The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts using the python command in steps 5 and 9 of the process. These commands interpolate variables ({rate}, {mde}, {n}) which are derived from user input or potentially from the loaded brand profile files. If these variables are not strictly validated as numeric values, they could be used for command injection.
Evidence:
python "${CLAUDE_PLUGIN_ROOT}/scripts/sample-size-calculator.py" --baseline-rate {rate} --mde {mde} ... (SKILL.md, Step 5)
python "${CLAUDE_PLUGIN_ROOT}/scripts/significance-tester.py" --control-visitors {n} ... (SKILL.md, Step 9)
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from multiple external configuration files and templates to define its behavior and parameters. This data is then used in the agent's logic and passed to shell-executed scripts, creating an attack surface for indirect injection.
Ingestion points: ~/.claude-marketing/brands/_active-brand.json, profile.json, _manifest.json, and various template/SOP files (SKILL.md, Step 1).
Boundary markers: None mentioned; the skill appears to load and apply these files directly into the context.
Capability inventory: The skill has the capability to execute Python scripts via the shell.
Sanitization: No explicit sanitization or validation of the ingested file content is mentioned before it is used in script parameters.
[DYNAMIC_EXECUTION]: The skill invokes external Python scripts (sample-size-calculator.py and significance-tester.py) at runtime. While these scripts are located within the ${CLAUDE_PLUGIN_ROOT} and are likely part of the skill's own package, they represent executable logic that runs outside the core LLM instructions.

creative-testing-framework