Agent Governance Toolkit

security、automation、monitoring

来源： manual_curated · Manually curated from the verified project README; no AI draft. · 2026/8/17

{"en":"AGT addresses the layer after a model decides to use a tool: deterministic enforcement and audit before the action reaches the target, rather than another safety prompt. It is relevant once agents touch real data or write-capable tools, but the Public Preview status means integrations should allow for API change.","zh":"AGT 解决的是“模型已经决定要调用工具以后，谁来做确定性拦截和审计”这一层问题，而不是再加一段安全 Prompt。它适合开始把 Agent 接入真实数据和写操作的团队，但当前仍是 Public Preview，接口稳定性要留余量。"}

来源： manual_curated · Manually curated from the verified project README; no AI draft. · 2026/8/17

{"steps":[{"en":"Prepare Python 3.11+ and install the full Python toolkit with `pip install \"agent-governance-toolkit[full]\"`.","zh":"准备 Python 3.11+，执行 `pip install \"agent-governance-toolkit[full]\"` 安装完整 Python 工具包。"},{"en":"Run `agt doctor` to check the installation, then write a YAML policy for the actions that need governance.","zh":"先运行 `agt doctor` 检查安装，再为需要治理的操作编写 YAML 策略。"},{"en":"Wrap tool functions with `govern()` and verify the allow, deny, and approval paths.","zh":"用 `govern()` 包装工具函数，验证允许、拒绝和审批路径是否符合预期。"},{"en":"Before production, add audit, identity, container isolation, SRE, or MCP security components according to the risk profile.","zh":"进入生产前再按风险需要增加审计、身份、容器隔离、SRE 或 MCP 安全组件。"}],"installAt":"local","difficulty":"medium"}

来源： manual_curated · Manually curated from the verified project README; no AI draft. · 2026/8/17

{"en":"Microsoft Agent Governance Toolkit (AGT) places agent tool calls, message sends, and delegations behind deterministic application-level policy checks. In the simplest Python flow, `govern()` wraps a tool function and evaluates a YAML policy before execution, allowing the host to permit, deny, or require approval while recording the decision.\n\nThe project expands into identity and trust, execution sandboxing, audit and compliance, SRE and kill switches, an MCP Security Gateway, and related governance components. SDKs are available for Python, TypeScript, .NET, Rust, and Go, with integrations for agent frameworks and developer surfaces such as Claude Code. The `agt` CLI covers installation checks, policy linting, compliance verification, and security-oriented validation.\n\nThe README explicitly labels the current release as Public Preview and warns that breaking changes may occur before GA. It also states that governance runs at the application middleware layer rather than providing OS-level isolation; separate containers are still recommended for production agent isolation.","zh":"Microsoft Agent Governance Toolkit（AGT）把 Agent 的工具调用、消息发送和委派操作放在确定性的应用层策略检查之后执行。最简单的 Python 用法可以用 `govern()` 包装一个工具函数，再通过 YAML 策略决定允许、拒绝或要求审批，并记录策略判定。\n\n项目进一步提供身份与信任、执行沙箱、审计与合规、SRE/kill switch、MCP Security Gateway 等组件，并有 Python、TypeScript、.NET、Rust 和 Go SDK，以及 Claude Code 等开发工具的接入方式。它也提供 `agt` CLI 用于安装检查、策略校验、合规验证和安全测试。\n\n当前 README 明确标注为 Public Preview，并提醒 GA 前可能有 breaking changes；同时文档说明治理发生在应用中间件层，并不等同于操作系统级隔离。生产环境仍建议将不同 Agent 放入独立容器。"}

来源： manual_curated · Manually curated from the verified project README; no AI draft. · 2026/8/17

v4.1.0

来源： GitHub 官方接口 · latest_release=v4.1.0 · 2026/9/12

MIT

来源： GitHub 官方接口 · license.spdx_id=MIT · 2026/9/12

否

来源： manual_curated · Manually curated from the verified project README; no AI draft. · 2026/8/17

{"en":"A governance toolkit for AI agents that adds policy checks, identity, audit trails, sandboxing, and SRE controls across multiple agent frameworks.","zh":"为 AI Agent 工具调用加入策略检查、身份、审计、沙箱与 SRE 控制的治理工具包，可接入多种 Agent 框架。"}

来源： manual_curated · Manually curated from the verified project README; no AI draft. · 2026/8/17

ai-apps

来源： 项目说明文档 · hint=ai-apps · 2026/8/13

cli、library_framework、ide_plugin

来源： manual_curated · Manually curated from the verified project README; no AI draft. · 2026/8/17

security、backend、devops

来源： manual_curated · Manually curated from the verified project README; no AI draft. · 2026/8/17

是

来源： 仓库文件 · dockerfile=true; compose=true · 2026/9/12

是

来源： manual_curated · Manually curated from the verified project README; no AI draft. · 2026/8/17

是

来源： 项目说明文档 · matched self-host keywords · 2026/8/13