library:ingest — Source Ingestion

Add source material to the vault's sources layer.

When to Use

• Adding new source material (PRDs, session notes, research papers)
• Initial vault population after library:config --init-vault
• Importing a batch of files from another location

Single Mode (default)

Takes one file, directory, or URL.

Step 1: Classify Source

Determine contamination tier:

• raw — unedited human-authored content (PRDs, meeting notes, specs)
• curated — human-edited but may include AI assistance
• llm-generated — primarily AI-generated content (session transcripts, compiled articles)
• external — third-party content (vendor docs, framework references)

Ask user to confirm tier if ambiguous.

Step 2: Categorize

Determine content category (e.g., "prds", "session-notes", "security-audits"). Suggest based on filename/content. Ask user to confirm.

Step 3: Ingest

Call library:vault:ingest with vault_path, source_path, tier, category.

Step 4: Report

Display: what was ingested, where it landed, what compile targets it affects.

Step 5: Rebuild Graph (if enabled)

If graphify.auto_rebuild: true in config, call library:graph:rebuild.

Batch Mode (--batch)

Loop for multiple sources:

1. Prompt: "Add a source path (or 'done'):"
2. User provides path → classify tier → categorize → ingest → report
3. Repeat until user says "done"
4. Display batch summary: total files, categories created, compile targets affected
5. Single graph rebuild at the end (not per-source)

Chained from config

When library:config --init-vault completes, it asks "Want to ingest sources now?" If yes, enters batch mode automatically.

Token Budget

Weight: Light-Medium Estimated context cost: ~1000 tokens (single), ~1500 tokens (batch) Subagent delegation: No

MCP Tools Used

• library:vault:ingest — classify and bucket source material
• library:graph:rebuild — rebuild knowledge graph (if enabled)
• library:config:get — read vault path and graphify settings