Selection Agent

全渠道选品 Agent — 关键词调研、SEO分析、竞品分析、TrendPlus数据。触发词：选品、关键词调研、SEO分析、竞品分析、keyword research、competitor analysis。执行前必须读取 AGENT_CONFIG.md。

Do not enable this skill without verification. The SKILL.md instructs the agent to read /tmp/powerful-trendplus/AGENT_CONFIG.md, source .env.local, and run a local Python script — actions that can expose local secrets or execute arbitrary code. Before installing, ask the publisher to: (1) provide a trusted source URL and SHA256 for scripts; (2) update metadata to list all required env vars and explain why each is needed; (3) publish the AGENT_CONFIG.md and run_full_research.py for review. If you must test, do so in an isolated sandbox or VM, and avoid granting production credentials (move keys to a test account). If you do install, inspect /tmp/powerful-trendplus and .env.local contents and remove or redact any secrets you don't want the agent to access.

Type: OpenClaw Skill Name: selection-agent Version: 1.0.0 The skill bundle contains highly risky instructions in SKILL.md that direct the AI agent to execute a Python script from a temporary directory (/tmp/powerful-trendplus/scripts/run_full_research.py) and source local environment variables (.env.local). It explicitly mentions that sensitive tokens and API keys (Semrush, FB_ADS_TOKEN, NOTION_API_KEY, GEMINI_API_KEY) are stored in these local files. While these behaviors facilitate arbitrary code execution and potential credential access, the lack of the actual script payload or explicit exfiltration logic makes it impossible to confirm malicious intent versus extremely poor security practices for a research tool.