The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands via git and glab to manage branch workflows and commit history. These commands rely on parameters like branch names and commit SHAs derived from the environment or user input. The instructions mitigate potential command injection risks by requiring the agent to verify remote branches and request explicit user confirmation before pushing changes or creating merge requests.
[PROMPT_INJECTION]: The skill processes data from remote git logs, creating a potential surface for indirect prompt injection. Ingestion points: Remote commit messages and branch names are retrieved during log and list operations. Boundary markers: Data is presented to the user in structured markdown tables for review. Capability inventory: The agent has the ability to perform git write operations, push to remote repositories, and create merge requests. Sanitization: The skill mandates user review and confirmation of all commit lists and merge request details before any remote actions are executed.
[DATA_EXFILTRATION]: No patterns of sensitive data exfiltration were identified. Network activity is limited to standard GitLab operations through authenticated tools.
[SAFE]: The skill follows established software engineering best practices for backporting code and provides clear safety rules to prevent common version control errors.

gitlab-cherrypick