review-design

Pass

Audited by Gen Agent Trust Hub on Aug 4, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using unvalidated user-supplied parameters, including URLs, PR numbers, and commit SHAs. This creates a potential command injection surface if the executing environment does not properly sanitize these arguments.
  • Evidence: Use of <pr_number>, <commit_sha>, and <url> variables in shell commands for gh, git, and agent-browser tools in SKILL.md (Phase 0 and Phase 3).
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes content from external, potentially untrusted URLs through browser snapshots and accessibility trees.
  • Ingestion points: Live URL capture via agent-browser (SKILL.md, Phase 3).
  • Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions for the captured external content.
  • Capability inventory: The agent has access to shell tools (git, gh, agent-browser), file reading capabilities, and the ability to spawn sub-agents.
  • Sanitization: No explicit sanitization or filtering of external content is described before the data is processed by sub-agents.
  • [EXTERNAL_DOWNLOADS]: The skill recommends the installation of an external CLI tool, agent-browser, from the npm registry to enable live auditing capabilities.
  • Evidence: Installation check and instructions in SKILL.md (Phase 1).
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 4, 2026, 10:47 AM
Security Audit — agent-trust-hub — review-design