sumsub-conformance
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted regulation documents (PDF or text) to extract requirements. While this presents an indirect prompt injection surface where malicious text could attempt to influence the agent's extraction or mapping logic, the skill mitigates this by enforcing a human-in-the-loop confirmation step before proceeding to the deterministic tracing phase.
- Ingestion points: Regulatory documents (PDF/text) read via the Read tool or scripts/render_regulation.py.
- Boundary markers: Explicit instructions to show extracted requirements to the user for verification in Step 1 of the procedure.
- Capability inventory: Execution of Bash/Python scripts (resolve_graph.sh, trace.py), network access to api.sumsub.com, and file system writes for artifacts.
- Sanitization: The conformance trace logic is deterministic and separate from the initial natural language extraction step.
- [COMMAND_EXECUTION]: The skill uses Bash scripts to orchestrate API calls (sumsub_curl.sh) and Python scripts for data processing. These operations are limited to the skill's own internal logic and the official Sumsub API domain.
- [EXTERNAL_DOWNLOADS]: The skill optionally references pymupdf (fitz) for PDF rendering if available, but includes a safe fallback to the environment's native tools if the library is not installed.
Audit Metadata