Purpose
Track the confirmed weaknesses from the 2 August 2026 three-agent Terra review of asdecided/core at fd1c20191351ba74524afd993290c6dfd1c56b16.
The review found no P0. The concentration of risk is MCP: transport hardening, protocol validation, audit correctness, and response-boundary enforcement. The Rust index, freshness, export, and Sentry paths were comparatively sound. The full decided-mcp suite passed while several invalid requests were still accepted, so negative and conformance coverage is part of the corrective programme rather than optional follow-up.
Recommended execution order
Phase 1 — secure and validate the MCP request boundary
These should be designed together so Origin policy, size/deadline controls, version negotiation, and JSON-RPC validation happen before era-specific dispatch. Valid frozen legacy bytes must remain unchanged.
Phase 2 — make compatibility independently testable
Core should consume the language-neutral fixtures directly. This can begin alongside Phase 1, but Phase 1 is not complete until its negative cases are represented in the shared fixture set or an explicitly Core-only integration suite.
Phase 3 — restore MCP product guarantees
Audit semantics and the response budget are user-facing trust contracts. Any intentional changes to ADR-084, ADR-098, or ADR-033 must be recorded as explicit superseding artifacts rather than silent implementation drift.
Phase 4 — make mutation safe
Root confinement lands first. Transactional multi-file replacement must not expand the filesystem attack surface.
Phase 5 — close operability and correctness gaps
Documentation should follow the resolved runtime contracts. Cross-platform runtime coverage then verifies the supported distribution surfaces.
Programme exit criteria
Purpose
Track the confirmed weaknesses from the 2 August 2026 three-agent Terra review of
asdecided/coreatfd1c20191351ba74524afd993290c6dfd1c56b16.The review found no P0. The concentration of risk is MCP: transport hardening, protocol validation, audit correctness, and response-boundary enforcement. The Rust index, freshness, export, and Sentry paths were comparatively sound. The full
decided-mcpsuite passed while several invalid requests were still accepted, so negative and conformance coverage is part of the corrective programme rather than optional follow-up.Recommended execution order
Phase 1 — secure and validate the MCP request boundary
These should be designed together so Origin policy, size/deadline controls, version negotiation, and JSON-RPC validation happen before era-specific dispatch. Valid frozen legacy bytes must remain unchanged.
Phase 2 — make compatibility independently testable
Core should consume the language-neutral fixtures directly. This can begin alongside Phase 1, but Phase 1 is not complete until its negative cases are represented in the shared fixture set or an explicitly Core-only integration suite.
Phase 3 — restore MCP product guarantees
Audit semantics and the response budget are user-facing trust contracts. Any intentional changes to ADR-084, ADR-098, or ADR-033 must be recorded as explicit superseding artifacts rather than silent implementation drift.
Phase 4 — make mutation safe
Root confinement lands first. Transactional multi-file replacement must not expand the filesystem attack surface.
Phase 5 — close operability and correctness gaps
Documentation should follow the resolved runtime contracts. Cross-platform runtime coverage then verifies the supported distribution surfaces.
Programme exit criteria
asdecided/specis the authoritative language-neutral MCP compatibility source.