saas-agent-integration-deep
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external business data, including agent archetypes, architecture descriptions, and business briefs. This creates a theoretical surface for indirect prompt injection where malicious instructions could be embedded in the business documents being analyzed.
- Ingestion points: SKILL.md specifies inputs from "Client intake", "Engagement owner", and "Technical leads" (business briefs, architecture diagrams, data rights).
- Boundary markers: The "Capability and Permission Boundaries" section explicitly restricts the agent to "read-only analysis and drafting" and forbids exposing client data, purchasing services, or granting autonomous permissions.
- Capability inventory: The skill contains no scripts, subprocess calls, or network operations. It is restricted to content composition and cross-referencing.
- Sanitization: The skill implements a "Quality Standards" and "Anti-slop gate" review process to ensure outputs are verified and professional.
- [SAFE]: The skill does not contain any executable code, scripts, or remote dependency installations. All references are to internal markdown files within the skill repository. No obfuscation, persistence mechanisms, or credential handling issues were identified.
Audit Metadata