AgentFuse Evidence Schema v0.1.0
This is the first minimal MVP release for AgentFuse Evidence Schema.
It provides a small, structured schema for representing guarded agent tool-call decisions, especially blocked tool calls that should be recorded as completed policy decisions rather than failed tool executions.
What is included
PolicyResolutionEvidencefor auditable policy matching and ambiguity handling.NonExecutionEvidencefor blocked calls withstatus=not_executed,execution=not_started, andpayload_executed=false.LayeredBoundaryDecisionfor server/tool/call-level policy decisions.SafeTraceMetadatafor sanitized trace metadata without raw sensitive inputs.- A
langgraph_bigtool.create_agent()wiring demo that emits a blocked SQL mutation evidence record without compiling, invoking, or streaming the graph. - Focused schema tests covering policy resolution, non-execution boundaries, safe trace metadata, and call-level decisive gates.
Boundaries
This release is schema/demo work.
It does not claim production runtime protection.
It does not compile, invoke, or stream a live agent graph.
It does not call model providers.
It does not execute SQL, network calls, credential handling, user-data access, or protected payload bodies.
Quick validation
python -m pip install -e .
python -m pytest tests/test_agentfuse_evidence_schema.py
python examples/external_integrations/langgraph_bigtool/dhms_guarded_tool_registry_demo.pyExpected demo verdict:
DHMS_REAL_LANGGRAPH_BIGTOOL_API_WIRING_DEMO_PASS