github-multi-repo

Warn

Audited by Gen Agent Trust Hub on Mar 3, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the Bash capability to execute shell commands like 'gh', 'npm', and 'git' across multiple local and temporary directories.
  • [REMOTE_CODE_EXECUTION]: The skill logic includes cloning repositories to '/tmp/' and running 'npm test'. This behavior allows any code defined in the test scripts of those repositories to execute on the agent's host system.
  • [EXTERNAL_DOWNLOADS]: The skill utilizes the GitHub CLI to download repository data and clone source code from external organizations.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection by reading data from external repositories. 1. Ingestion points: The skill fetches 'package.json' and 'CLAUDE.md' files from GitHub via the 'gh api' command. 2. Boundary markers: No specific delimiters or safety instructions are used when interpolating this external content into the agent's context. 3. Capability inventory: The skill has extensive capabilities, including full Bash access, repository modification, and PR creation. 4. Sanitization: There is no evidence of filtering or sanitization of the content retrieved from external repositories before it is processed by the agent.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 3, 2026, 05:44 AM