The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external content, specifically peer reviewer comments and academic papers, which introduces a potential surface for indirect prompt injection attacks. If these external inputs contain hidden instructions, they could influence the agent's behavior.
Ingestion points: The agent accepts peer reviewer comments and manuscript text as primary inputs as described in the 'INPUTS NEEDED' section of SKILL.md.
Boundary markers: There are no specific instructions to the agent to treat input text strictly as data or to use boundary markers to separate user instructions from external data.
Capability inventory: The skill is configured to use Write and Edit tools, which could potentially be misused to modify files if an injection is successful.
Sanitization: The instructions do not specify any sanitization, escaping, or validation of the input content before processing.

rebuttal