The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data from LaTeX files or pasted text and uses this to perform file-write operations. This creates a surface for indirect prompt injection attacks where instructions hidden in the data could influence the agent.
Ingestion points: The skill reads user-provided .tex files and pasted text descriptions in the 'Collect Context' step of the workflow.
Boundary markers: None identified. The skill does not define specific delimiters or include instructions to ignore instructions embedded within the processed text.
Capability inventory: The skill uses the Write tool to modify the content of the target .tex file, specifically replacing content within \caption{} commands.
Sanitization: No evidence of input validation, escaping, or filtering of the content read from files or provided by the user is mentioned in the workflow.

caption-skill