Plan

Auto-learns when to plan vs execute directly. Adapts planning depth to task type. Improves strategy through outcome tracking.

This skill appears coherent and low-risk for its stated purpose. Before installing, confirm two things with the skill author or your admin: (1) where and how outcome/tracking data will be stored (local files, agent workspace, or an external service) and whether those records contain sensitive details you don't want kept; (2) what exactly the 'promotion' / auto-start behavior does and whether you want the agent to be allowed to auto-execute promoted plan types without human confirmation. If you want tighter control, require human validation for promotions or restrict the agent's ability to write persistent records or execute tasks autonomously.

Type: OpenClaw Skill Name: plan Version: 1.0.0 The skill bundle, consisting of `_meta.json`, `SKILL.md`, `outcomes.md`, and `strategies.md`, is entirely focused on instructing the AI agent on how to improve its own planning, execution, and learning processes. The `SKILL.md` file, which is the primary prompt injection surface, contains no instructions for executing commands, accessing sensitive files, making network requests, exfiltrating data, or any other actions that could be considered malicious or even suspicious. All content is self-referential to the agent's internal meta-cognition and process improvement, posing no security risks to the host system or user data.