skill-atoms

Discoverable, invocable AI skill primitives — inputs, outputs, side effects, and trust boundaries declared.

Skill primitives for AI runtimes — discoverable, invocable units of capability that agents load on demand. Each skill bundles a description, an invocation contract, and the prompt or reference material the agent needs to perform a bounded task. Composable, versioned, and catalog-addressable.

Catalog at v0.1.0

• 0 atoms across 0 types
• 0 skill compositions
• 0 compatibility rules

Atom types

• Seeding in progress — skill schema definition in progress (sub-project 2)

Civilization-grade properties

• Typed — every atom, composition, and rule validates against a JSON Schema.
• Versioned — every atom has a semver version field; compositions pin by version.
• Machine-readable — /exports/catalog.json is the canonical manifest.
• Composable — compositions reference atoms by ID; references resolve in CI.
• Open — Apache-2.0 licensed.
• Durable — no external dependencies in the hot path.