github-multi-repo
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill extensively uses the
Bash()tool to execute shell commands for repository management, including cloning, branching, committing, and creating pull requests. - [INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection because it processes data from external repositories and executes associated lifecycle scripts.
- Ingestion points: The skill uses
gh repo cloneto pull entire repositories into the local environment andgh apito read file contents (e.g.,package.json,CLAUDE.md). - Boundary markers: No delimiters or isolation instructions are present to prevent the agent from obeying instructions found within the files it reads.
- Capability inventory: The skill has powerful capabilities including file system write access, network operations via GitHub API, and arbitrary shell command execution via
npm testandnpm update. - Sanitization: No validation or sanitization is performed on the content of the ingested repositories before executing their scripts.
- [EXTERNAL_DOWNLOADS]: The skill downloads code and metadata from GitHub repositories using the
ghCLI andgit. It also installs and updates packages from the NPM registry. - [REMOTE_CODE_EXECUTION]: The practice of cloning a repository and automatically running
npm testornpm updatewithin its directory constitutes a form of remote code execution, as the test scripts and post-install hooks are defined by the remote repository's author.
Audit Metadata