vision-expert

Pass

Audited by Gen Agent Trust Hub on Nov 5, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze content from target repositories, including source code, model configurations, and evaluation logs. This establishes an attack surface where malicious instructions could be embedded in the data processed by the agent. However, the workflow emphasizes evidence-driven evaluation and visual verification, which mitigates the risk of blindly following injected instructions. Ingestion points include target repo files identified in references/project_workflow.md and local reference books described in references/local_books.md. No explicit sanitization or boundary markers are defined for the repository content analysis.
  • [COMMAND_EXECUTION]: The skill instructs the agent to use shell commands to search for and inspect existing vision pipelines within a project environment. This includes searching for source files, model weights, and performance benchmarks as part of the detection phase in the workflow described in SKILL.md and references/project_workflow.md.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 5, 2026, 09:27 AM