review-change
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill automatically executes commands derived from the repository under review. Specifically, it runs typechecks, linters, and tests identified in 'agent instructions, then the package scripts, then CI'. If a repository contains malicious scripts in these locations, the skill will execute them during the 'Check the fixes' step.
- [COMMAND_EXECUTION]: The skill incorporates user-supplied arguments into shell commands, such as
git rev-parse --verify <ref>^{commit}. While standard for repository tools, this requires the execution environment to properly sanitize the<ref>argument to prevent command injection. - [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing untrusted data from the repository.
- Ingestion points: The skill reads
git diffoutput, commit messages,package.json, and various documentation files likeCONTRIBUTING.mdandCLAUDE.mdfrom the local file system. - Boundary markers: The
references/reviewer-brief.mdfile uses a triple-dash (---) separator to attempt to delimit instructions from the injected repository data. - Capability inventory: The skill has the capability to modify files in the working tree and execute arbitrary shell commands defined in project metadata.
- Sanitization: There is no explicit logic described to sanitize or escape the content extracted from the repository before it is passed to the subagent or executed as a command.
- [DYNAMIC_EXECUTION]: The skill dynamically determines which executable commands to run at runtime by parsing project-specific configuration files (e.g., linter and type-checker configs). This dynamic lookup of execution paths increases the risk if configuration files are tampered with.
Audit Metadata