The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes Python scripts (tests/test_review_packet.py and evaluations/2026-11-06/run_revision_exercise.py) that use subprocess.run() to invoke the Python interpreter. These calls are limited to executing the skill's own utility scripts (scripts/review_packet.py) and temporary synthetic evaluation scripts during testing and demonstration. There is no evidence of arbitrary command execution from untrusted inputs.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data in the form of scientific manuscripts, which represents an attack surface for indirect prompt injection. However, the SKILL.md file contains explicit defense instructions, mandating that the agent treat all text, code comments, and hidden strings within the manuscripts as material to be reviewed rather than as instructional commands. This aligns with safety best practices for document-processing agents.

light-review