Goal

Implement the core pre-execution guard for AI-agent payment proposals.

Deliverables

• Policy evaluation engine.
• Signed-intent validation.
• Freshness and expiry checks.
• Persistent nonce replay protection.
• Deterministic ACCEPT, HOLD, and BLOCK decisions.

Acceptance checks

• Valid signed intent returns ACCEPT.
• Replayed signed intent returns BLOCK / INTENT_REPLAYED.
• Invalid policy returns HOLD or BLOCK.
• Blocked proposals never reach the mock payment rail.

Grant context

Application: 2026-08-00b
Fund: NGI TALER
Reviewer docs: docs/NGI_TALER_REVIEWER_PATH.md