experiment-tracking-swanlab

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the logging of external media and text metrics, creating a theoretical attack surface where untrusted data could contain instructions meant to influence the agent. \n
  • Ingestion points: Untrusted content such as text, images, and audio is passed to swanlab.log() in SKILL.md and references/visualization.md.\n
  • Boundary markers: The provided code snippets do not include explicit prompt-level delimiters or 'ignore' instructions for the logged content.\n
  • Capability inventory: The skill's capabilities are limited to local file logging and network synchronization of metrics to a SwanLab server.\n
  • Sanitization: Standard serialization and media processing are handled by the swanlab library before storage or transmission.\n- [EXTERNAL_DOWNLOADS]: The skill includes documentation for installing the swanlab library and its associated media dependencies (pillow, soundfile) and framework integrations from established package registries.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 07:08 PM
Security Audit — agent-trust-hub — experiment-tracking-swanlab