Brain Ingest

Brain Ingest

Use this skill for one focused update to ontology-brain/.

Principles

• The source is canonical. If the brain and source disagree, the source wins.
• Store each fact once on its canonical page.
• Propagate references. A canonical page edit is incomplete when index rows, related pages, or typed links still point at stale meaning.
• Label uncertainty with [INFERENCE], [SPECULATION], or [UNVERIFIED].

Workflow

1. Read AGENTS.md and ontology.md.
2. Identify the source type and the fact owner.
3. Create or update a sources/ page when the input is durable enough to cite.
4. Create or update the relevant object or concept page.
5. Add typed links in frontmatter when a relationship matters to future navigation.
6. Search for the object name, stable ID, and page slug, then update stale references.
7. Regenerate index.md with python3 .claude/skills/brain-lint/scripts/build_index.py.
8. Append log.md with ## [YYYY-MM-DD] ingest | <title>.
9. Run python3 .claude/skills/brain-lint/scripts/wiki_link_check.py.

Page Placement

• Use objects/<object-type>/ for things with identity.
• Use concepts/ for reusable logic or patterns.
• Use sources/ for faithful summaries of raw inputs.

Create object-type folders only when needed. Prefer concrete singular names.

Reference Propagation Checklist

• Grep the identifier, title, and slug everywhere.
• Update stale prose.
• Update typed links if relationships changed.
• Update both ends of supersedes and superseded_by.
• Run link repair after renames or moves:

python3 .claude/skills/brain-ingest/scripts/wiki_link_repair.py --dry-run

Review dry-run output before running without --dry-run.

Gate

The change is not complete until the generated index is current and the lint gate reports 0 problems.