The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides implementation templates that ingest external data from audio URLs. This constitutes a potential attack surface where untrusted content (the transcription output) could be introduced into the agent's context if the templates are used in an automated workflow.
Ingestion points: The transcribeUrl method and associated functions in SKILL.md and references/implementation.md accept an external url or audioUrl as input.
Boundary markers: The provided code snippets do not implement specific delimiters or "ignore" instructions for the processed content to prevent the agent from following instructions found within transcripts.
Capability inventory: The skill environment is configured with file manipulation tools (Read, Write, Edit) and network access (Bash(curl:*)).
Sanitization: The sanitize helper in references/implementation.md validates URL structure and strips query parameters to prevent logging sensitive tokens, but it does not filter the content of the data fetched from the URL.

deepgram-prod-checklist