The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill sets up extraction prompts and evaluation criteria that process call transcripts, which constitute untrusted external data.
Ingestion points: Call transcripts retrieved via GET /v1/convai/conversations/{cid} as outlined in the design grounding section.
Boundary markers: The instructions do not explicitly mandate delimiters for the extraction prompts (config data_collection), though it notes that LLM extractions run independently for each field.
Capability inventory: The skill uses PATCH requests to update agent configurations, controlling how the platform processes conversation data.
Sanitization: While no technical sanitization is described, the skill provides a security-positive instruction to avoid copying transcripts to other logs and to respect zero-retention settings due to PII content.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill describes interactions with ElevenLabs' official API infrastructure.
The use of api.elevenlabs.io for configuration and transcript retrieval is a standard vendor operation.
Sensitive identifiers like $API_KEY and $AGENT_ID are treated correctly as placeholders for the user to provide, rather than being hardcoded.
The skill proactively warns against the exfiltration or improper logging of PII found in transcripts.

architect-post-call-data