partner-marketplace-strategy
Pass
Audited by Gen Agent Trust Hub on Sep 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest external data from the web (competitor listings, partner directories, and category benchmarks) to inform its strategic recommendations. This creates a surface for indirect prompt injection if external sources contain adversarial instructions.
- Ingestion points: Steps 4, 7, and 11 in
SKILL.mdinvolve using a web browsing tool to research competitors and verify fee schedules. - Boundary markers: The skill mitigates this risk through strict instructional boundaries, requiring the agent to "Flag provenance on every benchmark quoted" and to explicitly treat operator-published figures as "marketing" or "ceilings."
- Capability inventory: The skill's capabilities are restricted to generating a Markdown decision memo. It does not perform file-system writes, shell command execution, or network operations to untrusted domains beyond the browsing tool.
- Sanitization: The skill relies on analytical framing rather than technical sanitization to process external content safely.
- [SAFE]: The skill demonstrates a strong security posture by treating all external data with skepticism, requiring verification of sources, and explicitly defining disqualifying "readiness gates" for high-risk or low-viability scenarios.
Audit Metadata