The Agent Skills Directory

[SAFE]: The skill is a collection of markdown-based instructions and templates designed to help users draft natural voice conversations. Audit of all nine files confirms that no scripts are executed, no external code is downloaded, and no sensitive credentials are hardcoded or exfiltrated.- [PROMPT_INJECTION]: The templates include dynamic variables (e.g., {customer_name}, {callback_reason}) that ingest untrusted data from user_data or telephony transcripts, creating an indirect prompt injection surface. The skill addresses this by providing structural guidelines for guardrails and out-of-scope refusals to stabilize the agent's persona. Ingestion points: {customer_name}, {callback_reason}, and other user-defined variables in SKILL.md and asset templates. Boundary markers: Use of explicit sections (PERSONALITY, CONTEXT, FLOW, GUARDRAILS) helps structure the prompt, although the skill does not show specific delimiters for the interpolated variables. Capability inventory: Use of tool triggers like @fetch_calendar_slots, @book_calendar_slot, and @transfer_to_human. Sanitization: The templates do not explicitly specify sanitization methods for the dynamic variables, relying instead on high-level guardrails.

write-bolna-prompts