The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of reviewer comments and revision emails provided by the user. While the workflow includes structured translation and classification steps that act as natural boundaries, the skill lacks explicit sanitization of the input text beyond formatting it into LaTeX blocks.
Ingestion points: User prompts containing reviewer comments and revision emails; user-provided LaTeX (.tex) files.
Boundary markers: The skill mandates using LaTeX quote blocks (\begin{quote}) and italics (\emph{}) to delimit external comments.
Capability inventory: The skill can write response plans and LaTeX files to an output/ directory, though instructions require explicit user authorization before file writing.
Sanitization: No explicit sanitization or filtering of the external comment text is defined.
[COMMAND_EXECUTION]: The workflow documentation (workflow.md) mentions the use of grep commands as part of a manual 'Gate Sync Check'. These are presented as guidance for manual verification steps and are not executed automatically by the skill runtime or loader.
[DATA_EXFILTRATION]: The skill references shared group resources (style profiles, terminology) located in a relative parent directory (../ieee-mg-share/). This access is consistent with the group's organizational structure and does not involve unauthorized access to sensitive system files or credentials.

ieee-mg-response