The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process live transcription data, which constitutes an untrusted data source that could contain hidden instructions targeting the agent.
Ingestion points: The agent processes Turn messages received via WebSocket from wss://streaming.assemblyai.com/v3/ws (referenced in SKILL.md).
Boundary markers: The skill explicitly instructs the agent to "Exclude ... transcript text" from its final output, providing a functional boundary between raw data and agent response.
Capability inventory: The skill allows the use of Write and Edit tools, which could potentially be triggered by malicious commands found within the transcribed speech.
Sanitization: While the skill mandates excluding transcripts from output, it lacks technical sanitization or validation rules for the content of the transcripts before they are processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill references official documentation and configuration files from AssemblyAI to guide its operations.
Evidence: Multiple links to https://www.assemblyai.com/docs/ are included in references/official-docs.md for parameter verification and model selection.

assemblyai-core-workflow-b