The Agent Skills Directory

[SAFE]: The skill is entirely instructional and does not contain any executable scripts, automated tools, or network operation code. It mandates explicit user authorization for all data access and recruitment activities.- [PROMPT_INJECTION]: The skill incorporates specific defensive instructions (Rule 9) requiring the agent to treat external content strictly as data and to ignore any commands or instructions embedded within processed research material.- [DATA_EXFILTRATION]: Strict prohibitions are established in Rules 2, 3, and 5 against accessing sensitive system locations such as environment files (.env), private messages, or unauthorized CRM databases.- [INDIRECT_PROMPT_INJECTION]: While the skill defines a surface for processing external data, it mitigates risk through a defined evidence chain: 1. Ingestion points: Processes interviews, surveys, and support tickets (SKILL.md); 2. Boundary markers: Rule 9 explicitly instructs to ignore embedded commands; 3. Capability inventory: No code or automated tools are present in the skill; 4. Sanitization: Mandates pseudonymization, data minimization, and separation of raw evidence from interpretation.

customer-research