The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses node to run a local JavaScript file (scripts/significance.js) for statistical analysis.
[DATA_EXFILTRATION]: The scripts/significance.js script includes functionality to read files via the --control-file and --variant-file command-line arguments. It uses fs.readFileSync on the provided path without any validation or restriction, which could allow the agent to be tricked into reading and exposing sensitive files such as .env or configuration files.
[INDIRECT_PROMPT_INJECTION]: \n- Ingestion points: The agent is instructed in SKILL.md to read .agents/marketing-strategy.md and .agents/marketing-learnings.md to gather context. \n- Boundary markers: None. \n- Capability inventory: The agent has the ability to execute the included Node.js script. \n- Sanitization: None. The skill assumes the contents of these external files are safe and should be used to guide its behavior.

ab-testing