360 Secure SkillHub
360 安全技能中心

Sanna Governance

该 Skill 启用 Sanna 治理层,按 YAML 宪法自动拦截并评估每次工具调用,给出放行、阻断或升级人工审批三种结果,并对受治理调用自动生成 Ed25519 签名收据,用于证明执行时治理已生效。

nicallen-exd
nicallen-exd
热度 34.60
最近更新 2026-05-11 17:41:28
分类:安全依赖:纯提示词应用场景:工作流 / 报告撰写

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add sanna-governance
Metadata
name
sanna
description
Sanna governance — tool calls are governed transparently
metadata
{"openclaw":{"tags":["governance","security","constitution","receipts","compliance","audit","ed25519","cryptographic-proof","tool-gating","enforcement","guardrails","ai-safety","agent-safety","approval","escalation","policy","permissions","trust","verification","openclaw-plugin"],"summary":"Constitution enforcement and Ed25519-signed cryptographic receipts for every AI agent tool call. Define what actions agents can and cannot take in YAML constitutions. Three possible outcomes per tool call: allowed, blocked, or escalated for human approval. Every governed action produces a verifiable receipt proving governance was applied at the moment of execution.","requiresPlugin":"@sanna-ai/openclaw","requires":{"bins":["node"]}}}

Sanna Governance

This system has Sanna governance enabled. A constitution defines what actions are allowed, denied, or require escalation. Governance is applied automatically to every tool call — no special tool names or prefixes are needed.

How It Works

Call tools normally. The governance layer intercepts each call and evaluates it against the constitution before execution. There are three possible outcomes:

  • Allowed — the tool executes normally
  • Blocked — the tool is denied with an explanation of which rule was violated
  • Escalated — the tool requires human approval before it can proceed

Governed Tool Tiers

Tier Tools Risk Level
1 exec, bash, write, edit, apply_patch, process Modifies system state
2 browser, message, nodes Composite tools with high-risk actions
3 web_search, web_fetch, cron, gateway, sessions_send, sessions_spawn Audit trail

Tier 4 tools (read, image, canvas, sessions_list, sessions_history, session_status, memory_search, memory_get, agents_list) are not governed.

Receipts

Every governed action generates a cryptographic receipt that proves governance was applied. Receipts are generated and persisted automatically — no action is needed from the agent.