Skip to content

/draft - Content Creation
Runtime ~5 minutes
Reads Vault context (semantic search), voice-exemplars.md, anti-slop rules
Writes Stdout first, then theme file or clipboard on approval
Model Opus

What It Does

Generates outbound content - LinkedIn posts, stakeholder emails, thought leadership pieces, internal memos - synthesised from vault context. Turns your internal thinking into external positioning.

Why It Matters

Writing from scratch is slow. Writing from a vault full of refined thinking, meeting notes, and framework definitions is fast. The problem is that internal language doesn't translate directly to external audiences.

/draft bridges this gap. It mines your vault for source material, applies voice calibration from real writing samples, enforces anti-slop rules as hard constraints, and adapts framing per audience.

How It Works

graph LR
    A["/draft type topic"] --> B[Parse Input]
    B --> C[Mine Vault]
    C --> D[Voice Calibration]
    D --> E[Draft to stdout]
    E --> F{Approved?}
    F -->|Iterate| E
    F -->|Save| G[Theme file]
    F -->|Clipboard| H[Rich HTML copy]
  1. Parse - Identifies content type (linkedin, thread, email, memo), topic, and target audience
  2. Mine the vault - Semantic search for conceptual context plus grep for specific framework definitions, meeting proof points, and prior emails for tone calibration
  3. Voice calibration - Reads voice-exemplars.md and matches your actual writing cadence, not generic executive prose
  4. Draft - Writes to stdout first. Never saves until you approve.
  5. Present - Offers three paths: iterate, save to theme folder, or copy to clipboard as rich HTML

Content Types

Type Length Key Constraint
linkedin 150-250 words One idea. Hook in first line. End with a question, not a summary.
thread 3-5 posts Each stands alone but builds. First post hooks, last post provokes.
email Varies Stakeholder-specific framing from claude.md. Prose flow, not memo format.
memo 1-2 pages Decisions and asks up front. Evidence-backed. Target audience stated.

The Key Innovation

Three things separate this from "write me a LinkedIn post":

Voice calibration from exemplars. The skill reads voice-exemplars.md - real samples of your writing at its best. It matches your actual cadence: sentence length variation, verb choices, level of specificity. Not "write in a professional tone" but "match this specific person's writing patterns."

Anti-slop as hard constraints. Every banned word, phrase, and structure from the Anti-Slop Rules applies double to outbound content. No "delve", no em dashes, no "in the fast-paced world of". External content that reads like AI output defeats the purpose. These aren't suggestions - they're compile errors.

Batch options for short-form content. For LinkedIn hooks, taglines, and email subject lines, the skill generates 5-10 variants instead of iterating one proposal at a time. You pick faster from breadth than from repeated depth. Single-proposal iteration produces 5+ rejection cycles. Batch options resolve in one round.

The skill also applies challenge criteria during drafting, not after. Audience fit, confidentiality of specific numbers, tone for recipient - all checked before the first draft hits the screen. No waiting for a post-hoc /challenge to catch issues that should never have been written.

Example Usage

LinkedIn post about a framework:

/draft linkedin professional work architecture

Stakeholder update email:

/draft email project-update Jordan

LinkedIn thread:

/draft thread why most AI transformations fail

Customisation Guide

  • Voice exemplars - Create 99_System/context/voice-exemplars.md with 3-5 real writing samples. The skill calibrates against these. Without exemplars, it falls back to CLAUDE.md style rules, which are less precise.
  • LinkedIn voice vs internal voice - The skill translates internal language to external automatically. Technical labels become plain language. Basis points become percentages. Framework jargon becomes "one memorable label." Override this mapping in the SKILL.md if needed.
  • Email style - Follows the peer email style from CLAUDE.md: prose flow, warm opening, inclusive language, simple sign-off. Adjust per-theme in theme claude.md files.
  • Auto-copy - When you say "send it" or "copy it", the skill runs md2email for rich HTML clipboard copy. Configure the script path in your CLAUDE.md Tools section.