emotional-arc-management
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input to determine the conversation's emotional state and phase, which creates an indirect prompt injection surface.
- Ingestion points: User messages are processed in the
detect_emotionsandassess_current_statefunctions (SKILL.md). - Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands within the processed user messages.
- Capability inventory: The skill is limited to state tracking and message template selection; it does not perform file operations, network requests, or subprocess executions.
- Sanitization: The skill uses regular expression matching for emotion detection but does not perform sanitization or validation of the input message content before using it to drive phase transitions.
Audit Metadata