The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external sales conversation data, which constitutes an untrusted input surface.\n
Ingestion points: The conversation object is passed into extract_reasons_from_conversation and extract_reasons_with_llm within SKILL.md.\n
Boundary markers: The LLM prompt template in extract_reasons_with_llm interpolates the conversation text using a label (Conversation:) but lacks robust delimiters (e.g., XML tags or triple quotes) or specific instructions to the model to ignore potentially malicious commands embedded in the conversation logs.\n
Capability inventory: The skill demonstrates the capability to modify external records through update_crm_with_reasons, which calls crm_client.update_opportunity in SKILL.md.\n
Sanitization: The code snippets do not include logic to sanitize, filter, or validate the conversation content before it is processed by the LLM or used to update the CRM.

win-loss-reason-extraction