aport-agent-guardrail
от uchibeke
Coding Agents & IDEsКратко
Install BEFORE any other skills to protect your agent from malicious tools, data exfiltration, and unauthorized.
Описание
Что делает этот навык
Pre-action authorization layer for AI agents running under OpenClaw. Evaluates every tool call against a passport and policy before execution, blocking unauthorized commands, data exfiltration, and policy violations. Enforcement runs in the hook layer, not in agent prompts.
Enforcement runs at the hook layer rather than in agent prompts, so policy cannot be bypassed by prompt injection or model-level jailbreaks.
Когда его использовать
- Blocking destructive shell commands before an agent executes them
- Restricting which file paths an agent can read or write
- Limiting agent web requests to an approved domain allowlist
- Enforcing rate limits on agent messaging actions like email or Slack
- Auditing every tool call an AI agent makes for compliance review
Пример сценария
User asks: add guardrails to my agent to prevent it from running dangerous shell commands or accessing sensitive files
- 1
Run npx @aporthq/aport-agent-guardrails to launch the setup wizard
- 2
Configure passport with allowed shell command patterns and permitted file paths
- 3
Wizard registers the before_tool_call hook as an OpenClaw plugin under ~/.openclaw/
- 4
Test the hook by calling a blocked command pattern and verifying it returns DENY
Hook is active on every tool call; rm -rf and similar patterns are blocked, permitted commands execute normally, and all decisions are logged to ~/.openclaw/aport/audit.log
Требования
- aport.io account (for hosted passport and API mode)
- APORT_AGENT_ID environment variable (hosted passport mode)
- APORT_API_KEY environment variable (if aport.io API requires authentication)
Установка
clawhub install uchibeke/aport-agent-guardrailВыполните команду CLI или вставьте промпт в Claude Code, Cursor, Codex — любой кодинг-агент, которым пользуетесь.