Google Unveils Conductor: Gemini CLI Extension for Context-Driven Markdown and Workflows

Google has introduced a new open-source extension called Conductor for its Gemini CLI. This tool enhances AI code generation by providing a structured and context-driven workflow. Conductor allows for the storage of essential project knowledge, technical decisions, and work plans in versioned Markdown files within the repository, ensuring Gemini agents utilize these structured files rather than relying on informal chat prompts.

Transforming AI Coding with Conductor

Traditionally, AI coding is session-based. Users paste code into a chat and describe the task, but the context is lost after the session ends. Conductor addresses this issue by creating a persistent context directory within the repository. This directory captures crucial elements including product goals, constraints, technology stacks, workflow rules, and style guides as Markdown documents. Gemini reads these files during each execution, which allows for consistent AI behavior across different machines, shells, and team members.

Life Cycle of Development

The development process with Conductor operates under a clear lifecycle: Context → Spec and Plan → Implement. Instead of jumping directly from a natural language request to coding, Conductor creates a track, establishes a specification, formulates a plan, and only then executes it.

Installation of Conductor in Gemini CLI

Installing Conductor as a Gemini CLI extension is straightforward. Users can execute a single command:

gemini extensions install https://github.com/gemini-cli-extensions/conductor –auto-update

The optional –auto-update flag ensures that the extension remains in sync with the latest version. After installation, Conductor commands become available when working in a project directory.

Project Setup

The setup process begins with the command /conductor:setup. This command initiates an interactive session that builds the foundation for the project’s context. Conductor collects information about the product, target users, requirements, technology stack, and development practices. The system generates a conductor/ directory containing essential files like:

• conductor/product.md
• conductor/product-guidelines.md
• conductor/tech-stack.md
• conductor/workflow.md
• conductor/code_styleguides/
• conductor/tracks.md

These files help define how AI should approach the project, outlining target users, primary features, acceptable technologies, testing expectations, and coding standards. All documents reside in Git alongside the source code, making changes trackable and auditable.

Managing Work with Tracks

Conductor introduces the concept of tracks for managing units of work, such as features or bug fixes. New tracks can be created using the command:

/conductor:newTrack “Add dark mode toggle to settings page”

Each track generates a directory under conductor/tracks/ that contains:

• spec.md – Detailed requirements and constraints
• plan.md – Stepwise execution plan
• metadata.json – Identification and status information

Conductor also assists in drafting the spec and plan by leveraging existing context files, allowing developers to refine and approve them. This process ensures that all implementations adhere to a clear and version-controlled plan.

Execution of Plans

Once the planning phase is complete, users can initiate implementation using the /conductor:implement command. Conductor reads the plan.md file, selects the next pending task, and executes it according to the established workflow. Typical steps in this cycle include:

• Reviewing relevant files and context
• Proposing necessary code modifications
• Running tests as outlined in workflow.md
• Updating task statuses in plan.md and global tracks.md

Conductor incorporates checkpoints at key stages to allow for human verification, minimizing the risk of unreviewed changes.

Compatibility with Brownfield Projects

Conductor is adaptable for brownfield codebases in addition to new projects. When the setup command is executed in an existing repository, it helps extract implicit knowledge from the team and documents it in explicit Markdown format. Over time, the context directory becomes a comprehensive representation of the project’s architecture and constraints, with team behaviors codified in workflow.md, tech-stack.md, and style guide documents.

Conclusion

Conductor significantly enhances the development process within Gemini CLI by promoting context-driven development. This extension, available under an Apache 2.0 license, organizes project context as versioned Markdown files, maintains explicit execution plans, and streamlines team workflows. For those interested, more details can be found by visiting Filmogaz.com.