/projects/runbookguard
Runbook Guard
面向运营自动化 runbook 的安全层
/projects/runbookguard
面向运营自动化 runbook 的安全层
Runbook Guard 处理的核心挑战,是把面向运营自动化 runbook 的安全层变成有用、可治理且可解释的产品系统。
Runbook Guard 围绕明确边界、可检查工作流和可见交付技术栈组织,其中包括 Python、FastAPI、Typer和Pydantic。
难点在于让自主性、数据、工具和用户体验保持一致,同时不隐藏运营风险或证据。
关键创新是把架构当成产品界面:能力通过责任、证明路径和交付边界来说明。
Runbook Guard 展示了产品架构、智能体系统、数据边界、控制机制和交付纪律上的应用工程判断。
Runbook Guard 是关于面向运营自动化 runbook 的安全层的案例研究。面向运营自动化 runbook 的安全层 其深度在于让边界、证据、工作流和交付选择变得可读。
Runbook Guard 通过清晰职责来组织,避免用户、工具、数据和运行时动作混成一次不透明的模型调用。
该流程将规划、工具使用、状态和结果呈现区分开,让操作者能理解发生了什么以及为什么发生。
页面把证明、日志、产物和评审路径视为产品的一部分,而不是可选的实现备注。
Python 支撑实现叙事,周边技术栈承载持久化、界面、自动化和验证。
技术理由: Keeps the control plane approachable for SRE and platform automation scripts.
技术理由: Provides typed HTTP endpoints for plan, status, execute, health, and readiness.
技术理由: Gives operators a CLI that mirrors the API without duplicating command parsing.
技术理由: Makes runbook requests, plans, receipts, approvals, and audit bundles explicit contracts.
技术理由: Stores workflow state locally with almost no operational overhead.
技术理由: Packages the guardrail service for repeatable local and deployment checks.
技术理由: Keeps Python dependency resolution and command startup fast for operator workflows.
已实现工具: Accepts plan, execute, status, health, and readiness calls.
已实现工具: Runs validate, plan, execute, audit-list, and audit-show from the terminal.
已实现工具: Rejects malformed requests before they enter the state machine.
已实现工具: Pulls read-only evidence before any write action can be planned.
已实现工具: Applies allowlists, denylists, risk thresholds, and budget limits.
已实现工具: Holds higher-risk plans until a named human approves them.
已实现工具: Keep mutations behind a separate seam and emit receipts by default.
已实现工具: Confirm the intended outcome before the run is marked verified.
已实现工具: Seals request, evidence, plan, approvals, receipts, and verification into a reviewable record.