
Style Analysis
Define how prose voice is captured in split style reference files so writer and critic agents load the right `-f` chunks per scene or character.
Install
npx skills add https://github.com/haowjy/creative-writing-skills --skill style-analysisWhat is this skill?
- Derives style dimensions from the text instead of imposing a fixed taxonomy
- Splits style into files at boundaries where callers need one chunk without another
- Documents dialogue vs narration splits and scene-bound vs character-bound axes
- Frames each file as a context-selection unit for orchestrators passing `-f` to writers
- Reference skill for creating, updating, and evaluating project voice pattern files
Adoption & trust: 1 installs on skills.sh; 241 GitHub stars; 3/3 security scanners passed (skills.sh audits); trending (+100% hot-view momentum).
Recommended Skills
Journey fit
Canonical shelf is Build because the primary artifact is agent-facing style reference files and orchestration context boundaries—not the finished novel itself. agent-tooling fits skills that shape what context files agents receive and how multi-agent creative pipelines are wired.
Common Questions / FAQ
Is Style Analysis safe to install?
skills.sh reports 3 of 3 security scanners passed. Review the Security Audits panel on this page before installing in production.
SKILL.md
READMESKILL.md - Style Analysis
# Style Analysis How to analyze a project's prose and produce style reference files that writer and critic agents can use. ## Style Has Dimensions Style varies independently along multiple axes: some bound to a character, some to a scene type, some to both, some that cross everything. The first job is identifying what dimensions exist in *this* project, not arriving with a predetermined taxonomy. Look at the text: what varies independently? If a character's narration voice changes by scene type, that's two dimensions interacting. If action pacing stays consistent regardless of narrator, that's scene-bound. The text tells you where the boundaries are. ## File Splitting Style files are the unit of context selection: an orchestrator passes individual files to writers via `-f`. Every file boundary is a context decision: would an agent ever need this chunk without that chunk? Split where a caller would plausibly want one part without the other. A character with distinct dialogue and narration modes might need separate files. A character with a simple, consistent voice needs one. A scene type that works the same regardless of POV is its own file. ## What to Analyze Dimensions worth investigating: the text determines which matter: - **Sentence patterns**: length distribution, rhythm, how it shifts with emotional intensity. Fragment usage, compound tendencies. - **Interiority**: depth of internal monologue. Direct thought, indirect thought, stream of consciousness: when each activates. - **Vocabulary and register**: recurring word choices, domain language, register shifts. - **Dialogue patterns**: how characters sound distinct. Tag frequency, action beats, subtext delivery. - **Humor mechanics**: techniques, timing, what's played for laughs vs what's sacred. - **Emotional approach**: physical manifestation vs named emotions, how much space emotional moments get. - **Sensory detail**: privileged senses, how density shifts by scene type. - **Pacing and paragraph rhythm**: how paragraph length and whitespace shift between scene types. ## File Structure Each style file teaches a voice through principles, not catalogs: - **Principle**: the core insight in a few sentences. What's the pattern? Why does it work? - **Representative examples**: one or two with chapter citations showing the principle in action. - **Chapter pointers**: where to see more of the pattern in context. A writer who internalizes the principle produces natural variation. A writer following an exhaustive checklist produces something mechanical. Each file should be self-describing: a caller reading it should understand what it covers and when to load it. ## Patterns vs Problems Intentional patterns go in style files: the voice a writer should reproduce. Unconscious tics and inconsistencies go in the issues directory: problems for the critic to watch for and the author to address in revision. The test: would the author want a writer agent to reproduce this? If "for a moment" appears 29 times across 17 chapters, that's a tic. If an emotional technique works in chapters 2 and 15 but is absent from chapter 11, that's an inconsistency to log as an issue. ## Quality Tests 1. **Voice test**: could a writer, reading only this file and a scene brief, produce prose the author recognizes as their voice? 2. **Brevity test**: could a writer internalize this file in one read? If they need to keep it open as reference while drafting, it's over-prescribed.