astro-code-review

Pass

Audited by Gen Agent Trust Hub on Aug 20, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill incorporates a robust safety boundary that explicitly prevents the agent from modifying files, running project code, or executing scripts, ensuring the review process is entirely read-only and static.
  • [COMMAND_EXECUTION]: The use of shell commands is strictly limited to a whitelist of read-only Git operations (such as git diff and git status) necessary for identifying code changes, which prevents arbitrary command injection.
  • [EXTERNAL_DOWNLOADS]: Network activity is restricted to a single git fetch for metadata and targeted documentation lookups from well-known sources like MDN, providing necessary context without opening the agent to uncontrolled remote data.
  • [PROMPT_INJECTION]: The skill analyzes external code changes which could contain malicious content, but its strict non-execution policy and focus on generating a static report mitigate the risk of indirect prompt injection.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 20, 2026, 10:18 PM
Security Audit — agent-trust-hub — astro-code-review