The Agent Skills Directory

[COMMAND_EXECUTION]: The toolkit executes shell commands via execSync to render diagrams. Specifically, toolkit/src/mermaid-processor.ts invokes the mmdc (Mermaid CLI) utility to convert diagram code into PNG images for publication. It also uses execSync for environment checks (checking tool versions).
[INDIRECT_PROMPT_INJECTION]: The skill has an indirect injection surface due to its data ingestion pipeline.
Ingestion points: scripts/fetch_hotspots.py fetches trending topics from Weibo, Toutiao, and Baidu search suggestions. toolkit/src/youmind-api.ts retrieves content from the user's YouMind knowledge base boards.
Boundary markers: The skill lacks explicit boundary markers or instructions to ignore embedded commands within the fetched hotspot data.
Capability inventory: The skill possesses capabilities including createDraft (publishing content to a remote platform) and execSync (operating system command execution).
Sanitization: While the LLM processes this data during the topic selection and writing steps, there is no deterministic sanitization of the input strings before they enter the agent context.
[EXTERNAL_DOWNLOADS]: The skill makes several network requests to well-known services and the vendor's infrastructure.
Fetches real-time hotspot data from weibo.com, toutiao.com, and baidu.com.
Downloads predefined cover assets from the vendor's CDN (marketing-assets.youmind.com).
Communicates with the YouMind OpenAPI for knowledge retrieval and image generation tasks.
[AUTONOMY_CONCERN]: The instructions in SKILL.md contain directives that reduce user oversight, such as Critical Quality Rule #11, which mandates automatic publishing to WeChat drafts without asking the user for confirmation.

youmind-wechat-article