Validate Agent

Agent验证与质量保证。减少静默系统错误，产出可重现的证据。 触发条件：(1) 完成任务时 (2) 修复问题后 (3) 提供答案前 (4) 跨会话承诺时

This skill is an instruction-only agent QA checklist and looks coherent. Before installing: (1) confirm you are comfortable with the agent writing/reading a local file named promises.md (cross-session storage and expiry behavior), and ensure the agent has appropriate filesystem permissions and the intended directory for that file; (2) review and tune thresholds (4s, 30s validation, 3 proactive messages/day) to match your environment; (3) if you need strict auditing, modify the skill to log to a controlled location or declare the config path explicitly; (4) no credentials or network calls are requested, so network/exfiltration risk from the skill itself is low. If you want higher assurance, request the author add an explicit config-path requirement for promises.md and clarify where logs are kept and how expiration is enforced.

Type: OpenClaw Skill Name: validate-agent Version: 1.0.0 The skill bundle consists of metadata and instructional guidelines (SKILL.md) aimed at improving agent reliability and quality assurance. It provides procedural advice for the AI agent, such as performing single-variable modifications and tracking commitments in a 'promises.md' file, but contains no executable code, network requests, or instructions that would lead to unauthorized data access or system compromise.