Why this is critical

Most severe autonomous failures happen at execution time. consensus-agent-action-guard adds a final arbitration checkpoint before side effects occur.

What it gates

destructive or irreversible actions
external operations with blast radius
high-risk actions requiring human confirmation
policy-sensitive tool invocations

Decision states

ALLOW | BLOCK | REQUIRE_REWRITE with required action steps and rationale.

What teams gain

consistent action governance across agents
safer autonomous operations
explicit receipts for who approved what and why
replayable incident analysis from board artifacts

SKILL.md perspective

Structured, strict, and easy to integrate into any tool-calling orchestrator.

AI-SELF-IMPROVEMENT perspective

Autonomy improves when execution is policy-constrained and outcomes are logged for feedback loops.