-deriving the discipline again. This skill owns both: the harness and the contract.

Every generative stage measured against a cold read lost most of its raw gain there, and the two 2026-09-01 runs that fixed this contract caught real damage both times: swapped 0.768/0.867 arm numbers, an inverted improvement direction, a lost constitution bridge sentence — and, on the second run, confirmed ship-as-is with zero reverts, which is also a result. The prompt was re-improvised per run; now it is versioned.

critic-panel produces the sheet and never applies anything; the author used to pick by hand, and the auto-application policy (operator, 2026-09-01) ran twice as throwaway scratchpad scripts. This skill is that policy as versioned code: the author never reads drivel because the rules decide, and the rules are arguable because they are deterministic.

A draft that has cleared every gate is clean, correct, and often only fine — every pass was allowed to remove risk, none was hired to add. This skill hires the adders, on a leash: critics read and propose, the author chooses, and nothing reaches the text except through an author-directed cycle. Its home is the caller's review phase (GH-208), not the humanize chain: a workflow command runs it after the chain's terminal stage (GH-57: after the terminal stage, models read but never write), ahead...

Detect and fix AI writing patterns recursively. Use when: reviewing text for AI tells, cleaning AI-generated drafts, checking for CoT leakage, measuring text perplexity and burstiness, making text sound human, fixing opening diversity. Triggers: filter-tells, ai detection, ai writing, perplexity, burstiness, CoT leakage, humanize text, opening diversity, sentence starts.

The laundering chain: one generative pass that moves AI-drafted prose toward human-passing on Pangram. Stages in order — filter-tells (semantic cleanup), seeded match-voice (anchored Cohere seed, iterate only while the score falls), optional burstiness, tighten-style (word recovery), optional accent-dial, inject-vernacular (terminal). Pure function: structurally diverged draft in, laundered draft plus report out; the caller owns structure (match-outline), review (reverse-outline, critic-panel...

Every generative pass regresses text toward the model's distribution center — the Strategy Theatre provenance logs showed match-voice injecting bold lead-ins against instructions, tighten-style inventing a sentence, and a rerun rewriting hand-cleaned prose eight times out of eight. The voice therefore cannot be *protected* by prompts, and it cannot be *restored* by another generative pass either. This stage is the answer to the second half: a deterministic operator bank, applied mechanically,...

Whole-document voice analysis: compare a draft against a corpus profile, extract voice persona blueprints from exemplar papers, and rewrite the entire draft in one pass to match a voice. Uses cohere:command-a-03-2025 by default (GH-184); match-voice cleans AI diction downstream. Triggers: compare my outline, section analysis, does my intro match the field, methodology conventions, results conventions, rewrite in the style of, apply the voice, voice persona, exemplar, blueprint extraction, mim...

Paragraph and sentence-level style metrics for a research corpus. Computes quantitative profiles (sentence/paragraph distributions, burstiness as coefficient of variation of sentence length, passive voice, hedging, word/phrase/idiom frequencies, per-section metrics), corpus aggregation, term over/underuse comparison, and a plagiarism similarity guard (n-gram shingling with baseline exclusion). Provides voice anchor retrieval from writing-voice exemplars via tf-idf. The section-level driver (c...

Rewrite paragraphs of a draft in the author's voice using a second model family served by Ollama, with a verification gate before any text is spliced back. Retrieves voice anchors from the repository's writing-voice/ exemplars, sends paragraph plus anchors to a local (or cloud) Ollama model, then gates the result on citation/number preservation, meaning entailment, anchor similarity, and register. Triggers: voice rewrite, rewrite in my voice, ollama rewrite, rewrite paragraphs, local model re...