|
| 1 | +# Agent Governance Toolkit v1.0.0 |
| 2 | + |
| 3 | +**The missing security layer for AI agents.** Runtime policy enforcement, identity mesh, execution sandboxing, and reliability engineering — in one toolkit. |
| 4 | + |
| 5 | +## Highlights |
| 6 | + |
| 7 | +- 🛡️ **10/10 OWASP Agentic Top 10 coverage** — full compliance mapping across all ASI-01 through ASI-10 risks |
| 8 | +- 🔐 **Microsoft Entra Agent ID integration** — bridge DID identity with enterprise Zero Trust via Entra Agent ID |
| 9 | +- 📦 **AI-BOM v2.0** — full AI supply chain tracking: model provenance, dataset lineage, weights versioning |
| 10 | +- 🏛️ **CSA Agentic Trust Framework** — compliance mapping across all 5 ATF pillars (15/15 requirements) |
| 11 | +- ✅ **OpenSSF Scorecard hardened** — pinned dependencies, CodeQL SAST, Dependabot, signed workflows |
| 12 | + |
| 13 | +## Packages |
| 14 | + |
| 15 | +| Package | Description | Install | |
| 16 | +|---------|-------------|---------| |
| 17 | +| **Agent OS** | Stateless governance kernel with policy engine, VFS, and MCP proxy | `pip install agent-os-kernel` | |
| 18 | +| **AgentMesh** | Zero-trust identity mesh with DID, trust scoring, delegation chains | `pip install agentmesh-platform` | |
| 19 | +| **Agent Hypervisor** | Execution rings, resource limits, kill switch, saga orchestration | `pip install agent-hypervisor` | |
| 20 | +| **Agent SRE** | SLOs, error budgets, circuit breakers, chaos engineering | `pip install agent-sre` | |
| 21 | +| **Agent Compliance** | Unified installer and compliance documentation | `pip install ai-agent-compliance` | |
| 22 | + |
| 23 | +## Security & Compliance |
| 24 | + |
| 25 | +| Framework | Coverage | |
| 26 | +|-----------|----------| |
| 27 | +| [OWASP Agentic Top 10 (2026)](docs/OWASP-COMPLIANCE.md) | 10/10 risks covered | |
| 28 | +| [CSA Agentic Trust Framework](docs/compliance/csa-atf-mapping.md) | 15/15 requirements | |
| 29 | +| [NIST AI RMF](https://www.nist.gov/artificial-intelligence/ai-risk-management-framework) | Govern, Map, Measure, Manage | |
| 30 | +| [Singapore MGF for Agentic AI](docs/analyst/singapore-mgf-mapping.md) | Zero-trust, accountability, oversight | |
| 31 | +| [EU AI Act](https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai) | Risk classification, audit trails, human oversight | |
| 32 | + |
| 33 | +## Key Features in v1.0.0 |
| 34 | + |
| 35 | +### Identity & Trust |
| 36 | +- Cryptographic DID identity (`did:mesh:`) with Ed25519 key pairs |
| 37 | +- Microsoft Entra Agent ID adapter (sponsor accountability, lifecycle management, Conditional Access) |
| 38 | +- Trust scoring with decay, delegation chains with capability narrowing |
| 39 | +- SPIFFE/SVID workload identity support |
| 40 | + |
| 41 | +### Governance |
| 42 | +- Policy-as-code engine (strict/permissive/audit modes) |
| 43 | +- MCP Governance Proxy for tool call interception |
| 44 | +- Approval workflows with quorum logic and expiration |
| 45 | +- Prompt injection detection and PII protection |
| 46 | + |
| 47 | +### Runtime Security |
| 48 | +- Execution rings (Ring 0–3) with graduated privilege |
| 49 | +- Kill switch for instant agent termination |
| 50 | +- Saga orchestration with automatic rollback |
| 51 | +- Joint liability scoring (Shapley values) |
| 52 | + |
| 53 | +### Reliability (SRE) |
| 54 | +- Agent-specific SLOs (correctness, safety, latency, cost) |
| 55 | +- Circuit breakers with cascading failure detection |
| 56 | +- Chaos engineering framework for AI agents |
| 57 | +- Cost anomaly detection with per-agent budgets |
| 58 | + |
| 59 | +### Supply Chain |
| 60 | +- AI-BOM v2.0 — model provenance, dataset tracking, weights versioning |
| 61 | +- SLSA-compatible build provenance for model artifacts |
| 62 | +- CycloneDX ML-BOM export support |
| 63 | + |
| 64 | +## External Submissions |
| 65 | + |
| 66 | +45 integration proposals submitted across the ecosystem: |
| 67 | +- **Merged:** GitHub Copilot (×3), Dify (×1) |
| 68 | +- **Under Review:** Microsoft Agent Framework, Google ADK, AutoGen, CrewAI, LangChain, OpenAI Swarm, MetaGPT, Anthropic, MCP, OpenLit, OWASP, LF AI, CoSAI, AAIF |
| 69 | + |
| 70 | +See [docs/PROPOSALS-INDEX.md](docs/PROPOSALS-INDEX.md) for the full list. |
| 71 | + |
| 72 | +## Quick Start |
| 73 | + |
| 74 | +```bash |
| 75 | +pip install ai-agent-compliance[full] |
| 76 | +``` |
| 77 | + |
| 78 | +```python |
| 79 | +from agent_os import StatelessKernel, ExecutionContext |
| 80 | + |
| 81 | +kernel = StatelessKernel() |
| 82 | +ctx = ExecutionContext(agent_id="my-agent", policies=["read_only"]) |
| 83 | +result = await kernel.execute(action="query_db", params={"table": "users"}, context=ctx) |
| 84 | +``` |
| 85 | + |
| 86 | +## License |
| 87 | + |
| 88 | +[MIT](LICENSE) — © Microsoft Corporation |
0 commit comments