The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the source_file (markdown articles). This content is dynamically adapted by specialized agents and subsequently passed to execution tools, creating a surface for indirect instructions to influence the agent's behavior.
Evidence Chain:
Ingestion points: The source_file parameter in SKILL.md allows the input of arbitrary article content.
Boundary markers: No explicit boundary markers or delimiters are defined in the templates to separate untrusted content from system instructions.
Capability inventory: The skill can execute shell commands (wechatsync, biliup, curl) and Python scripts to perform network operations and file modifications.
Sanitization: The skill mentions a 'Sensitive-word preflight' scan, which acts as a basic filter, but it does not prevent logic-based prompt injections.
[COMMAND_EXECUTION]: The skill provides templates for executing various shell commands to manage platform authentication and content synchronization. This includes wechatsync for multiple platforms, biliup for video uploads, and curl commands to interact with a local xiaohongshu-mcp daemon.
[EXTERNAL_DOWNLOADS]: The instructions direct the agent to utilize third-party tools and repositories that are not managed by trusted organizations. Specifically, it references xpzouying/xiaohongshu-mcp for Xiaohongshu publishing and the Nemo2011/bilibili-api Python library for Bilibili automation.

marketing-multi-platform-publisher