russellian-style

Files (128)

• .gitignore
• SKILL.md
• ai-vocabulary-supplement.json · assets/
• index.json · assets/feynman-corpus/
• index.json · assets/hoskinson-corpus/
• -IBwDvxMnBc.md · assets/hoskinson-corpus/transcripts/
• 0QtQGzqAIiU.md · assets/hoskinson-corpus/transcripts/
• 3_wJOF4Y8gU.md · assets/hoskinson-corpus/transcripts/
• 3bY66Zgr8Cs.md · assets/hoskinson-corpus/transcripts/
• 6n_CxVxOHUQ.md · assets/hoskinson-corpus/transcripts/
• 9d3bp33AJFk.md · assets/hoskinson-corpus/transcripts/
• ALG0OkFTg0A.md · assets/hoskinson-corpus/transcripts/
• EDyQEUZ0PlY.md · assets/hoskinson-corpus/transcripts/
• F0lR_u7BVho.md · assets/hoskinson-corpus/transcripts/
• HvIAgDEUC4o.md · assets/hoskinson-corpus/transcripts/
• ILglnEC0iqU.md · assets/hoskinson-corpus/transcripts/
• JilEb42q-CI.md · assets/hoskinson-corpus/transcripts/
• P2sNyAZLLqg.md · assets/hoskinson-corpus/transcripts/
• VCm8sYmrr_Y.md · assets/hoskinson-corpus/transcripts/
• WWUCgaKO77k.md · assets/hoskinson-corpus/transcripts/
• aO1S6MOrZcE.md · assets/hoskinson-corpus/transcripts/
• eilQyJJQY9c.md · assets/hoskinson-corpus/transcripts/
• fqrAzBAi64c.md · assets/hoskinson-corpus/transcripts/
• g7uySEgt06c.md · assets/hoskinson-corpus/transcripts/
• k8a6tX53YPs.md · assets/hoskinson-corpus/transcripts/
• uZgDxPCXgPo.md · assets/hoskinson-corpus/transcripts/
• ugmOeNlxwqs.md · assets/hoskinson-corpus/transcripts/
• w2bhIQdzeI4.md · assets/hoskinson-corpus/transcripts/
• xiDHg-Ql2hU.md · assets/hoskinson-corpus/transcripts/
• zaTcYGkY7YU.md · assets/hoskinson-corpus/transcripts/
• zhFTO1jYjbk.md · assets/hoskinson-corpus/transcripts/
• index.json · assets/russell-corpus/
• russell-delta-profile.json · assets/
• russellian-rules-v2.json · assets/
• russellian-rules.json · assets/
• style-pass-report.template.md · assets/
• narrative-editorial.md · assets/system-prompts/
• polemic.md · assets/system-prompts/
• technical-exposition.md · assets/system-prompts/
• pyproject.toml
• before-after-examples.md · references/
• how-i-write-maxims.md · references/
• logical-atomism-for-writers.md · references/
• negative-triggers.md · references/
• russell-corpus-map.md · references/
• russellian-style-guide.md · references/
• russellian-vitality-guide.md · references/
• section-headers.md · references/
• triadic-voice-guide.md · references/
• __init__.py · scripts/
• build_delta_profile.py · scripts/
• delta_math.py · scripts/
• lint_ai_staccato.py · scripts/
• lint_ai_vocabulary.py · scripts/
• lint_burstiness.py · scripts/
• lint_common.py · scripts/
• lint_concrete_instance_density.py · scripts/
• lint_epistemic_precision.py · scripts/
• lint_footnotes.py · scripts/
• lint_hedges.py · scripts/
• …and 68 more

SKILL.md (first lines)

# russellian-style

Rewrites technical prose in Bertrand Russell's analytic style and audits markdown for hedging, passive voice, modifier bloat, and rhythm defects. For authors of the book suite and anyone enforcing signal density on accuracy-bearing text.

## What it owns

- Sentence- and paragraph-level prose discipline.
- The Russellian style catalog and rule registry.
- Deterministic linters for hedges, passive voice, signal density, parallel structure, rhythm, and listicle abstraction.
- The `style-pass-report.md` artifact and its acceptance metrics.
- Refusal protocol for non-accuracy genres.

## What it does NOT own

- Source ingestion, claim ledgers, wiki synthesis — owned by `book-knowledge`.
- Chapter drafting, release bundles, PDF/EPUB rendering — owned by `book-compose`.
- Editorial persona reviews — owned by `book-review`.
- Question answering against a drafted manuscript — owned by `book-qa`.
- Generic AI-writing tells (em-dash overuse, rule-of-three, AI vocabulary) — owned by `humanizer`.

