clear-context
Fail
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: HIGHPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONPRIVILEGE_ESCALATION
Full Analysis
- [PROMPT_INJECTION]: The skill contains instructions that tell subagents to bypass user confirmation and safety prompts if specific flags are present in a state file. It uses phrases like 'DO NOT pause for user confirmation' and 'Do not ask the user for confirmation'.
- [INDIRECT_PROMPT_INJECTION]: The subagent's behavior is controlled by the content of a local file (.claude/session-state.md), which can be used to inject malicious instructions. * Ingestion points: .claude/session-state.md. * Boundary markers: None. * Capability inventory: All available tools. * Sanitization: None.
- [PRIVILEGE_ESCALATION]: The 'dangerous' mode explicitly instructs the agent to skip permissions and confirmation dialogs, allowing autonomous execution of sensitive operations.
Recommendations
- AI detected serious security threats
Audit Metadata