The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions and evaluation files contain explicit rules preventing the agent from following user prompts that ask to skip evidence or fabricate certainty.
[INDIRECT_PROMPT_INJECTION]: Although the skill processes external user data, it implements an evidence discipline policy that requires classifying all claims into facts, inferences, or assumptions, reducing the risk of malicious instructions in the data being executed.
[COMMAND_EXECUTION]: The skill references local internal scripts for routing. These are standard architectural components of the agent framework and do not represent a security risk.
[DATA_EXFILTRATION]: No network access or attempts to read sensitive files or credentials were detected.

marketing-experimentation