The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill manages GitLab access tokens securely. Authentication is handled via a dedicated resolver in connections.ts that injects credentials into the PRIVATE-TOKEN header. The connector supports host overrides via the GITLAB_HOST environment variable, allowing users to target self-managed or GitLab Dedicated instances securely. Network operations are restricted to gitlab.com or the user-provided host.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes untrusted data from GitLab repositories and issues.
Ingestion points: Tools such as scripts/getFile.ts, scripts/getIssue.ts, and scripts/getJobLog.ts read content directly from external sources.
Boundary markers: The references/gitlab-formatting.md file provides guidance to the agent on how to handle GitLab-flavored markdown and auto-linking.
Capability inventory: The skill includes extensive write capabilities, such as scripts/commitFiles.ts, scripts/mergeMergeRequest.ts, and scripts/triggerPipeline.ts, which require careful agent oversight.
Sanitization: The connector utilizes Zod schemas for input and output validation to ensure data integrity and structural consistency.
[DYNAMIC_EXECUTION]: The cli.js entry point utilizes node:child_process.spawnSync to execute the connector's internal scripts. This is a standard architectural pattern used by the vendor to maintain compatibility across different Node.js versions and execution environments; it does not constitute arbitrary command execution as the targets are restricted to the skill's own local files.

gitlab