gitlab-cli-skills

Pass

Audited by Gen Agent Trust Hub on Nov 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from GitLab, such as issue titles and CI/CD job logs, which could potentially contain malicious instructions. The author has implemented several mitigations:
  • Ingestion points: Data is fetched in scripts/create-mr-from-issue.sh (issue titles) and scripts/ci-debug.sh (job logs).
  • Boundary markers: Scripts explicitly wrap external content in --- BEGIN/END EXTERNAL CONTENT --- markers to inform the agent that the data is untrusted.
  • Capability inventory: Scripts utilize git and glab binaries for project management.
  • Sanitization: The create-mr-from-issue.sh script employs regex-based sanitization to clean issue titles before they are used in shell commands or branch names.
  • [COMMAND_EXECUTION]: The skill automates GitLab workflows by executing glab and git commands. Specifically, scripts/mr-review-workflow.sh allows for the execution of test commands (e.g., npm test). To mitigate the risk of arbitrary code execution, the script enforces a strict allowlist (ALLOWED_COMMANDS) that only permits standard, well-known test runners.
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions and automation for downloading the official glab binary and its optional cosign dependency from trusted sources like Homebrew and GitLab's official releases. The glab-duo sub-skill also manages updates for the GitLab Duo AI binary from official GitLab infrastructure.
  • [DATA_EXPOSURE]: The skill is designed to handle GitLab authentication. It follows best practices by retrieving tokens via the official glab auth token command or reading from the standard glab configuration file (~/.config/glab-cli/config.yml), avoiding insecure hardcoded credentials.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 22, 2026, 01:12 AM