The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the glab and git command-line tools to interact with GitLab repositories. These operations are performed using direct command execution in bash scripts and the subprocess.run module in Python.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources, such as GitLab issue titles and merge request diffs. While this presents an attack surface where a malicious user could embed instructions in an issue title, the skill mitigates this by:
Explicitly warning the agent about untrusted content in scripts/create-mr-from-issue.sh.
Sanitizing issue titles (removing non-alphanumeric characters) before using them in branch names.
Providing a dedicated Python helper for posting inline comments to avoid failures that could lead to misinterpretation of context.
[DYNAMIC_EXECUTION]: The skill instructions and scripts use python3 to process JSON data and generate API payloads. This is a standard and well-controlled use of the Python interpreter for data transformation.
[REMOTE_CODE_EXECUTION]: The automated review script (scripts/mr-review-workflow.sh) allows the execution of test commands. However, it implements a strict allowlist (npm test, pytest, cargo test, etc.) to prevent the execution of arbitrary or malicious code.

glab-mr