The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to fetch venue details from external URLs (CFP pages and author guidelines) to verify rebuttal budgets and deadlines. This is a legitimate functional requirement for maintaining accuracy against changing venue rules.
[PROMPT_INJECTION]: The instructions include a specific security guardrail advising the agent that addressing text to automated review tools (prompt injection) is an ethics violation. No adversarial injection patterns were found in the skill content.
[DATA_EXFILTRATION]: The skill explicitly forbids uploading, posting, or submitting content to external systems. It mandates that review text be processed transiently and never committed to repositories or public artifacts, ensuring data confidentiality.
[COMMAND_EXECUTION]: The provided Python scripts (check_budget.py, check_coverage.py, venueyaml.py) use only standard library modules and perform deterministic checks on local files. They do not execute arbitrary shell commands or untrusted code.
[DYNAMIC_EXECUTION]: The check_budget.py script uses the zlib module to inflate PDF streams for the purpose of counting pages in compiled LaTeX rebuttals. This is a standard and safe implementation for PDF parsing without external dependencies.

write-rebuttal