wiki-narrate writes a readout on one topic from an Obsidian vault. You pick a voice (briefing, plain-language or lecturer), and the skill builds a ledger of claims with supporting page links before it writes any prose.
Every factual sentence carries an adjacent citation, inferred connections and conflicts are marked, and a Coverage footer lists cited pages and gaps. Output is shown in chat and saved to a readouts folder only when you pass the save flag.
What You Get
- A readout titled with the topic and voice, in that voice’s fixed section order
- Adjacent citations on each factual sentence, as wikilinks or Markdown links
- Inference and ambiguity markers where evidence is indirect or conflicting
- A Coverage footer with cited pages, count of inferred statements and gaps
- An optional saved file under the readouts folder
Common Use Cases
- Getting a one-page briefing on a topic before a meeting
- Explaining a technical subject to a newcomer using only your own notes
- Producing a progressive lecture outline from a cluster of related pages
What Makes It Unique
It keeps a claim ledger and audits citations before writing, so it never closes a gap with outside knowledge. An approved-facts gate checks statements about the owner or their company against a facts file and drops retired claims, and an owner voice file can shape prose without adding claims. Saved readouts are derived output, excluded from later retrieval and kept out of the index and manifest.
Under The Hood
The skill is instructions plus one reference file of voice outlines. It reads vault pages and, when a QMD collection is configured, uses it to find candidates. It runs the obsidian-wiki command line tool to check facts, show and check the owner voice, log the run and refresh the hot cache. The only file write besides the log is the saved readout. It does not fetch web pages or call remote APIs.