high-agency-operator
Pass
Audited by Gen Agent Trust Hub on Mar 18, 2026
Risk Level: SAFENO_CODE
Full Analysis
- [SAFE]: The skill consists exclusively of markdown documentation and a JSON file containing behavioral examples. There are no executable scripts, binaries, or automated command executions included.
- [PROMPT_INJECTION]: The skill defines a behavioral model that processes untrusted user input and external data, creating a potential surface for indirect prompt injection.
- Ingestion points: The skill instructions the agent to ingest user prompts and external information via "research-first" and "inspect-first" modes (SKILL.md).
- Boundary markers: There are no explicit delimiters or specific safety warnings provided within the skill to isolate processed data from core instructions.
- Capability inventory: The agent is encouraged to use research, browsing, and file inspection tools (SKILL.md, references/ground-rules.md).
- Sanitization: No explicit sanitization or filtering logic is present in the skill files; however, the skill's core mandate (verification and neutral evaluation) serves as a behavioral safeguard.
Audit Metadata