The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process and refine untrusted user-provided content, such as research grant drafts and preliminary data, which creates an attack surface for indirect prompt injection.
Ingestion points: User research drafts, data tables, and solicitation text provided in the chat environment are ingested as input (referenced in the Agent Operating Procedure in SKILL.md).
Boundary markers: There are no explicit markers or instructions defining boundaries between untrusted user data and the agent's internal instructions.
Capability inventory: The skill utilizes Bash (shell access), Write, and Edit tools as specified in the SKILL.md configuration.
Sanitization: The instructions do not specify sanitization or escaping of user-provided content before it is processed or potentially used in generated shell commands.
[COMMAND_EXECUTION]: The skill provides templates and instructions for the agent to execute shell commands using the Bash tool, specifically for invoking an external Python script (generate_schematic.py) to generate project figures. If the agent interpolates user-provided descriptions into these command strings without rigorous validation, it could lead to command injection.

research-grants