gemini-review

Pass

Audited by Gen Agent Trust Hub on Feb 27, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted external data which creates a surface for indirect prompt injection. Malicious instructions embedded in code comments, commit messages, or PR descriptions could attempt to override the reviewer's persona or bias the review results.
  • Ingestion points: Untrusted data enters the prompt via git diff (branches/commits), cat (untracked files and context files), and gh pr view (PR descriptions) in scripts/gemini-review.sh.
  • Boundary markers: The script uses descriptive delimiters such as === PR DESCRIPTION ===, ADDITIONAL CONTEXT:, and CODE CHANGES TO REVIEW: to separate untrusted data from instructions.
  • Capability inventory: The skill executes local commands (git, gh, gemini) and the Gemini model is explicitly set to plan (read-only) mode, preventing the injection from performing modifications.
  • Sanitization: No sanitization or escaping is applied to the ingested content before it is interpolated into the large language model prompt.
  • [COMMAND_EXECUTION]: The script scripts/gemini-review.sh invokes several system commands including git, gh (GitHub CLI), and a gemini CLI tool. These operations are required for the skill's primary purpose but rely on the integrity of the user's PATH and the security of the underlying tools.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 27, 2026, 04:29 PM