agent-sentinel
بواسطة jimmystacks
AI & LLMsموجز
The operational circuit breaker for this agent.
الوصف
ما الذي تفعله هذه المهارة
AgentSentinel adds a budget enforcement and action safety layer to AI agents. It intercepts potentially destructive or expensive commands before execution, requiring explicit approval when thresholds are exceeded. An optional cloud connection enables real-time dashboards and human-in-the-loop approval workflows.
Unlike simple prompt-level guardrails, AgentSentinel enforces limits at the tool-call level and can escalate to a human approver before irreversible actions complete.
متى تستخدمها
- Blocking an agent from deleting files without confirmation
- Capping API spend per session to avoid runaway costs
- Requiring human approval before an agent executes unknown code
- Monitoring cumulative cost across a long multi-step coding task
- Pausing an automated workflow when a sensitive data transfer is detected
مثال على سير العمل
User asks: clean up all temp files in /tmp
- 1
Agent calls check_safety with cmd='rm -rf /tmp' and estimated cost
- 2
sentinel_wrapper.py evaluates the command against budget and restriction rules
- 3
check_safety returns APPROVAL_REQUIRED for a destructive delete
- 4
Agent calls request_approval with the action and reason
- 5
Human approves or denies via dashboard; agent proceeds only on approval
Files deleted only after explicit human confirmation, with cost logged to the session.
المتطلبات
- AGENT_SENTINEL_API_KEY environment variable (from agentsentinel.dev dashboard, required for cloud features)
التثبيت
clawhub install jimmystacks/agent-sentinelشغّل أمر واجهة سطر الأوامر، أو الصق النص في Claude Code أو Cursor أو Codex أو أي وكيل برمجة تستخدمه.