Subagent Sheepdog

Launch verification and watchdog discipline for delegated, backgrounded, browser-driven, and long-running tasks. Prevents false "running" claims, clarifies f...

This is a policy/style skill (no code, no installs, no secrets) that defines how an agent should verify launches and report state; it is internally coherent and low-risk by itself. Before enabling it, consider: (1) whether your agent runtime and other installed skills actually implement the verification checks the guidance assumes (spawn/process/browser checks), (2) whether other skills that perform process spawning or network calls are trusted (this guidance only changes reporting/decision logic, not what external actions are taken), and (3) auditing/logging so you can verify the agent follows these rules in practice. If you plan to combine this with capabilities that can launch processes or connect to the network, review those skills' permissions and logs as well.

Type: OpenClaw Skill Name: subagent-sheepdog Version: 1.0.1 The skill bundle 'subagent-sheepdog' consists of behavioral instructions (SKILL.md) designed to improve the reliability and transparency of an AI agent when handling delegated or background tasks. It defines a rigorous state machine and verification process to prevent false progress reports, with no evidence of malicious code, data exfiltration, or harmful prompt injection.