## Linters

- `scripts/lint_hedges.py` — hedge vocabulary detection against the rule registry.
- `scripts/lint_passive_voice.py` — passive constructions via spaCy dependency parse.
- `scripts/lint_signal_density.py` — adjective and adverb ratio per sentence against budget.
- `scripts/lint_parallel_structure.py` — grammatical-opening parity across bullet lists.
- `scripts/lint_sentence_rhythm.py` — sentence-length variance and cadence defects.
- `scripts/lint_listicle_abstract.py` — abstract-noun listicles masquerading as content.
- `scripts/lint_ai_staccato.py` — the negation-affirmation wall ("X is not Y. X is Z.") and "This is …" stacking. the highest-signal AI tell; passes every other linter while reading as machine prose. Also flags antithesis-as-default-closer and a repeated antithesis signature phrase (the negation/contrast tic the persona panel kept catching above the sentence).
- `scripts/lint_footnotes.py` — orphaned footnotes: an inline `[^x]` marker with no `[^x]:` definition, or a definition no inline marker references. Deterministic and invisible to the prose linters; a draft can carry every footnote definition and none of the inline citations and still pass every style gate. Wired into `chapter_contract_check` as `footnote_orphan_count` (gated at 0).
- `scripts/lint_burstiness.py`, `scripts/lint_sentence_rhythm.py` — uniform sentence length / repeated openings.
- `scripts/lint_ai_vocabulary.py`, `scripts/lint_concrete_instance_density.py`, `scripts/lint_epistemic_precision.py`, `scripts/lint_paragraph_motion.py` — vitality linters (absence-of-the-positive-move detectors).

## Delta calibration

- `scripts/score_russell_delta.py <file>` — Burrows Delta vs the 50-paragraph Russell profile; verdict "within / at the edge of / outside Russell's range". Advisory.
- `scripts/score_russell_delta.py --diagnose <file>` — the **calibration levers**: the most-frequent words ranked by divergence (z-score) and direction. Over-used emphatic absolutes ("never", "cannot", "no") and bare "it"/"and" drive Delta up; under-used subordinators ("of", "which", "but", "if") drive it down. Calibrate by *prose moves* (more subordination, fewer absolutes, named subjects), never by hunting single words — cutting one under-used word can raise the Delta. See the vitality guide.

## Acceptance-threshold guidance (for chapter contracts)

The linters are candidate-detectors; a chapter contract that gates them all at `== 0`
will fail on genuine analytic prose. Calibrate the contract, not the prose:

- `hedge_count` counts deontic permission ("the model **may** reason") and counterfactual/possibility "could" as candidates, which the no-VAGUE-hedging policy permits. Gate a small budget, not `== 0`. (Deontic/epistemic modality is not separable by a clean rule.)
- `passive_voice_ratio` uses the precise true-passive detector; good analytic prose runs ~0.10–0.13 (legitimate "are performed", "is owed"). Gate `< 0.15`, not `< 0.10`.
- `modifier_budget_violations` flags precise paired adjectives ("private and lossy"). Allow a small budget.
- Keep the anti-slop / correctness gates strict at `0`: `ai_fingerprint_total`, `listicle_abstract_count`, `rhythm_violations`, `citation_token_count`.

## Style guide

`references/russellian-style-guide.md` is the authoritative catalog, organised across vocabulary, voice, atomicity, flow, and structure.

When prose is compliant but lifeless, read `references/russell-corpus-map.md` and compare against the 50-paragraph Russell corpus indexed in `assets/russell-corpus/index.json`.

For a fused target that adds Feynman's intuition and Hoskinson's momentum to
Russell's rigor, read `references/triadic-voice-guide.md`. It draws on three
exemplar corpora: `assets/russell-corpus/index.json`,
`assets/feynman-corpus/index.json` (pointers only), and
`assets/hoskinson-corpus/index.json` (built by `tools/build-voice-corpus`). The
discipline linters remain the quality floor for all three voices.

## References index

- `references/russellian-style-guide.md` — authoritative style catalog.
- `references/russellian-vitality-guide.md` — vitality and rhythm rules for compliant-but-lifeless prose.
- `references/russell-corpus-map.md` — guide to the 50-paragraph Russell corpus.
- `references/triadic-voice-guide.md` — Russell × Feynman × Hoskinson fusion guide for the three-voice target.
- `references/how-i-write-maxims.md` — Russell's seven sentence-craft maxims.
- `references/logical-atomism-for-writers.md` — IF/AND IF/THEN decomposition for tangled conditionals.
- `references/before-after-examples.md` — worked rewrites paired with the rule each one applies.
- `references/section-headers.md` — taxonomy of section/subsection titles (four use / three ban categories) with the EpochPoET case study.
- `references/negative-triggers.md` — genres in which the skill refuses to activate.

## Composes with

- `book-knowledge` — ingests sources whose claims this skill later polishes.
- `book-compose` — calls this skill as the prose-polish stage of chapter drafting.
- `book-review` — runs persona reviews on prose already compliant with this skill.
- `book-qa` — queries chapters drafted under this style.
- `book-thesis` — provides reasoning layer for entailment-backed prose discipline.
- `humanizer` — strips residual AI-writing tells after this skill runs.

## Usage

- `/russellian-style` — invoke on the current selection or supplied path.
- "Apply Russell style to this paragraph" — single-passage rewrite.
- "Run the linters on chapter 3" — lint-only pass, no rewrite.
- "Russell pass on `draft.md`" — full pass with `style-pass-report.md`.

## Tests

Run with `pytest` from the skill root. Fixtures in `tests/fixtures/`; trigger calibration in `tests/trigger_tests.yaml`.

From CharlesHoskinson/russellian-book-suite, MIT.

More from CharlesHoskinson/russellian-book-suite