The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external Quarto (.qmd) and Beamer (.tex) files provided by the user via arguments.
Ingestion points: The skill uses Read and Grep tools to load the content of the slide files into the agent's context for analysis (Step 1).
Boundary markers: The instructions lack explicit boundary markers or delimiters to separate the untrusted slide content from the agent's auditing instructions, and there are no warnings to the agent to ignore instructions embedded in the file content.
Capability inventory: The skill allows the agent to use Agent, Task, and Write tools, which provides a significant capability surface if the agent obeys instructions hidden within the audited decks.
Sanitization: There is no mention of sanitizing, escaping, or validating the slide content before the agent analyzes it or compiles a report.
[COMMAND_EXECUTION]: The skill involves the execution of external tools like quarto render and LaTeX compilers (Steps 2 and 3). While these are core to the skill's purpose of visual auditing, executing these tools on untrusted files represents a local execution surface.

visual-audit