agent-atoms

bootstrap · v0.1.0 agent-atoms.com

At a glance

Specatoms-spec/v1
LicenseApache-2.0

Links

README

agent-atoms

Agent primitives canonicalized — personas, tool definitions, capability declarations, role boundaries, isolation constraints — across LangChain, AutoGen, CrewAI, Olympus and beyond.

agent-atoms is a *-Atoms catalog in the Convergent Systems ecosystem. It defines what exists in its domain — typed, versioned, machine-readable, composable, and open — so runtimes (and humans) can stand on shared infrastructure instead of reinventing it.

Structure

agent-atoms/
├── ATOMS.yml              # Catalog manifest
├── atoms/                 # Reusable building blocks
├── agents/            # Compositions assembled from atoms
├── rules/                 # Typed constraint vocabulary
├── schemas/               # Catalog-specific JSON Schemas
├── exports/               # CI-generated machine-readable exports
└── docs/                  # Human-readable documentation

Atom types

  • persona
  • tool-definition
  • capability-declaration
  • role-boundary
  • isolation-constraint

Rule types

  • capability-grant
  • isolation-rule
  • communication-pattern
  • supervision-hierarchy

Runtime consumers

olympus

How to consume

Machine-readable exports are published in exports/ on every release:

  • exports/manifest.json — lightweight discovery (name, version, counts)
  • exports/catalog.json — full catalog dump (every atom, composition, rule)

Exports are deterministic, signed, and versioned. See ATOMS.yml for the manifest and the conformance spec.

How to contribute

  1. Read ATOMS.yml to understand the catalog's atom types, compositions, and rules.
  2. Add a new atom under atoms/<type>/ or a composition under agents/<name>/.
  3. Open a PR. CI validates the schema, references, and exports.
  4. Larger structural changes go through the XAIP process.

Ecosystem

License

Apache-2.0 — see LICENSE.