Self-Governor

LLM 通用内部自裁决技能。在关键节点判断"当前这一层最优的下一步动作是什么"，再让主链继续执行。触发条件：(1) 路径分叉时——多个可行方案且无明显优先级；(2) 高代价动作前——搜索/生成/发布等消耗军费或不可逆操作；(3) 连续两步无明显增益时——进展停滞、输出质量未提升。禁止：改写主任务、输出并列动作、长...

This skill appears coherent and low-risk: it only implements a small decision filter and asks nothing extra. Before installing, verify provenance (no homepage or identifiable author), test it in a safe/staging agent to ensure the host correctly enforces the required Update callback (to avoid loop triggers), and confirm you trust the agent's orchestration logic to enforce the skill's 'must not' rules. If you allow autonomous invocation, monitor logs initially to ensure the skill doesn't get inserted in inappropriate places. If provenance or publisher identity matters to you, ask the publisher for more info or prefer a skill with an auditable source.

Type: OpenClaw Skill Name: self-governor Version: 1.0.0 The 'self-governor' skill is a meta-cognitive framework designed to improve LLM agent decision-making at critical junctures, such as path branching or task stagnation. It defines a structured input/output loop and a set of logical actions (e.g., anchor, search, synthesize, degrade_continue) to ensure the agent remains focused on its goals efficiently. The documentation (SKILL.md and references/) contains no evidence of malicious intent, data exfiltration, or harmful prompt injection; instead, it includes explicit constraints to prevent the agent from hijacking tasks or creating bureaucratic delays.