data-masking

Pass

Audited by Gen Agent Trust Hub on Sep 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from logs, databases, and APIs which acts as a surface for indirect prompt injection. Malicious instructions embedded within this data could attempt to subvert the masking logic or influence the agent's behavior.
  • Ingestion points: Raw input is processed by Python functions (mask_text_presidio), JavaScript middleware (sanitizeLog), and shell scripts (mask-db-for-dev.sh).
  • Boundary markers: The provided code examples lack explicit delimiters or instructions to treat input strictly as data and ignore any embedded natural language instructions.
  • Capability inventory: The skill environment involves shell command execution (via psql/pg_dump), file system access, and complex NLP processing.
  • Sanitization: While the skill focuses on masking PII, it does not include sanitization layers to detect or prevent prompt injection attacks contained within the processed text.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 8, 2026, 05:54 PM
Security Audit — agent-trust-hub — data-masking