task-tracker
Pass
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: LOWPROMPT_INJECTION
Full Analysis
- Indirect Prompt Injection (LOW): The skill processes user messages to detect context switches and transition task states, creating a surface for manipulating agent behavior through crafted input.\n
- Ingestion points: The
current_messageis evaluated by the abandonment detection logic in the workflow.\n - Boundary markers: No explicit delimiters are present to separate user-provided content from the skill's logic.\n
- Capability inventory: The skill is limited to internal state tracking and conversational interventions; no execution or exfiltration capabilities were identified.\n
- Sanitization: The logic does not demonstrate sanitization or validation of the processed user input.
Audit Metadata