write-from-vault
Write from Vault
Inspired by Matt Pocock's draft writing-fragments, writing-shape, and writing-beats skills in mattpocock/skills.
Your notes are the quarry. The article is the building. Every claim keeps its receipt.
Writing from a vault is not summarising it. Notes are atomic and written for you; an article is one argument written for a reader. This skill moves between the two in three stages, and the claim ledger carries the evidence across so nothing in the final piece is unsourced, superseded, or stronger than its source.
Before you start
Read .agent/context/knowledge-vault.md first (fallback: docs/agents/knowledge-vault.md). If missing, run /setup-knowledge-vault before continuing. If missing and .agent/context/knowledge-vault.example.md exists, ask the user to copy it and fill in their own path; never use the placeholder. Also read vault Indexes/Context Index.md when present.
Settle three things with the user, one question at a time, each with your recommended answer:
- Question: the one question the article answers for the reader (not a topic: "Why do coding agents forget, and what fixes it?", not "agent memory").
- Reader: who, and what they already know (the grounded concepts at the start).
- Form and length: blog post, internal memo, README section, talk outline; rough word count.
Create the draft workspace (see DRAFT-WORKSPACE.md):