The Agent Skills Directory

[COMMAND_EXECUTION]: The skill leverages the Bash tool to perform essential repository operations. This includes using git for status, branch tracking, and pushing, as well as the GitHub CLI (gh) for creating pull requests. These commands are executed within the user's workspace to automate standard development workflows.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting external data such as file diffs and commit messages. This data is used to synthesize PR titles and descriptions. The skill implements a 'source of truth' policy by prioritizing the raw diff over commit messages to ensure accuracy and mitigate the risk of being misled by malicious commit metadata.
[INDIRECT_PROMPT_INJECTION]: The skill architecture includes security-centric configuration like disable-model-invocation: true, which prevents the AI agent from autonomously creating pull requests without a direct user command. This preserves human-in-the-loop oversight for remote state mutations.
[COMMAND_EXECUTION]: Step 4 of the workflow executes project-specific quality checks such as linters and type-checkers via shell commands. The results of these checks are surfaced to the user to prevent the fabrication of passing test statuses.

create-pr