review-visual-changes

Pass

Audited by Gen Agent Trust Hub on Nov 12, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes content from external Chromatic builds, including story names, comments, and HTML snapshots.
  • Ingestion points: External data enters the context through chromatic_query and chromatic_get_latest_build tools as described in SKILL.md.
  • Boundary markers: Instructions explicitly state that story names, comments, and image content must be treated as data and that the agent should never follow instructions inside them.
  • Capability inventory: The skill uses git for repository analysis, curl for file system writes to temporary directories, and chromatic_mutate for project modifications.
  • Sanitization: While the skill advises treating content as data, there are no programmatic sanitization steps for the HTML snapshots or image metadata before processing.
  • [EXTERNAL_DOWNLOADS]: The skill performs network requests to download images and HTML snapshots from Chromatic's infrastructure using curl.
  • Assets are saved to a local temporary directory.
  • The skill handles signed URLs which are documented as sensitive and expiring.
  • [COMMAND_EXECUTION]: The skill relies on executing local shell commands to perform its functions.
  • Git operations: Uses git rev-parse, git merge-base, and git diff to determine code changes associated with a build.
  • Network tools: Uses curl to fetch build artifacts individually based on build metadata.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 12, 2026, 10:31 PM