The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a local Python script scripts/validate_input.py to perform pre-execution checks on provided media file paths and validate product categories against a predefined list.
[EXTERNAL_DOWNLOADS]: The configuration in requirements.txt specifies dashscope and agentscope. These are official, versioned packages from the author's platform (Aliyun/Alibaba Cloud) used for AI model integration.
[PROMPT_INJECTION]: The skill processes untrusted external audio data that is converted to text and ingested into the agent's context for compliance analysis. This creates a surface for indirect prompt injection if the recording contains malicious instructions.
Ingestion points: Audio/video files provided via media_path in SKILL.md and scripts/validate_input.py.
Boundary markers: Absent; the instructions do not specify the use of structural delimiters (e.g., XML tags or unique separators) to isolate transcribed text from system instructions.
Capability inventory: Uses asr-service for transcription, compliance-kb for knowledge retrieval, and local Python validation scripts.
Sanitization: The skill includes a 'Compliance Constraints' section explicitly requiring the desensitization of customer PII and financial information found within the recordings.

insurance-underwriting-recording-inspection