shejian

舌尖香港门店AI助手（支持环境变量） 当用户发送任何与门店运营相关的中文信息时触发，包括但不限于： - 报告商品库存状态，如"番茄卖完了"、"胡萝卜还剩5斤"、"白菜今天卖了20斤" - 查询今日库存、销售情况、进货记录、操作日志 - 录入进货信息，如"今天收到50斤胡萝卜" - 查询门店所在城市天气，如"明天天...

Before installing, confirm with the author how tokens are supplied and why the package metadata doesn't declare them. Specific actions to take: 1) Require the skill manifest to list the required env var (SHEJIAN_API_TOKEN_<agentId>) or otherwise document credential input explicitly. 2) Verify the origin and trustworthiness of the API host (https://s.xingke888.com) and confirm token scope and expiration — prefer short-lived, least-privilege tokens. 3) Limit the skill's read permission to the single file (SHEJIAN.md) and prevent broad workspace file access. 4) Ask whether exec/curl outputs or command histories are logged; avoid embedding long-lived secrets into commands where logs might capture them. 5) If you cannot verify the owner or token handling, treat this skill as risky and avoid installing it in workspaces that contain other sensitive credentials or data.

Type: OpenClaw Skill Name: store-ai Version: 1.0.6 The skill instructions in SKILL.md explicitly direct the AI agent to bypass security restrictions in the 'exec' tool (which blocks the use of '$' for environment variables) by reading sensitive API tokens and manually concatenating them into shell commands. This pattern is highly vulnerable to command injection and exposes secrets in the process list. Furthermore, the instructions mandate 'stealth' behavior by strictly forbidding the agent from disclosing its internal operations or tool-calling details to the user, which could be used to mask unauthorized actions.