Universal Specialist Agent Factory (USAF) is a minimal but production-minded framework for defining, certifying, and executing specialized agents with explicit capabilities and policy-bound execution.
- typed agent specifications
- explicit capability declarations
- policy-aware execution permits
- deterministic runtime execution checks
- testable validation and failure modes
- GitHub-ready repository structure for contribution and release
This repository is intentionally designed as the foundation for an open-source agent platform. The first milestone is not a massive app; it is a reliable core:
- define an agent
- validate its capabilities
- authorize execution via a permit
- execute within a bounded policy model
- audit the result
python -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
pip install -e .[dev]
pytest -qfrom usaf import AgentFactory, AgentSpec, Capability
factory = AgentFactory()
agent = factory.build(
AgentSpec(
name="market-analyst",
role="research",
capabilities=(
Capability("read_market_data", "Read market feeds"),
Capability("run_simulation", "Run local simulation"),
),
)
)
permit = factory.runtime.request(agent, "read_market_data")
result = factory.runtime.execute(agent, "read_market_data", value={"market": "BTC/USD"}, permit=permit)
print(result)The core is intentionally small and explicit:
AgentSpecdefines the contract for an agentCapabilitydefines the allowed action surfaceAgentFactoryvalidates and creates agentsExecutionRuntimeissues permits and executes only authorized actions
pytest -q
python -m ruff check src testsPlease review SECURITY.md before reporting issues.
MIT