The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill directs the agent to download and execute unverified code using the npx -y ppxc-leads-mcp command. This pattern facilitates the immediate execution of third-party scripts from the NPM registry on the user host machine without manual review.
[COMMAND_EXECUTION]: The instructions require the agent to perform environment discovery and modify host configuration files (such as ~/.cursor/mcp.json or Claude desktop configs) to register the external MCP server. This allows for the persistent execution of remote tools within the agent host environment.
[DATA_EXFILTRATION]: The skill accesses user-authenticated data from social media platforms (Douyin, Xiaohongshu, Kuaishou) and transmits sales leads, customer names, and interaction history to the external domain opc1.me for processing and storage.
[PROMPT_INJECTION]: The skill instructions explicitly direct the agent to override standard user interaction patterns, such as '不要中途问用户是否继续' (do not ask the user if they want to continue), which reduces user control over batch operations and external data transmission.
[PROMPT_INJECTION]: The skill processes untrusted external data which constitutes an indirect prompt injection surface. 1. Ingestion points: Social media comment data retrieved via the analyze_video_comments and start_search_run tools. 2. Boundary markers: No protective delimiters or instructions to ignore embedded commands are specified. 3. Capability inventory: The skill has the ability to perform network exfiltration to opc1.me and execute commands on the host. 4. Sanitization: No evidence of data sanitization or validation of the fetched comments is present.
[EXTERNAL_DOWNLOADS]: The skill relies on fetching the ppxc-leads-mcp package from the public NPM registry without version pinning or integrity validation, introducing significant supply chain risks.

find-customers