The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted marketing content provided by the user or from external files, which could contain hidden instructions intended to bias the evaluation.
Ingestion points: The content variable and file paths referenced in the log action within SKILL.md.
Boundary markers: Absent; instructions do not specify the use of delimiters or warnings for the agent to ignore embedded instructions in the test content.
Capability inventory: The skill utilizes shell execution for Python scripts (eval-runner.py, prompt-ab-tester.py) as defined in SKILL.md.
Sanitization: No sanitization or filtering of the input content is described before it is passed to the evaluation scripts.
[COMMAND_EXECUTION]: The skill operates by executing shell commands that incorporate user-provided inputs.
Script execution: Relies on python to run scripts located in ${CLAUDE_PLUGIN_ROOT}/scripts/ for administrative and evaluation tasks.
Argument Interpolation: Variables such as {name}, {content}, {label}, and {description} are interpolated directly into shell command arguments, which could lead to command injection if the underlying scripts do not handle these parameters securely.

prompt-test