# Governance Checklist > Maturity levels and assessment criteria for ACS implementation - **Category**: guardrails-acs - **Canonical URL**: https://designpattern.fyi/guardrails-acs/governance-checklist/ --- ## Description Maturity levels and assessment criteria for ACS implementation ## Additional Notes # Governance and Maturity Checklist Use this checklist to gauge where an agent — or your whole agent estate — sits. --- ## Foundational ### Basic Instrumentation - [ ] **User Message and Agent Response hooks instrumented** on every production agent - Every agent has input and output boundary hooks - Hooks are firing consistently and reliably - Basic policy evaluation is working - [ ] **Guardian Agent exists as a distinct, testable service** - Guardian Agent is deployed and accessible - Policy evaluation logic is centralized - Service has monitoring and health checks - [ ] **Basic trace events reach a central log** - Session start/end events are captured - Tool calls are logged - Trace data is stored and queryable ### Foundational Capabilities At this level, you can: - **Monitor** basic agent interactions - **Enforce** simple input/output policies - **Detect** obvious security issues - **Audit** agent activity at a basic level ### Foundational Gaps Typical gaps at this level: - No tool-level controls - No memory or knowledge protection - No structured security event mapping - No component inventory - No formal governance processes --- ## Managed ### Extended Instrumentation - [ ] **Tool Call Request/Result hooks instrumented** with parameter-level (not just tool-name) policy - Tool calls are evaluated before execution - Parameters are inspected, not just tool names - Tool results are screened before agent consumption - [ ] **Memory and Knowledge hooks instrumented** - Memory write operations are gated - Memory read operations are screened - Knowledge retrieval is inspected - Both read and write paths are protected ### Security Integration - [ ] **Trace events mapped to OCSF** and flowing into the enterprise SIEM - Security-relevant events are converted to OCSF format - Events are forwarded to existing SIEM infrastructure - Security team can query and analyze agent events - [ ] **Static AgBOM exists for every production agent** - Initial component inventory is complete - AgBOM is captured in CycloneDX or SPDX format - Component lists are reviewed and approved ### Governance Processes - [ ] **Verdict logs reviewed on a fixed cadence** - Guardian Agent decisions are reviewed regularly - False positive/negative rates are tracked - Policies are tuned based on review findings ### Managed Capabilities At this level, you can: - **Control** tool usage with parameter-level precision - **Protect** memory and knowledge from poisoning - **Integrate** agent security with existing security operations - **Inventory** and track agent components - **Review** and tune security policies systematically ### Managed Gaps Typical gaps at this level: - No protocol-level controls (MCP, A2A) - Static rather than dynamic inventory - No enterprise-specific classifiers - No explicit compliance mapping - No red-team testing program --- ## Optimized ### Protocol-Level Controls - [ ] **MCP and A2A protocol-level hooks instrumented** for every agent that uses them - MCP outbound connections are gated - MCP inbound traffic is screened - A2A task delegation is authenticated - A2A messages are authorized ### Dynamic Inventory - [ ] **AgBOM is dynamic** — regenerates automatically on capability discovery - New tool connections trigger AgBOM updates - MCP server connections are detected and recorded - Model switches are captured in inventory - AgBOM stays current without manual intervention ### Enterprise Integration - [ ] **Tier 3 enterprise classifiers (PII/PHI, cost, jurisdiction) plugged into the Guardian Agent** - Organizational data classification is integrated - Cost monitoring and alerting is active - Jurisdiction-specific rules are enforced - Industry-specific requirements are addressed ### Compliance Mapping - [ ] **Explicit compliance crosswalk maintained** (EU AI Act Article 14, NIST AI RMF) - ACS controls are mapped to regulatory requirements - Compliance evidence is systematically collected - Legal and risk teams understand the technical implementation - Audit-ready documentation is maintained ### Security Testing - [ ] **Red-team testing against the full ASI01–ASI10 catalog** is part of the release process - New agents are tested against all ASI scenarios - Red-team findings are addressed before deployment - Testing is repeated regularly, not just at release - Test results inform policy improvements ### Community Engagement - [ ] **Learnings fed back to the open ACS/AOS specification** - Implementation issues are reported to the community - Feature requests are submitted based on production experience - Best practices are shared with other adopters - Contributions are made to the open-source project ### Optimized Capabilities At this level, you can: - **Control** protocol-level agent interactions - **Maintain** real-time component inventories - **Enforce** organization-specific security requirements - **Demonstrate** compliance with regulatory requirements - **Validate** security controls through adversarial testing - **Influence** the evolution of the ACS/AOS standards ### Optimized Characteristics At this level, your implementation: - **Is comprehensive** — covers all ACS capabilities - **Is integrated** — works seamlessly with existing security infrastructure - **Is dynamic** — adapts automatically to changes - **Is validated** — tested through adversarial scenarios - **Is compliant** — mapped to regulatory requirements - **Is contributory** — improves the broader ecosystem --- ## Maturity Assessment ### Scoring Your Maturity Use this scoring system to assess your overall maturity: | Level | Criteria Count | Weight | |---|---|---| | Foundational | 3 items | 1 point each | | Managed | 5 items | 2 points each | | Optimized | 6 items | 3 points each | **Maximum Score:** 35 points ### Maturity Levels - **0-5 points:** *Initial* — Basic experimentation, no production deployment - **6-15 points:** *Foundational* — Production deployment with basic controls - **16-25 points:** *Managed* — Comprehensive controls with security integration - **26-35 points:** *Optimized* — Full ACS implementation with compliance validation ### Improvement Path Identify gaps at your current level and prioritize: 1. **Complete all items at your current level** before moving up 2. **Focus on high-impact items first** — security-critical controls 3. **Address dependencies** — some items require others to be complete 4. **Plan for governance** — processes are as important as technology --- ## Implementation Timeline ### From Initial to Foundational (1-3 months) - Deploy Guardian Agent service - Instrument input/output hooks on critical agents - Set up basic trace collection - Establish initial governance processes ### From Foundational to Managed (3-6 months) - Extend instrumentation to tool calls - Add memory and knowledge hooks - Integrate with SIEM and security operations - Create initial AgBOM for all agents - Establish regular policy review cadence ### From Managed to Optimized (6-12 months) - Add protocol-level controls (MCP, A2A) - Implement dynamic AgBOM updates - Integrate enterprise classifiers - Map controls to compliance requirements - Establish red-team testing program - Engage with ACS/AOS community --- ## Continuous Improvement ### Regular Assessments Schedule maturity assessments: - **Quarterly** for Foundational implementations - **Bi-annually** for Managed implementations - **Annually** for Optimized implementations ### Assessment Focus Areas Each assessment should review: - **Technology coverage** — are all required hooks instrumented? - **Policy effectiveness** — are controls working as intended? - **Integration quality** — is security integration effective? - **Governance processes** — are reviews and testing happening? - **Compliance status** — are regulatory requirements met? ### Improvement Planning Use assessment results to: - **Identify gaps** — missing capabilities or processes - **Prioritize improvements** — focus on highest-impact items - **Plan next steps** — create roadmap for advancement - **Track progress** — measure improvement over time --- ## Key Takeaways 1. **Three maturity levels** — Foundational, Managed, Optimized 2. **Clear criteria** — specific, measurable items at each level 3. **Progressive advancement** — complete each level before moving up 4. **Balanced focus** — technology, process, and compliance 5. **Continuous improvement** — regular assessments and planning 6. **Community engagement** — contribute back at higher maturity levels --- ## Next Steps - **Ecosystem & Roadmap** — understand how ACS is evolving and where it's headed - **Limitations & Critique** — understand the current constraints and considerations