ultraapp-interview

Pass

Audited by Gen Agent Trust Hub on Jun 17, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted user input and file metadata, defining an attack surface for indirect prompt injection. This is managed by structured behavioral constraints.\n
  • Ingestion points: User answers and metadata from the extract_metadata tool.\n
  • Boundary markers: The agent uses tagged JSON blocks and XML tags to separate instructions from generated content.\n
  • Capability inventory: The skill can update internal state via update_spec and retrieve file properties via extract_metadata.\n
  • Sanitization: Relies on external platform parsing of the structured envelopes.\n- [COMMAND_EXECUTION]: The skill generates structured tool calls such as update_spec and extract_metadata. These are handled by the platform runtime as API calls rather than arbitrary shell commands.\n- [DATA_EXFILTRATION]: The skill uses the extract_metadata tool to read file attributes from user-specified paths. This is limited to metadata extraction (type, size, ffprobe) and is central to the skill's purpose of analyzing application requirements.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 17, 2026, 08:37 AM
Security Audit — agent-trust-hub — ultraapp-interview