Autonomously executes research, maintenance, security, reflection, knowledge, and optimization tasks during idle time.

This skill is vague about what 'exploration' entails and gives the agent broad, unspecific permission to do system cleanup, security review, and optimizations during idle time. Before installing: 1) Demand a precise, written policy of allowed actions (which directories are in-scope/out-of-scope, whether it may run shell commands, whether it may modify or delete files). 2) Require explicit user confirmation for any destructive or network-capable actions. 3) Prefer a version that limits operations to read-only analysis unless the user opts in to write/delete operations. 4) Ask for logging/audit hooks and a way to revoke or pause the skill. 5) If you cannot get specifics, avoid enabling autonomous invocation or do not install the skill — leave it as user-invoked only and test in a sandbox. Additional information that would raise confidence to 'benign': a detailed SKILL.md enumerating permitted commands, file path restrictions, network endpoints (if any), and explicit safeguards (confirmation prompts, dry-run modes, and logging).

Type: OpenClaw Skill Name: exploration-mode-skill Version: 1.0.0 The skill bundle defines an 'Exploration Mode' in SKILL.md that instructs the AI agent to perform 'autonomous task execution' during idle time, including 'system cleanup' and 'system optimization.' This represents a high-risk capability by encouraging the agent to operate without human-in-the-loop oversight or specific task constraints. While the bundle lacks executable code or explicit malicious intent, the broad permission for autonomous system-level actions is a significant security risk in an agentic framework.