Ontology Brain

This is a scoped Markdown brain for durable, agent-maintained knowledge. It uses the Palantir Foundry Ontology as a modeling lens: objects are pages, properties are frontmatter, relationships are typed links, source material is summarized once, and reusable logic becomes concepts.

The point is modeling discipline inside plain Markdown so knowledge compounds instead of being re-derived every time.

Start Here

• AGENTS.md: operating protocol for agents.
• ontology.md: the local ontology definition.
• index.md: generated catalog of pages.
• log.md: append-only action log.

Structure

objects/   object pages, grouped by concrete object type
concepts/  reusable methods, principles, and logic
sources/   faithful summaries of raw source material

Scripts live as scoped skill assets:

.claude/skills/brain-lint/scripts/
.claude/skills/brain-ingest/scripts/

Pattern

Most document workflows answer from raw material every time. This brain instead compiles durable knowledge into a maintained graph. New source material updates canonical pages, typed links, source summaries, and the generated index.

Each object page has one canonical home. Other pages link to it instead of copying its facts.

Maintenance

Regenerate the index after page changes:

python3 .claude/skills/brain-lint/scripts/build_index.py

Run the health check after changes:

python3 .claude/skills/brain-lint/scripts/wiki_link_check.py

The brain uses relative Markdown links, not Obsidian wikilinks, so it stays readable in Obsidian, editors, and agent sessions.