The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a local Python script to perform audience simulation and message testing.
Evidence: Found in SKILL.md, the process uses python "${CLAUDE_PLUGIN_ROOT}/scripts/audience-simulator.py" with various actions such as list-panels, create-panel, and test-message.
[DYNAMIC_EXECUTION]: The skill constructs shell commands at runtime by interpolating variables into command strings.
Evidence: SKILL.md uses placeholders like {slug}, {name}, {id}, and {variants} within the shell command arguments passed to the python interpreter.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it processes untrusted user-supplied marketing copy and segment data that could contain malicious instructions.
Ingestion points: User-provided "Message variants" (headline, body, CTA) and audience segment definitions.
Boundary markers: None described in the instructions to delimit user content from instructions.
Capability inventory: Reads sensitive brand profile files from ~/.claude-marketing/ and executes a Python subprocess via audience-simulator.py (documented in SKILL.md).
Sanitization: No explicit sanitization or validation of the input content is mentioned.

message-test