edgespeak-name-speakers
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted transcript and roster data which could potentially contain malicious prompt instructions. However, it mitigates this risk by using a local script (
speaker_names.py) to generate structured JSON evidence reports, filtering and truncating raw text before it is audited by the agent. - Ingestion points: The agent reads transcript content through the
inspectworkflow and thespeaker-evidence.jsonreport generated from user-provided files. - Boundary markers: The workflow relies on structured data outputs and explicit audit steps to separate raw content from identity conclusions.
- Capability inventory: The skill allows for local file reading and writing (via the provided script) and uses agent reasoning to resolve identities.
- Sanitization: The script truncates text samples to 320 characters and uses specific regex patterns to extract clues rather than passing the entire raw transcript to the agent in one prompt.
Audit Metadata