| name | seed |
| description | Seed an OKF bundle from a structured source — a SQLite database, an OpenAPI or JSON-Schema file, or a directory tree — minting one conformant concept per table, endpoint/definition, or file. Use when the user says "seed from", "import my database schema", "catalog this repo", "build concepts from this OpenAPI spec", or "generate a knowledge base from <source>". For importing one markdown file use `add`; for one non-markdown file use `convert`.
|
| argument-hint | <source> [into <bundle path>] |
| user-invocable | true |
| allowed-tools | Bash Read |
okfm: seed
Generate many concepts at once from a structured source. The engine is offline
and dependency-free; every write previews first; generated concepts are
curated_by: agent until a human verifies them.
- Identify the source and the bundle. The engine auto-detects the mode
(
sqlite / openapi / tree); pass --mode to override, and --dest-dir
to choose the subdirectory (defaults to a slug of the source name).
- Preview:
python "${CLAUDE_PLUGIN_ROOT}/scripts" seed <source> --bundle <root> --dest-dir <subdir> --dry-run --json
Show the user the list of concepts that would be created and where.
- On their OK, run the same command without
--dry-run. Existing files are
skipped unless --force. The engine sets timestamp and curated_by: agent,
regenerates the index tree, and appends a log.md entry.
- Run
check and report. Remind the user the new concepts are agent-drafted —
they should review and flip verified ones to curated_by: human.
Modes
- sqlite — one
Table concept per table, with a # Schema column listing and a sqlite://<db>#<table> resource.
- openapi — one
API Endpoint per path+method (OpenAPI), or one Schema per definition (JSON Schema), or one Schema for a bare schema document.
- tree — one
File concept per file, directory structure preserved; small text files embed their content; binaries and dot-directories (.git, node_modules, …) are skipped.
For ongoing growth from the web, use the enrich skill.
