The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The skill executes a local shell script as part of its primary task.
Evidence: SKILL.md instructs the agent to execute scripts/reveal.sh.
Analysis: The script content is transparent and benign, only echoing a static string. It does not attempt privilege escalation, persistence, or network communication.
[PROMPT_INJECTION] (SAFE): The skill uses authoritative instructional language to guide the agent.
Evidence: SKILL.md contains phrases like 'MUST follow these exact steps' and 'IMPORTANT RULES'.
Analysis: This is classified as natural instructional language intended for task completion rather than an attempt to bypass safety filters or extract system prompts.
[DATA_EXPOSURE] (SAFE): The skill reads a local asset file.
Evidence: assets/secret.txt is read by the agent.
Analysis: The file is a local resource belonging to the skill package and contains non-sensitive data. No system files or environment variables are targeted.

secret-code