Agent Governance Toolkit v1.0.0¶

[!IMPORTANT] Community Preview Release — This release is for testing and evaluation purposes only. Packages published to PyPI are not official Microsoft-signed releases. Official Microsoft-signed packages via ESRP Release will be available in a future release.

The missing security layer for AI agents. Runtime policy enforcement, identity mesh, execution sandboxing, and reliability engineering — in one toolkit.

Highlights¶

🛡️ 10/10 OWASP Agentic Top 10 coverage — full compliance mapping across all ASI-01 through ASI-10 risks
🔐 Microsoft Entra Agent ID integration — bridge DID identity with enterprise Zero Trust via Entra Agent ID
📦 AI-BOM v2.0 — full AI supply chain tracking: model provenance, dataset lineage, weights versioning
🏛️ CSA Agentic Trust Framework — compliance mapping across all 5 ATF pillars (15/15 requirements)
✅ OpenSSF Scorecard hardened — pinned dependencies, CodeQL SAST, Dependabot, signed workflows

Packages¶

Package	Description	Install
Agent OS	Stateless governance kernel with policy engine, VFS, and MCP proxy	`pip install agent-os-kernel`
AgentMesh	Zero-trust identity mesh with DID, trust scoring, delegation chains	`pip install agentmesh-platform`
Agent Runtime	Execution rings, resource limits, kill switch, saga orchestration	`pip install agentmesh-runtime`
Agent SRE	SLOs, error budgets, circuit breakers, chaos engineering	`pip install agent-sre`
Agent Compliance	Unified installer and runtime policy enforcement	`pip install agent-governance`
Agent Marketplace	Plugin lifecycle management for governed agent ecosystems	`pip install agentmesh-marketplace`
Agent Lightning	RL training governance with governed runners and policy rewards	`pip install agentmesh-lightning`

Security & Compliance¶

Framework	Coverage
OWASP Agentic Top 10 (2026)	Architecture mapping
CSA Agentic Trust Framework	Conformance assessment
NIST AI RMF	Govern, Map, Measure, Manage
EU AI Act	Risk classification, audit trails, human oversight

Key Features in v1.0.0¶

Identity & Trust¶

Cryptographic DID identity (did:mesh:) with Ed25519 key pairs
Microsoft Entra Agent ID adapter (sponsor accountability, lifecycle management, Conditional Access)
Trust scoring with decay, delegation chains with capability narrowing
SPIFFE/SVID workload identity support

Governance¶

Policy-as-code engine (strict/permissive/audit modes)
MCP Governance Proxy for tool call interception
Approval workflows with quorum logic and expiration
Prompt injection detection and PII protection

Runtime Security¶

Execution rings (Ring 0–3) with graduated privilege
Kill switch for instant agent termination
Saga orchestration with automatic rollback
Joint liability scoring (Shapley values)

Reliability (SRE)¶

Agent-specific SLOs (correctness, safety, latency, cost)
Circuit breakers with cascading failure detection
Chaos engineering framework for AI agents
Cost anomaly detection with per-agent budgets

Supply Chain¶

AI-BOM v2.0 — model provenance, dataset tracking, weights versioning
SLSA-compatible build provenance for model artifacts
CycloneDX ML-BOM export support

External Submissions¶

45 integration proposals submitted across the ecosystem: - Merged: GitHub Copilot (×3), Dify (×1) - Under Review: Microsoft Agent Framework, Google ADK, AutoGen, CrewAI, LangChain, OpenAI Swarm, MetaGPT, Anthropic, MCP, OpenLit, OWASP, LF AI, CoSAI, AAIF

See docs/proposals/ for the full list.

Quick Start¶

pip install agent-governance[full]

from agent_os import StatelessKernel, ExecutionContext

kernel = StatelessKernel()
ctx = ExecutionContext(agent_id="my-agent", policies=["read_only"])
result = await kernel.execute(action="query_db", params={"table": "users"}, context=ctx)