Policy plugins for OpenClaw agent gateways. Each plugin is a standalone, dependency-light package under its own directory.
Minimal, dependency-free policy plugin. Requires human approval for selected tools (e.g. web_search, web_fetch, browser) and for exec calls whose working directory or referenced paths fall outside configured workspace roots. Renders a structured approval card (action, target, agent's reasoning, reason, risk) since the host may not display raw tool parameters.
TypeScript policy plugin that enforces a bounded number of tool-call attempts per agent run. Once the budget is exhausted, further tool calls are blocked and the agent must produce a user-visible checkpoint (result, blockers, revised plan) before the run ends.