The skill uses prompt injection to execute arbitrary commands like
Claims to do
lark-cli 共享规则: 本技能指导你如何通过lark-cli操作飞书资源, 以及有哪些注意事项。
Actually does
The skill executes `lark-cli` commands for configuration initialization (`config init`), user authentication (`auth login` with `--domain` or `--scope`), and identity management (`--as user/bot`). It also provides commands to update the `lark-cli` tool and its associated skill using `npm update` and `npx skills add`. The skill guides on handling permission errors by directing users to Feishu developer console URLs or suggesting `auth login` commands.
npx skills add https://github.com/larksuite/cli --skill lark-sharedDeBERTa classifier detected prompt injection (confidence: 1.00)
The skill instructs the AI to execute `npm update -g` and `npx skills add` commands. This allows for arbitrary code execution, modification of the agent's environment, and introduces a supply chain risk if the package registries or skill repositories are compromised.
npm update -g @larksuite/cli && npx skills add larksuite/cli -g -y
The skill instructs the agent to "proactively suggest to the user to update" when a new version is detected. This could be leveraged in a social engineering attack to trick a human overseer into approving a malicious update, especially when combined with the critical command execution vulnerability.
主动提议帮用户更新
[](https://mondoo.com/ai-agent-security/skills/github/larksuite/cli/lark-shared)<a href="https://mondoo.com/ai-agent-security/skills/github/larksuite/cli/lark-shared"><img src="https://mondoo.com/ai-agent-security/api/badge/github/larksuite/cli/lark-shared.svg" alt="Mondoo Skill Check" /></a>https://mondoo.com/ai-agent-security/api/badge/github/larksuite/cli/lark-shared.svgSkills can read files, run commands, and access credentials. Mondoo helps organizations manage the security risks of AI agent skills across their entire fleet.