Plant Wilting Monitoring Skill | 植物枯萎监测技能

What to consider before installing/using this skill: - Mismatch between policy and code: SKILL.md forbids reading local 'memory' files and LanceDB, but the code will read/write YAML config files and can create/use a local SQLite DB under the workspace data directory. If you rely on that prohibition for privacy, do not trust it until code is modified and audited. - Undeclared environment variables: The code reads OPENCLAW_WORKSPACE, OPENCLAW_SENDER_OPEN_ID, OPENCLAW_SENDER_USERNAME, and FEISHU_OPEN_ID (and the ApiEnum/config may use API keys and base URLs). These are not listed in the skill's declared requirements. Confirm where open-id comes from and whether your environment will expose sensitive IDs to the skill. - External endpoints and data upload: By default config files point to lifeemergence/open API hosts. The skill uploads files or tells a remote API to download a URL. Verify the API endpoints (and privacy policy) before sending images/videos; do not upload sensitive images to an untrusted backend. - Large shared library and unrelated code: The package contains a full face_analysis skill and a heavy smyx_common dependency tree. That increases attack surface and may include functionality unrelated to plant monitoring. Inspect skills/smyx_common/scripts/util.py and api_service.RequestUtil to confirm exactly what is sent to remote servers (headers, tokens, full payload). - Persistence: The skill may create local database and config files under the workspace; if you need zero local persistence, run the skill in an isolated environment or sandbox. Actionable steps: 1) Inspect skills/smyx_common/scripts/util.py (RequestUtil) to see endpoints, headers, and whether it includes any credential exfiltration or telemetry. Confirm where files are uploaded and exactly which fields (open-id, filenames, etc.) are sent. 2) If you will use historical-report features, ask the publisher to document how open-id is used, how long reports are retained, and to provide a privacy statement for the remote API. 3) Run the skill in an isolated test workspace (with a throwaway OPENCLAW_WORKSPACE and without providing real credentials) to observe network calls before using with production data. 4) If you cannot audit the RequestUtil and API behavior, avoid uploading sensitive images or providing real identifiers (open-id) to the skill. What would change this assessment: seeing a trimmed package that only contains the plant-wilting scripts (no unrelated face_analysis), an explicit declaration of env vars and the API endpoints, or a clear README describing exactly what RequestUtil transmits and how data is stored/retained would raise confidence. Conversely, if RequestUtil sends environment contents or arbitrary local files to remote endpoints, the verdict would escalate.