ingest

/llm-wiki:ingest — Ingest a Source into the Bundle

Entry point for the supervised ingest loop. Detects the source type and delegates to the appropriate adapter skill.

Usage

/llm-wiki:ingest <source> [--auto]
/llm-wiki:ingest <source1> <source2> ... [--auto]

--auto: skip review pauses; fan out across all sources; one review at the end.

Source type detection

Input	Routes to
project.dataset (contains exactly one .)	ingesting-bigquery
URL starting with http:// or https://	ingesting-web
Local path to a file or directory	Direct read → authoring-concepts
seeds.txt or seeds.example.txt	ingesting-web (file of seed URLs)

What happens

See ingesting-sources skill for the full supervised loop. Summary:

1. Detect source → activate the appropriate skill
2. Adapter lists concepts → show plan to owner → wait for confirmation
3. Fan out okf-concept-enricher subagents (one per concept)
4. Report diff summary → wait for owner review
5. If multiple sources: proceed to the next only after owner go
6. Suggest /llm-wiki:index and /llm-wiki:log

With --auto

Skip pauses. Fan out all enrichers concurrently. Present one consolidated review at the end. Use when you trust the source and want unattended batch processing.

After ingest

Always remind the owner to:

/llm-wiki:index   — regenerate index.md files
/llm-wiki:log     — record what was ingested
/llm-wiki:validate — confirm conformance