discover-attack-surface

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests external project data to perform taint analysis, which creates a potential vulnerability surface.\n
  • Ingestion points: The skill reads partition plans from .opentaint/tracking/rules/plans/ and project dependency JAR files from .opentaint/project/dependencies/.\n
  • Boundary markers: There are no boundary markers or instructions specified to delimit external content or to prevent the agent from obeying instructions embedded in the dependency names, method signatures, or plan metadata.\n
  • Capability inventory: The skill uses shell commands like unzip, javap, and grep to inspect binaries and writes YAML tracking units to the .opentaint/tracking/rules/ directory.\n
  • Sanitization: No sanitization logic is provided to validate class paths, method names, or signatures extracted from the untrusted plan files before they are interpolated into shell commands.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 02:13 AM
Security Audit — agent-trust-hub — discover-attack-surface