Agent tools & model relay protocols.

Off-surface execution — anything that leaves the runtime through a tool call or a model relay — is the most permissive boundary in the system and therefore the most strictly governed. The two protocols partition the surface:

• ATL-001 (Agent Tools & Logic Protocol) governs executable tool logic that a runtime agent invokes.
• MRP-001 (Model Relay & MCP Protocol) governs the message-class interface between models, relays, and MCP handlers.

Both are Tier 2 — Statutory. Both have status ACTIVE. Neither grants execution authority on its own; authority remains artifact-derived per the governance framework.

The two protocols overlap at runtime/scripts/agent_tools/relay/mcp/. ATL-001 owns tool logic; MRP-001 owns relay/MCP framing. A handler that lives at the overlap is bound by both.

Header (verbatim)

Governance Tier: Tier 2 — Statutory
Authority ID:    ATL-001
Subsystem:       runtime_agent_tools
Status:          ACTIVE

Scope

This protocol governs executable logic under runtime/scripts/agent_tools/ and runtime/scripts/agent_tools/relay/mcp/. It does not grant execution authority. Authority remains artifact-derived per governance documents.

Structural requirements (verbatim)

• Tools MUST run through governed orchestration surfaces for Gate-Exec claims.
• MCP-facing tools MUST enforce active_envelope_hash gating before mutation-capable operations.
• Tool outputs intended for governance/audit surfaces MUST be deterministic and machine-parseable.
• Drift or heartbeat outputs are evidentiary/diagnostic and MUST NOT be treated as permission artifacts.

Header (verbatim)

Governance Tier: Tier 2 — Statutory
Authority ID:    MRP-001
Subsystem:       model_relay_mcp
Status:          ACTIVE

Scope

This protocol governs runtime/scripts/model_relay/, runtime/agents/mcp/, and runtime/scripts/agent_tools/relay/mcp/.

Interface compliance (verbatim)

• MCP handlers/messages MUST remain schema-compatible with canonical message classes in MCP message modules.
• Relay and observer outputs (for example dist/rael_observe_report.json) are diagnostic/evidentiary only.
• Any mutation-capable MCP path MUST enforce active_envelope_hash and fail closed when invalid.

Both protocols name the same gate: active_envelope_hash. This is the load-bearing connector control. Every mutation-capable path through ATL-001 or MRP-001 must:

1. Read the active envelope hash from the runtime state.
2. Verify the envelope is current and the path is admitted under it.
3. Fail closed when the envelope hash is missing, expired, or does not admit the path.

An expired or missing envelope is not a soft warning. The mutation path must abort. Read-only and observer paths are exempt because they cannot affect governed state.

Both protocols draw a hard line between evidentiary outputs and permission artifacts. Diagnostic outputs are useful for audit, replay, and operator triage. They do not grant authority and may not be substituted for an envelope.

Violations are governance boundary breaches. Both protocols escalate identically:

• ATL-001 §3: violations are treated as governance boundary breaches and must be escalated through Phase 6/6.5 controls.
• MRP-001 §3: contract violations, envelope-gate bypass, or failed relay integrity checks trigger Phase 6/6.5 compliance escalation.

Phase 6/6.5 admission is documented in meetLab Operations · IX. Both protocols are authorized under the meetLab Governance Framework v4.1.0.

Before promoting a tool, relay, or MCP handler to a governed surface, walk this list. Every item must be satisfied; none may be deferred.