docs: propose Agent Action Boundary Evidence model

[dinpd]

Summary

• publish the experimental Agent Action Boundary Evidence community draft 0.1
• define a vendor-neutral trust model, evidence lifecycle, identifiers, core records, invariants, privacy, replay, and failure semantics
• map the draft non-normatively to OpenTelemetry, CloudEvents, MCP, and AgentPass
• link the proposal and its public feedback thread from the repository README

Community discussion

Discussion and interoperability feedback will remain open in #43 after this PR merges.

Validation

• python -m pytest — 99 passed
• python scripts/check_manifests.py 'examples/customer-support-*-agent.yaml' examples/provider-mcp-support-agent.yaml --max-risk 75 — passed
• all 8 JSON examples parse successfully
• local Markdown links resolve
• external standards references resolve
• whitespace and credential-pattern scans passed
• human editorial review approved the standalone framing, non-normative AgentPass mapping, telemetry/authority separation, and feedback questions

Risk

Documentation only. The draft repeatedly identifies itself as experimental and not an adopted standard. It includes no credentials, raw tool arguments, personal identifiers, or production endpoints.