open-code-review-delegate

Pass

Audited by Gen Agent Trust Hub on Aug 29, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the open-code-review tool from the official Alibaba Group NPM repository if it is not present in the environment.
  • Evidence: npm install -g @alibaba-group/open-code-review (SKILL.md).
  • [COMMAND_EXECUTION]: The skill workflow relies on executing several CLI tools to perform its primary function, including repository metadata extraction and file reading.
  • Evidence: Use of ocr delegate, git diff, git show, and cat <path> across the review workflow steps (SKILL.md).
  • [INDIRECT_PROMPT_INJECTION]: As a code review tool, the skill ingests and processes untrusted code files and git history which may contain adversarial instructions designed to influence the agent's logic.
  • Ingestion points: Processes workspace files, untracked files, and git diffs (SKILL.md, Steps 1-4).
  • Boundary markers: The documentation recommends using structured checklists and grouping files by rules, though explicit prompt boundaries for the LLM are not provided.
  • Capability inventory: Accesses the file system (cat), interacts with the git CLI (git), and performs tool installation (npm).
  • Sanitization: The skill provides proactive security advice in the 'Recovering Oversized Background Context' section, instructing the agent to use shell-safe arguments and avoid double-quoted templates to prevent command injection (SKILL.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 29, 2026, 06:17 AM
Security Audit — agent-trust-hub — open-code-review-delegate