ecom-image2

Pass

Audited by Gen Agent Trust Hub on Aug 21, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill workflow processes user-supplied natural language and interpolates it into structured JSON templates. There is a surface for indirect prompt injection where untrusted user input could be crafted to manipulate the final prompt sent to the image generation model.
  • Ingestion points: User request input defined in Step 1 of SKILL.md.
  • Boundary markers: The instructions lack explicit boundary markers or warnings to the agent to treat user input as data rather than instructions during interpolation.
  • Capability inventory: The skill possesses command execution capabilities via the Bash tool and Codex CLI.
  • Sanitization: While the provided helper script (scripts/imagegen.sh) uses jq for some sanitization, the agent's internal assembly of the JSON payload in Step 3 does not have explicit sanitization guidelines.
  • [COMMAND_EXECUTION]: The skill dynamically constructs and executes shell commands by piping agent-generated JSON data into the Codex CLI tool. This dynamic execution surface is necessary for the skill's primary function but represents a potential vector if the generated payloads are maliciously manipulated.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 21, 2026, 09:47 AM
Security Audit — agent-trust-hub — ecom-image2