email-request-intake
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill acknowledges and addresses the risk of processing untrusted data from email bodies, attachments, and linked web pages. It provides instructions to the agent to treat this content as data rather than instructions and to ignore embedded requests that attempt to expand scope or reveal secrets.
- Ingestion points: The skill explicitly reads email bodies, attachments, and linked URLs (SKILL.md).
- Boundary markers: It mandates the use of <EXTERNAL_DATA> tags to isolate untrusted content from the rest of the agent context.
- Capability inventory: The instructions mention capabilities such as sending emails, updating task trackers, and command execution, but specify that these must be gated by a written authorization policy and live user confirmation.
- Sanitization: The skill emphasizes structured extraction with schema validation and safe attachment inspection in a controlled environment.
Audit Metadata