Summary

Add provenance metadata and admission validation for generated capability tools before they enter the OpenHuman tool registry.

Background

Generated tools are now supported as wrappers, but OpenHuman needs a generic admission layer that can reject malformed or unsafe generated tools before the model can see or call them.

Acceptance criteria

• Generated tools can carry provider id, capability id, source digest, risk level, and policy surface metadata.
• Admission validation rejects missing or invalid provenance when enforcement is enabled.
• Admission validation rejects unsafe tool names, duplicate names, invalid schemas, missing risk metadata for write/external capabilities, and disabled/untrusted providers.
• Diagnostics report whether a generated tool was admitted or rejected and why.
• Existing generated tools without provenance continue to work when admission enforcement is disabled.
• Add focused Rust tests for allowed tools, rejected tools, duplicate names, provider trust failures, and diagnostics output.

Non-goals

• Do not add any runtime-specific bundle format.
• Do not execute external code.
• Do not add frontend install/import flows.