Frequently Asked Questions

Common questions about the Agentic Cognitive Governance Protocol.

General Questions

What is ACGP?

ACGP is a protocol for governing AI agents through continuous evaluation of their reasoning and actions. See Concepts Overview for details.

When should I use ACGP?

Use ACGP when you have autonomous agents that: - Make decisions without constant oversight - Take actions with real-world consequences - Need governance beyond simple access controls

When should I NOT use ACGP?

Don't use ACGP for: - Simple scripted bots (if/else logic) - Read-only systems - Sub-10ms latency requirements

Implementation Questions

Which conformance level should I use?

  • Claimable now: Standard
  • Optional additive badge: Regulated Controls Badge
  • Dev Mode: Development and testing only; not a conformance profile
  • Safety-Critical: Future profile track retained for architecture continuity; not claimable in alpha

See ACGP-6 Conformance.

What v1.0.0-alpha.2 Does Not Promise Yet

  • full non-Python local runtime parity
  • external identity provider federation
  • independent ARS attestation
  • broad multi-message-type batch portability
  • future-track Safety-Critical external claims

These boundaries are intentional for the current alpha release and should not be interpreted as protocol defects in the active Standard claim surface.

What Governance Tier should I choose?

Use Risk Assessment & Governance Tiers to determine the appropriate tier based on your agent's autonomy, adaptability, and continuity.

How much latency does ACGP add?

  • Governance Tier GT-0: ~10ms typical, <50ms max
  • Governance Tier GT-1: ~20ms typical, <100ms max
  • Governance Tier GT-2: ~50ms typical, <150ms max (recommended for production)
  • Governance Tier GT-3: ~100ms typical, <200ms max
  • Governance Tier GT-4: ~200ms typical, <350ms max
  • Governance Tier GT-5: ~500ms typical, <1000ms max

Technical Questions

Can I use ACGP with my existing agent?

Yes! ACGP is framework-agnostic. See Custom Protocol Integration.

Does ACGP work with MCP?

Yes! See MCP Integration Guide.

How do I handle interventions?

Implement handlers for the five primary intervention levels (OK, Nudge, Escalate, Block, Halt) plus orthogonal Flag semantics that can be combined with any primary level. See Concepts Overview.

Can agents recover from HALT?

HALT stops the agent completely. Recovery requires manual investigation and restart.

Trust and Security

How does trust scoring work?

The normative runtime mechanism is Trust Debt, not Trust Score. Trust Debt tracks intervention pressure over time, can tighten runtime handling immediately, and can trigger Governance Tier review or queued ARS reassessment. See Trust Debt & Runtime Posture.

Is ACGP secure?

ACGP includes security best practices. For high-security needs, use Safety-Critical controls as design guidance, but note that Safety-Critical conformance claims are not available in v1.0.0-alpha.2. See Security Hardening.

Are audit logs tamper-proof?

Standard Conformance uses structured logging. Safety-Critical controls add stronger cryptographic and durability expectations but are not claimable for v1.0.0-alpha.2.

Getting Help

Still have questions?

Quick Start Troubleshooting