run-fusion-cloud-recipes
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- Indirect Prompt Injection Surface: The skill processes external recipes and resolves dependencies, which represents a surface where untrusted data could influence agent behavior.
- Ingestion points: Data enters the context through fusion_cloud_job_prepare and fusion_cloud_batch_prepare when resolving dependencies, typed inputs, and variant IDs in the SKILL.md file.
- Boundary markers: The skill provides explicit instructions to freeze exact source versions, configurations, and hashes, which serves as a significant mitigation against unauthorized changes or unintentional obedience to embedded instructions.
- Capability inventory: The skill allows for preparing and submitting cloud jobs, polling job status, and reconciling artifacts and costs.
- Sanitization: Security is maintained through admission budgets, trusted output checks, and a reconciliation adapter to validate final billing and job outcomes.
Audit Metadata