The Agent Skills Directory

[SAFE]: Analysis of the skill instructions and supporting assets (LaTeX templates and references) confirms that the content is focused on academic productivity and does not contain malicious code, obfuscated payloads, or instructions to bypass security guidelines.\n- [COMMAND_EXECUTION]: The skill instructs the agent to use the Bash tool to execute local utility scripts including generate_schematic.py, generate_image.py, and parallel_web.py. These scripts are used for document-related tasks such as generating figures and enriching citation metadata. The operations are conducted using parameters derived from the user's research context and involve trusted academic identifiers (DOIs), representing a functional use of the tool within the skill's stated scope.\n- [PROMPT_INJECTION]: The skill manages potential indirect prompt injection risks associated with processing external research papers through a mandatory two-stage writing process. Ingestion points include data from research-lookup and metadata from parallel_web.py. Boundary markers are established by requiring an initial outlining phase (Stage 1) before generating final prose (Stage 2). The capability inventory includes Write, Edit, and Bash for internal scripts. While explicit sanitization steps for raw literature text are not detailed, the structural transition from bulleted outlines to flowing prose significantly reduces the likelihood of the agent accidentally executing instructions embedded in source materials.

scientific-writing