The Agent Skills Directory

[SAFE]: The skill defines a legitimate organizational structure for research leadership. All instructions focus on scientific workflow management, decision synthesis, and publication-quality writing without introducing security risks.
[PROMPT_INJECTION]: Instructions regarding the PI's authority to 'disregard' or 'override' feedback are restricted to scientific judgment (e.g., choosing a simpler analysis method over a complex one). They do not attempt to bypass safety filters, ignore system prompts, or override agent constraints.
[DATA_EXFILTRATION]: No patterns of data exfiltration or sensitive file access were detected. The skill uses standard tools (Read, Write, Edit, Skill) for their intended purposes in managing research documentation and team coordination.
[REMOTE_CODE_EXECUTION]: No instances of remote code execution, dynamic command generation, or unsafe package installations are present. The skill consists entirely of markdown documentation and templates. References to software libraries (e.g., scanpy, pyDESeq2) are provided as guidance for implementation tasks delegated to other specialists.
[COMMAND_EXECUTION]: No direct shell command execution or subprocess spawning was identified within the skill's instructions.

principal-investigator