email-verification
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from contact CSV files, which creates a vector for indirect instructions that could potentially influence agent behavior.
- Ingestion points:
SKILL.md(Step 2: Read the input CSV). - Boundary markers: The instructions do not define delimiters or specific boundary markers to isolate the ingested CSV content.
- Capability inventory: The agent is capable of making network requests to verification and sequencer APIs (Steps 3 and 6) and writing cleaned results to local files (Step 5).
- Sanitization: No sanitization or validation of the email strings is specified before they are passed to API endpoints.
- [CREDENTIALS_UNSAFE]: Step 0 directs the agent to "Ask for their API credentials." Handling raw secrets within the conversation flow is a risk as they may be stored in session history or logs.
- [EXTERNAL_DOWNLOADS]: The skill is instructed to "Fetch or read the provider's API documentation" at runtime. This requires the agent to access and interpret content from external, third-party web resources.
Audit Metadata