The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could potentially contain malicious instructions intended to influence the agent's behavior.
Ingestion points: The skill ingests message content provided by the user and recipient data from external sources such as CSV files (Process step 2).
Boundary markers: The skill implements a robust boundary by requiring a mandatory "Execution Summary" and an explicit typed confirmation ("yes") from the user before any message is sent. This ensures human oversight of the final output.
Capability inventory: The agent has the capability to write to local audit logs, execute local scripts, and interact with external messaging APIs (Twilio/Brevo).
Sanitization: While the skill validates phone number formats (E.164) and checks message lengths, there is no explicit mention of sanitizing or escaping the content of personalization fields or message bodies to prevent the execution of embedded instructions.
[COMMAND_EXECUTION]: The skill utilizes local shell commands to manage approvals and track execution performance.
Evidence: It invokes local Python scripts including approval-manager.py, brand-voice-scorer.py, and execution-tracker.py located within the ${CLAUDE_PLUGIN_ROOT}. These are used for administrative and auditing purposes rather than executing arbitrary user input.

send-sms