Adaptive writing-style plugin for Claude. Learns your voice from real edits over time, in addition to whatever you captured up front via cortex's /setup-voice.
- Draft-first workflow — every written output (email, social post, doc) is presented as a draft for your review. Never sent directly.
- Edit detection — when you edit a draft inline in chat OR externally (Gmail, Drive, LinkedIn), the plugin compares draft-vs-final and looks for style patterns.
- Pattern recognition — only proposes style-guide updates after a pattern appears 2+ times. One-offs don't pollute the guide.
- Per-medium style files — separate
style-email.md,style-social.md, etc., so what works for a LinkedIn post doesn't get applied to a client email. - Two-stage triage — cheap classifier decides if an edit reveals a pattern; expensive synthesis only runs if it does. Cost-efficient at scale.
- Audit command —
/style-reviewscans your style files for contradictions, stale rules, and unvalidated patterns. Run monthly.
| Plugin | What it does | How writing-style integrates |
|---|---|---|
cortex /setup-voice |
One-time capture of high-level voice descriptors (3 words, banned phrases, hook patterns) | Baseline. writing-style refines over time on top of this. |
| bizdev-outreach | Drafts per-contact outreach | When you edit a draft before sending, run /style-learn to feed the pattern back. |
lead-engine /lead-draft |
Drafts 3-touch DM cadence | Same — /style-learn after you tweak. |
news-curator post-assembler |
Drafts LinkedIn roundup posts | Same — /style-learn after you publish your edited version. |
| All drafting plugins | Read voice rules at draft time | Read <config-root>/voice.md (canonical) + medium-specific <config-root>/style-{medium}.md (refined by writing-style). |
Recommended: via the BrightWayAI marketplace.
/plugin marketplace add BrightWayAI/nucleus
/plugin install writing-style@nucleus
Run /setup-style. The interview:
- Reads
<config-root>/voice.mdif it exists (from cortex's/setup-voice). Uses it as baseline. - Asks you for 5-10 writing samples across mediums (work email, casual email, LinkedIn post, doc, slack DM, etc.) — OR pulls them from Gmail/Drive if connectors are available.
- Analyzes patterns: voice, structure, sentence style, common phrases, "what you never do."
- Writes/refines
<config-root>/voice.md(high-level) and creates medium-specific files (style-email.md,style-social.md) if 3+ samples exist for that medium.
Saved at:
<config-root>/voice.md— high-level voice (shared with all drafters)<config-root>/style-email.md— email-specific (writing-style only)<config-root>/style-social.md— social-specific<config-root>/style-{medium}.md— any other medium
Plus plugin config at <config-root>/plugins/writing-style.user-context.md (edit-detection paths, confidence threshold, per-medium tracking opt-ins).
| Command | Purpose |
|---|---|
/setup-style |
Bootstrap or update style guides. Pulls writing samples, analyzes, writes initial files. |
/style draft [medium] [purpose] |
Draft something in your voice. Reads voice + medium-specific style files. Presents as draft for review. |
/style-learn |
Analyze a draft-vs-final diff and propose style-guide updates if a pattern emerges. Run after sending an edited version of any draft. |
/style-review |
Audit existing style files. Flags contradictions, unvalidated rules, suggested additions based on recent writing history. Run monthly. |
MIT.