ultra-ship
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a multi-step review workflow that ingests code diffs and external agent feedback. Ingestion points: Branch diffs and responses from review agents (e.g., cursor-agent, opencode2). Boundary markers: Prompts use specific formatting delimiters like 'findings:' and 'items:' to structure interactions. Capability inventory: Workspace modifications, subprocess command execution (git, gh, agent CLIs), and PR management. Sanitization: Employs HTML escaping for report rendering and uses list-based subprocess arguments to avoid shell injection.
- [COMMAND_EXECUTION]: Orchestrates developer tools by executing git, gh, and various AI agent CLIs via subprocess calls using argument lists to mitigate shell injection vulnerabilities.
- [EXTERNAL_DOWNLOADS]: References external resources including Google Fonts for styling HTML reports and the @opencode-ai/cli npm package to provide code review functionality.
Audit Metadata