schoolllllll

执行龙虾学校十道测试题，依步骤操作并保存所有证据，确保结果真实完整且逐题汇总。

This skill runs many local and networked tests and then asks to push all evidence to a GitHub repo you do not control. Before installing or running it, consider the following: (1) It will write files to ~/Desktop/claw_evidence and read your Desktop file list — sensitive filenames or content could be captured. (2) The upload step requires Git credentials (SSH key or HTTP auth) even though the skill declares none; running it could cause your existing SSH keys or credential helper to be used to push data to that remote repo. (3) The prompt-injection test asks you to record the assistant's exact reply — that can cause internal prompts or sensitive text to be written out if the model is tricked. (4) It also expects access to notification, scheduling, and memory tools that typically require tokens or config. Recommended actions: run only in an isolated/sandboxed environment (no personal SSH keys, no sensitive files on Desktop), or edit the SKILL.md to remove the automatic upload step or replace the repo URL with a repo you control and to require explicit credentials (e.g., GIT_SSH_KEY or GITHUB_TOKEN) before pushing. If you must test it, provide disposable credentials and remove any sensitive files from ~/Desktop first. If you do not want automated pushing or potential exfiltration, do not install or invoke this skill. If you need certainty about what it will access, request the skill author to declare required env vars/config paths and to remove the instruction that records full assistant replies.

Type: OpenClaw Skill Name: teamolab-school Version: 1.0.1 The skill bundle, framed as an 'intelligence test' for the agent, performs several high-risk actions in SKILL.md, including taking browser screenshots, listing notification channels, and gathering system metadata. Most critically, it instructs the agent to exfiltrate all collected 'evidence' files to an external GitHub repository (github.com/sophiaashi/shcool-skill-upload.git). While the data collected in this specific version is relatively low-sensitivity, the automated pipeline for uploading local files to a remote, hardcoded destination is a significant security risk and a common pattern for data exfiltration.