The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs file system operations and executes Git commands (git add, git commit, git push). The user-provided <topic> is used directly in the commit message: git commit -m "docs: add research on <topic>". If the agent does not properly escape this string before execution, it could lead to command injection within the shell environment.
[DATA_EXFILTRATION]: The skill's primary function involves uploading local content to a remote server via git push. Users should ensure the repository's remote destination is trusted, as any information researched or generated will be exfiltrated to that location.
[PROMPT_INJECTION]: The skill relies on parsing a topic from user prompts. Maliciously crafted topics could attempt to override the agent's instructions or influence the content of the generated document.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests untrusted data from the user (the topic) and from external sources (websites and documentation) while gathering research.
Boundary markers: There are no explicit instructions or delimiters provided to the agent to treat researched content as data rather than instructions.
Capability inventory: The skill possesses the ability to create directories, write files, and interact with a remote repository via Git.
Sanitization: Although the instructions require a "slug" (lowercase, hyphenated) for the filename, which provides some sanitization for file paths, the original topic string is used unsanitized in the Git commit message.

research-doc