account-health-check
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, including conversational dumps, CRM exports, and uploaded PDF files, to generate health assessments. This processing of user-supplied or external content creates an attack surface for indirect prompt injection.
- Ingestion points: The skill explicitly accepts CRM tables, messy notes, and uploaded PDFs in SKILL.md (Workflow Step 1) and references/input-rules.md (Rule 1).
- Boundary markers: SKILL.md contains a mandatory 'Untrusted content' section that instructs the agent to treat fetched or uploaded text as data and never as instructions, with a requirement to report any ignored instructions to the user.
- Capability inventory: The skill possesses no dangerous capabilities; it does not perform network operations, file system modifications, or arbitrary code execution.
- Sanitization: The skill implements a defensive strategy by requiring the agent to explicitly state when it has ignored instructions embedded within processed data.
Audit Metadata