The Agent Skills Directory

[SAFE]: The skill is entirely instructional and does not contain any executable code, shell scripts, or commands that perform network exfiltration or sensitive file access.
[SAFE]: While the skill describes a workflow that involves reading external review artifacts, it provides prompt templates with explicit constraints (e.g., 'Do not edit any other files') to limit the agent's scope and mitigate potential indirect instruction issues.
[SAFE]: All file operations are confined to specific designated paths for review artifacts (e.g., 'ephemeral/reviews/'), following standard project organization practices.

consensus