小爱舆情AI标注（OpenAI兼容）

Incrementally AI-labels unmarked records in Feishu bitable, adding fields like type, sentiment, competitor mention, platform, brand safety, and content safety.

This package will read records from your Feishu bitable and (per code) requires you to provide APP_ID and APP_SECRET to write labels back — that part aligns with its purpose. However, the runtime wrapper currently enforces providing OPENAI_API_KEY, OPENAI_BASE_URL and OPENAI_MODEL, and will send the record text to whatever model gateway you configure. SKILL.md does not declare these required inputs nor does the declared network permission allow arbitrary LLM hosts. Before installing or running: 1) Treat OPENAI_BASE_URL as a sensitive endpoint — only point it to a trusted, audited gateway you control (or don't set it if you don't want to send data externally). 2) If you expect the stdin/fallback mode, confirm with the author or update label_skill.py — as distributed it forces external LLM usage. 3) Prefer running in a restricted/sandbox environment and inspect/run the code locally with non-production data first. 4) Ensure the Feishu APP_ID/APP_SECRET you supply follow least privilege (scoped tokens) since the skill will use them to read/write the bitable. 5) If you need this skill for internal use but cannot trust external LLM gateways, ask the maintainer to: a) document required env inputs in SKILL.md, b) make stdin fallback actually optional (not failing when openai vars are missing), and c) adjust the network permission list to include any required LLM hosts or explicitly state they are user-controlled.

Type: OpenClaw Skill Name: yuqing-label-skill Version: 1.2.0 The skill bundle is a functional tool designed to perform automated AI labeling on Feishu (Lark) Bitable records. The Python scripts (bitable_labeling_skill.py and label_skill.py) implement standard OAuth authentication with Feishu and interact with OpenAI-compatible APIs to classify data based on sentiment, brand safety, and device types. The system prompts provided in the markdown files are highly specific to the labeling task (focused on Xiaomi's Xiaoai assistant) and do not contain any instructions intended to subvert the agent or exfiltrate sensitive data. All network activity is aligned with the stated purpose of the skill.