This pull request introduces a new, modular Gemini CLI automation system for GitHub, enabling secure, role-based AI workflows for pull request review, issue triage, and general invocation. It adds a central dispatch workflow that parses user commands and routes them to the appropriate specialized workflow, each governed by strict security and operational protocols.

The most important changes are:

1. Gemini Dispatch Workflow and Command Routing

• Added .github/workflows/gemini-dispatch.yml, a new central workflow that listens for comments or reviews invoking @gemini-cli. It parses the command (/review, /triage, or general invocation), checks author permissions, acknowledges the request, and dispatches to the correct sub-workflow (gemini-review, gemini-triage, or gemini-invoke). If the command is unrecognized or fails, it posts a fallback error message.

2. Specialized Gemini Command Prompts and Protocols

• Added .github/commands/gemini-review.toml: Defines the persona, security constraints, and detailed review protocol for the AI code review agent, including severity guidelines, comment formatting, and strict boundaries on what the agent can and cannot do.
• Added .github/commands/gemini-invoke.toml: Defines the persona, workflow, and security constraints for general-purpose AI development tasks, emphasizing plan-approve-execute-report cycles and safe handling of untrusted input.
• Added .github/commands/gemini-triage.toml: Defines a specialized prompt for issue triage, focusing on label selection from a provided list and secure shell command usage.

3. Security and Operational Safeguards

• All prompts and workflows strictly prohibit command substitution in shell commands, enforce tool exclusivity (no direct git/shell for repo operations), and require careful handling of untrusted input to prevent injection or accidental execution. [1] [2] [3] [4]

These changes collectively establish a robust, extensible, and secure foundation for AI-driven automation in GitHub repositories.