lowband
Pass
Audited by Gen Agent Trust Hub on Aug 4, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill contains instructions that override standard agent behavior and response formatting ("standing protocol: it applies to EVERY turn", "lowband owns its downgrade policy"). These instructions are intended for UX optimization during specific user states. The skill explicitly states that the agent "gains NO extra authority" and must continue to seek approval for high-impact, destructive, or external actions, mitigating the risk of safety protocol bypass.- [COMMAND_EXECUTION]: The instructions permit the agent to autonomously handle "in-scope, safe, reversible choices" such as small refactors or test fixes without user escalation. While this increases agent autonomy, the scope is restricted to non-destructive operations and explicitly requires user approval for any high-risk operations, maintaining a safe execution environment.
Audit Metadata