The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes Python scripts to execute local shell commands such as latexmk and xelatex. These operations are used for the intended purpose of initializing and compiling LaTeX projects into PDF deliverables as described in the documentation.
[DYNAMIC_EXECUTION]: The script scripts/analyze_pilot.py uses importlib.util to dynamically load a local validation module (scripts/validate.py). This is used for structural verification of the skill's internal evaluation data and does not involve untrusted remote sources.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-supplied research notes and manuscripts. It implements explicit boundary markers and integrity instructions (e.g., 'Non-negotiable boundaries') to ensure the model adheres to scientific facts and prevents data fabrication, effectively mitigating risks associated with untrusted text processing.
[DATA_EXPOSURE]: The package includes extensive evaluation logs and records of past model interactions used for development. A review of these logs confirmed they contain only synthetic research data and fictional teaching examples, with no exposure of sensitive user data or system configurations.

light-write