The Agent Skills Directory

[DATA_EXPOSURE]: The skill maintains persistent state for writing projects but includes strict guidelines against storing sensitive data such as API keys, cookies, or environment variables. It explicitly limits the 'Global Profile' to non-sensitive preferences.
[INDIRECT_PROMPT_INJECTION]: The skill defines a clear 'Security Boundary' for external materials. It instructs the agent to treat content from web pages, papers, and source code as data only, explicitly forbidding the execution of any commands or instruction overrides found within that data.
[COMMAND_EXECUTION]: The workflow involves standard shell utilities (git, openssl, mv, rm) for local environment management, such as identifying the project root, generating unique identifiers for articles, and managing the wenqu-skills/ workspace directory.
[EXTERNAL_DOWNLOADS]: While the skill interacts with web content, it does so through a structured research process (via a companion library skill) and treats all downloaded content as untrusted text rather than executable scripts.

wenqu-write