The Agent Skills Directory

[SAFE]: A thorough security analysis of the skill's instructions and references found no malicious patterns, unauthorized data access, or dangerous command execution. The skill's use of local file access and network fetching is entirely consistent with its purpose of academic paper evaluation.\n- [PROMPT_INJECTION]: The indirect prompt injection surface was evaluated and found to be handled safely within the skill's operational context.\n
Ingestion points: The skill retrieves data from well-known and trusted academic sources, including OpenReview forums and official conference guidelines, to inform its simulations.\n
Boundary markers: While technical delimiters are absent, the skill's instructions require the agent to paraphrase and focus on specific technical criteria, which significantly reduces the risk of the agent following instructions embedded in the fetched text.\n
Capability inventory: The skill uses standard tools such as Bash, Write, and Edit for maintaining project-local memory and report generation.\n
Sanitization: Instructions specifically direct the agent to ground criticisms in the provided paper evidence and to avoid direct quotation of external review text, providing a natural defense against injection content.

paper-reviewer-simulator