-
Notifications
You must be signed in to change notification settings - Fork 3.1k
Open
Description
Proposal: Integrate AAR-MCP-2.0 RC1 (Verifiable Agent Interaction Layer)
I’m publishing AAR-MCP-2.0 Core Spec (RC1): a verifiable interaction layer for MCP/agent tool calls.
It provides tamper-evident journals + checkpoint signatures + conformance vectors, and exports an audit bundle reviewers can independently verify.
Why this matters
- Observability logs are not evidence. We need non-repudiable action receipts for high-risk tools (write config, transfer funds, etc.).
- This RC1 focuses on fail-closed enforcement and format-stable verification, not vendor lock-in.
RC1 entry (spec bundle + sha256 + conformance gate)
- Repo: joy7758/aro-audit
- RC1 Review入口已在 README 顶部(含 spec bundle + sha256)
- Conformance Gate: boundary vectors (base OK / attestations-only OK / predicate-tamper FAIL)
Review workflow
- Please review via GitHub Discussions:
- Or open an issue using our RC1 review template.
Ask
- I’d like feedback on:
- Record types + digest boundary definition
- Checkpoint semantics (range, merkle root, signature)
- Tool-level dependency policy (soft/strict/hard-gate)
- Conformance vectors coverage (what’s missing)
If your project exposes MCP tools, I can provide a minimal wrapper and a 30s demo bundle to validate integration.
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels