orchestrating-fusion
Pass
Audited by Gen Agent Trust Hub on Aug 4, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides detailed instructions for a 'Fusion' workflow that emphasizes least privilege and verification. It explicitly notes that prompt-based instructions are not a strong security boundary and advocates for platform-level sandboxing, read-only roles, and user approval for any destructive or external operations.
- [PROMPT_INJECTION]: The skill defines a workflow that processes untrusted data (codebases) and delegates tasks based on those inputs, which represents an indirect prompt injection surface (Category 8).
- Ingestion points: The 'lead' agent reads user requests and codebase files in Step 1; the 'executor' agent reads listed skill files from the file system in Step 5.
- Boundary markers: The workflow uses structured text envelopes (such as the OBJECTIVE and STATUS blocks) to separate coordination metadata from content.
- Capability inventory: The agents being orchestrated are described as having potential shell access and file-write tools, depending on the host platform environment.
- Sanitization: The 'lead' agent is required to perform a comprehensive 'gate' check (Step 6), which includes reading back all changed paths, validating symbols, and independently verifying outcomes before reporting success to the user.
Audit Metadata