docs: clarify audit receipt integration points

[maxpetrusenkoagent]

Summary

• Document how structured logging can feed cryptographic action receipt systems
• Point enterprise governance integrations at MessageEvent, ToolCallEvent, LLMCallEvent, and InterventionHandler boundaries
• Clarify that AutoGen stays receipt-format neutral while supporting AAR-style adapters

Verification

• uv run poe markdown-code-lint
• uv run poe docs-check (fails on pre-existing optional autogen-ext autodoc imports: cv2, chromadb, mem0, redisvl, llama_cpp, ollama, semantic_kernel, graphrag, json_schema_to_pydantic)

Related to #7353

Created with: Hermes Agent

[yun520-1]

@maxpetrusenkoagent The audit receipt integration points you're documenting align well with HeartFlow's approach. We use a 4-stage audit trail that maps directly to MessageEvent / ToolCallEvent boundaries:

1. Pre-execution — log the intent + expected outcome (maps to InterventionHandler)
2. Execution — log the raw tool call + parameters (maps to ToolCallEvent)
3. Post-execution — log the result + verification verdict (maps to MessageEvent)
4. Decision trace — which decision rules fired and why (HeartFlow-specific, could be an extension to LLMCallEvent)

For the AAR-style adapter format, HeartFlow outputs structured receipts as JSON with confidence scores on every verification step. Happy to share the schema if it helps with the format-neutral approach you're aiming for.

HeartFlow: github.com/yun520-1/mark-heartflow-skill