deep-learning-experiment-workflow-skill

Pass

Audited by Gen Agent Trust Hub on Mar 27, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The workflow directs the agent to execute shell commands and scripts for git operations, data probes, and training/evaluation entrypoints throughout the lifecycle (specifically documented in stages/00-bootstrap/bootstrap-checklist.md, stages/01-investigation/investigation-guide.md, and stages/04-implementation/implementation-template.md).
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by requiring the ingestion of untrusted external data.
  • Ingestion points: The agent is directed to read external URLs, open-source repositories, and research papers as noted in stages/01-investigation/investigation-guide.md.
  • Boundary markers: The templates and guides do not include explicit instructions or markers to delimit external content or warn the agent to disregard instructions embedded in those sources.
  • Capability inventory: The agent possesses capabilities to execute shell commands and modify the local filesystem across several stages, including stages/01-investigation/investigation-guide.md and stages/04-implementation/implementation-template.md.
  • Sanitization: There is no mention of sanitizing or validating external content before it is processed by the agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 27, 2026, 07:35 PM