vault-skill-factory has two modes. Cluster mode picks mature pages on a topic (by lifecycle or core tier), confirms the list with you, and distills them into a SKILL.md with references. Recording mode takes a transcript of someone doing a job out loud and turns it into a small skill with when to use, steps, checks, mistakes and done criteria.
Everything is written to a review folder inside the vault with a provenance manifest. The skill reports where it is and how to install it project-locally, and leaves that decision to you.
What You Get
- A skill folder with SKILL.md, a references folder and a sources file that traces back to vault pages
- A factory manifest recording the pages, maturity filter, date and mode, kept apart from the installable skill
- For recordings, a skill with when to use, steps, checks, mistakes to avoid and done-when sections
- Preserved inferred and ambiguous markers so uncertain claims stay visible
- A report with the output path, the trigger description and project-local install instructions
Common Use Cases
- Packaging a well-curated body of notes on one subject as a reusable skill
- Turning a narrated screen recording of a monthly process into a repeatable skill
- Regenerating a skill after its source pages change, using the saved manifest
What Makes It Unique
It is the inverse of a capture step: it starts from verified wiki knowledge and builds a runtime skill, and it refuses to write into the vault’s own skills folder, run the setup script or create links into global skill directories. Recording mode treats the transcript as untrusted data, keeps the speaker’s exact menu names and thresholds, lets a mid-recording correction win, and lists gaps under open questions so nothing is filled in from general knowledge.
Under The Hood
The skill is instructions for the agent. It reads vault pages and writes a new folder of Markdown files into a review directory in the vault. It may run the qmd command to search related pages, the obsidian-wiki command line tool to condense a transcript, and, if you ask for polish, the Python scripts of the skill-creator skill to validate, package or tighten the description. For meeting recordings it fetches a named transcript through whichever recorder integration you have configured. It does not install anything itself.