The Agent Skills Directory

[Command Execution]: The skill utilizes subprocesses to detect local git repository settings. This is implemented safely using fixed argument lists to prevent command injection and includes timeouts to bound execution.
[Indirect Prompt Injection]: As the skill retrieves and displays GitLab merge requests and CI logs, it processes untrusted data. It addresses this surface by sanitizing output through a robust redaction filter and enforcing size limits to prevent oversized payloads from being processed by the agent.
[Credential Security]: The tool implements secure storage for OAuth tokens, employing owner-only directory structures and restricted file modes (0600). It further prevents secret exposure by scrubbing sensitive tokens from all diagnostic and response data.
[Security Testing Seed Analysis]: Automated scans detected suspicious URLs; however, these are located in test files as data seeds for verifying that the skill correctly rejects unsafe OAuth verification URIs.

gitlab