The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources including GitLab issue titles, merge request descriptions, and CI/CD job logs. Malicious actors could potentially embed instructions in these fields to influence agent behavior.
Ingestion points: Data is ingested via glab issue view, glab ci trace, and glab mr view across multiple sub-skills.
Boundary markers: The skill explicitly implements --- BEGIN/END EXTERNAL CONTENT --- delimiters in scripts like scripts/ci-debug.sh and scripts/create-mr-from-issue.sh to isolate untrusted content.
Capability inventory: The skill possesses significant capabilities including the ability to create/delete repositories, approve merge requests, and execute predefined test commands through glab and git CLI tools.
Sanitization: Scripts such as scripts/create-mr-from-issue.sh include logic to slugify and sanitize issue titles before using them in branch names.
[COMMAND_EXECUTION]: The script scripts/mr-review-workflow.sh executes a test command provided as an argument. However, this is mitigated by a strict, hardcoded allowlist (ALLOWED_COMMANDS) that restricts execution to standard test runners like npm test, pytest, and cargo test. Arbitrary command injection is prevented by validating the input against this fixed list.

gitlab-cli-guide