open-design-blog-factory

Pass

Audited by Gen Agent Trust Hub on Aug 27, 2026

Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill uses specific formatting and directives to override the agent's default behavior, including tags like <author_preferences priority="highest"> and explicit instructions to "follow these first" in the event of contradictions.
  • [DATA_EXPOSURE]: The instructions expose an absolute local file path (/Users/ashleyli/.codex/skills/blog-indexing-automation/SKILL.md), which reveals the internal directory structure and the system username ashleyli to the agent context.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and process data from external sources, presenting a risk of indirect prompt injection.
  • Ingestion points: The skill retrieves content from external GitHub issues and pull requests through the gh issue list and gh pr view commands (e.g., in the Topic discovery & scoring section).
  • Boundary markers: There are no delimiters or instructions to treat the data retrieved from GitHub as untrusted, increasing the risk of the agent obeying malicious instructions embedded in issue comments.
  • Capability inventory: The skill environment allows for the execution of several shell-based tools, including pnpm, gh, tsx, and pngquant.
  • Sanitization: No input validation or sanitization is performed on the data fetched from the external repository before it is processed by the agent.
  • [COMMAND_EXECUTION]: The skill frequently instructs the agent to execute local command-line utilities, such as pnpm for build tasks, gh for repository interaction, and pngquant for image optimization.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 27, 2026, 05:41 PM