| name | content-draft |
| description | Draft posts, newsletters, and pages in the user's own voice from _context/voice.md. Triggers on - draft a post, write a post, linkedin post, newsletter draft, write content, draft content, tweet, thread. |
Content draft
Write in the USER'S voice, never in generic AI voice. The voice file is law.
Process
- Load
_context/voice.md. If it still says STATUS: TEMPLATE, stop and offer system-setup first; drafting without a voice file produces generic slop.
- Load
_context/product.md, _context/icp.md, and _context/positioning.md so the draft serves the actual audience.
- Mine for material before inventing any: recent daily-log wins,
_ops/decisions.md, wiki pages on the topic, content/published/ for angles already used (avoid repeats).
- Draft. Match the samples in voice.md: tone, rhythm, vocabulary, opening style, closing style, formatting. Respect the never-use list absolutely.
- Save to
content/drafts/YYYY-MM-DD-slug.md with frontmatter: type: draft, platform, topic, created.
- Show the draft in chat for copy-paste. Offer one revision pass.
Publishing
- When the user says it shipped, move the file to
content/published/ and add published: date and link to its frontmatter.
- If a sibling
*-website/ folder exists in the workspace and the piece is a blog post, offer to convert it to MDX in the website's content/blog/ following that repo's frontmatter conventions (read an existing post there first).
Rules
- One platform per draft; a LinkedIn post and a tweet about the same idea are two files.
- True stories from the system beat invented examples; if there is no real material, ask for a story rather than fabricating one.