wiki-capture saves what you learned in a chat. Full mode classifies the content as a synthesis, concept, source summary, decision or session, rewrites it as declarative present-tense knowledge, and writes a cross-linked page in the right folder with summary, provenance and confidence frontmatter.
Quick mode drops findings into a raw staging folder in under a minute, and correction mode records one atomic fix to a derived claim while leaving the original source untouched.
What You Get
- A cross-linked wiki note with title, tags, sources, summary, provenance and lifecycle frontmatter
- Quick mode raw files, one per topic cluster, ready for later promotion with the ingest skill
- Correction records with a source hash check that proves the original was not modified
- Profile facts and open threads saved for the next session, with confidence and expiry dates
- A confirmation showing the saved path, title and type
Common Use Cases
- Saving a debugging conclusion as soon as a session reaches it
- Recording an architectural decision together with its rationale
- Correcting a wiki claim that a teammate or a test later disproved
What Makes It Unique
Quick mode has a keep-or-skip gate so it can run from a session-end hook without filling the staging folder, leaning toward skipping when automatic and toward keeping when you call it by hand. Correction mode compares a hash of the immutable source before and after the write and reports each downstream consumer as open, complete or not applicable. Named-vault routing lets a phrase such as save this to research choose the vault.
Under The Hood
The skill is instructions for the agent. It reads the vault index and writes Markdown notes, raw staging files and tracking records inside the vault. It runs the obsidian-wiki command line tool through the shell to sync the log, index and hot cache, to set profile facts and to add or close todo items. Correction mode computes file hashes of the source with shell tools. A configured QMD collection is refreshed with the qmd command. It does not fetch web pages or call remote APIs.