The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources that could potentially contain malicious instructions intended to influence the agent's behavior.
Ingestion points: Customer data is gathered from CRM and analytics MCPs, including transaction histories and engagement events.
Boundary markers: None are specified to delimit external data from instructions.
Capability inventory: The skill has the capability to read local files, execute a Python script (campaign-tracker.py), and interact with multiple MCPs.
Sanitization: There is no explicit mention of sanitizing or escaping the data pulled from external systems before it is processed or stored.
[COMMAND_EXECUTION]: The skill executes a local Python script using dynamic arguments.
Evidence: In Step 8 of the process, it runs python "${CLAUDE_PLUGIN_ROOT}/scripts/campaign-tracker.py" --brand {slug} --action save-insight --data '{"type":"cohort", ...}'. The {slug} parameter is derived from the _active-brand.json file, and other data may originate from user input or processed results.
[DYNAMIC_EXECUTION]: The skill uses local environment variables and computed paths to execute its logic.
Evidence: It relies on ${CLAUDE_PLUGIN_ROOT} to locate and execute internal scripts, which is a standard pattern for skill functionality but involves runtime path resolution.

cohort-analysis