dupe
Pass
Audited by Gen Agent Trust Hub on Feb 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION] (LOW): The skill instructs the agent to execute curl commands to interact with api.dupe.com, involving command-line construction using user-provided inputs.
- [DATA_EXFILTRATION] (LOW): The skill performs network operations (POST requests) to a non-whitelisted domain (api.dupe.com). While consistent with its stated purpose, it involves sending user-provided data to an external service.
- [PROMPT_INJECTION] (LOW): The skill is vulnerable to indirect prompt injection because it ingests untrusted user input and processes it without sanitization. Ingestion points: User-provided productUrl and imageUrl. Boundary markers: Absent; user input is interpolated directly into JSON payloads. Capability inventory: Execution of curl commands. Sanitization: Absent; the instructions explicitly command the agent not to modify the user-provided URL in any way.
Audit Metadata