gs-review
Pass
Audited by Gen Agent Trust Hub on Nov 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes several git commands (e.g.,
git status,git diff,git merge-base) to determine the scope of the code review. It also runs project-local test and lint commands as part of the 'QA execution' role. - [DATA_EXFILTRATION]: The skill sends code diffs to external LLM providers, specifically the Grok and Claude CLI tools, to perform 'second-opinion' reviews. This is a core feature of the skill and involves well-known services.
- [INDIRECT_PROMPT_INJECTION]: The skill processes instructions and role definitions from a
REVIEW.mdfile located within the repository being reviewed. - Ingestion points: The file is identified in
SKILL.mdand its contents are parsed inreferences/review-md.md. - Boundary markers: The skill instructions explicitly tell the agent to treat the file content as data, map it to specific categories, and ignore any other requests or instructions within the file.
- Capability inventory: The skill has the ability to execute shell commands (
gitand project checks) and interact with external CLI tools. - Sanitization: The skill mitigates risks by reading the
REVIEW.mdfile from the git base commit (to prevent a malicious change from altering its own review rules) and by filtering out any commands that are not read-only tests or linters. - [DYNAMIC_EXECUTION]: The skill executes shell commands defined in the
Checkssection of the project'sREVIEW.mdfile. While it filters for non-destructive commands, this represents the execution of logic defined in external repository data.
Audit Metadata