Google releases Conductor: a context-driven Gemini CLI extension that shops data as markdown and orchestrates agent workflows

Google launched Conductor, an open supply preview extension to Gemini CLI that turns AI code era into structured, context-driven workflows. Conductor shops product data, technical selections, and work plans as versioned Markdown in a repository and drives Gemini brokers from these recordsdata as a substitute of advert hoc chat prompts.

From chat-based coding to context-driven improvement

Most present AI coding is session-based. In the event you paste code right into a chat to explain a activity, the context disappears when the session ends. Conductors deal with it as a central problem.

As an alternative of short-term prompts, Conductor maintains a persistent context listing inside the repository. Seize your product targets, constraints, tech stack, workflow guidelines, and magnificence information as markdown. Gemini then reads these recordsdata each time it runs. This makes AI conduct reproducible throughout machines, shells, and crew members.

Conductor additionally enforces a easy lifecycle:

Context → Specification and Planning → Implementation

This extension doesn’t bounce straight from pure language requests to code modifying. First create the monitor, create the specs, generate the plan, after which execute.

Putting in Conductor on Gemini CLI

Conductor runs as a Gemini CLI extension. Set up is a single command.

gemini extensions set up https://github.com/gemini-cli-extensions/conductor --auto-update

of --auto-update flag is non-obligatory and retains the extension in sync with the newest launch. After set up, it is possible for you to to make use of Conductor instructions inside the Gemini CLI if you’re in your mission listing.

Establishing the mission with /conductor:setup

The workflow begins with project-level setup.

This command runs an interactive session that establishes the fundamental context. Conductor asks questions on your product, customers, necessities, expertise stack, and improvement practices. From these solutions, conductor/ a listing containing a number of recordsdata, for instance:

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

These artifacts outline how the AI ​​causes about your mission. Describe target market, high-level performance, accepted applied sciences, testing expectations, and coding conventions. They reside in Git together with the remainder of the supply code, so modifications to the context are reviewable and auditable.

Vans: Specs and plans as first-class deliverables

Conductor introduction monitor Represents a unit of labor, resembling a characteristic or bug repair. Create a monitor utilizing:

or a brief description:

/conductor:newTrack "Add darkish mode toggle to settings web page"

For every new monitor, Conductor creates a listing in: conductor/tracks/<track_id>/ Contains:

• spec.md
• plan.md
• metadata.json

spec.md Keep detailed necessities and constraints to your truck. plan.md Comprises a step-by-step execution plan divided into phases, duties, and subtasks. metadata.json Shops identifiers and standing info.

Conductor makes use of present context recordsdata to help in drafting and planning specs. The developer then edits and approves them. The essential level is that every one implementations should comply with an express, versioned plan.

Applied by /conductor:implement

As soon as your plan is prepared, Give management to the agent.

conductor reads plan.mdchoose the following pending activity and run the configured workflow. Typical cycles embody:

1. Examine related recordsdata and context.
2. Recommend code modifications.
3. Carry out checks or checks based on conductor/workflow.md.
4. Replace activity standing plan.md and international tracks.md.

This enhancement additionally inserts checkpoints at part boundaries. At these factors, the conductor pauses for human verification earlier than persevering with. This prevents the agent from making use of large-scale refactorings that haven’t been reviewed.

A number of operational instructions assist this stream.

• /conductor:standing View monitor and activity progress.
• /conductor:evaluate Helps validate accomplished work towards product and magnificence pointers.
• /conductor:revert Roll again a monitor, part, or activity utilizing Git.

Undos are outlined by way of tracks quite than uncooked commit hashes, making reasoning simpler in multi-change workflows.

Brownfield tasks and crew workflows

Conductor is designed to work with brownfield codebases in addition to new tasks. when operating /conductor:setup For present repositories, contextual classes are a technique to extract implicit data out of your crew into express Markdown. Over time, as extra tracks are executed, the context listing turns into a compact illustration of the system’s structure and constraints.

Crew-level behaviors are encoded as follows: workflow.md, tech-stack.mdtype information file. Any engineer or AI agent utilizing Conductor with that repository will inherit the identical guidelines. That is helpful for imposing testing methods, linting expectations, and imposing accredited frameworks amongst contributors.

The context and plan are in Git, so you may evaluate, talk about, and modify your code in the identical course of as your supply recordsdata.

Necessary factors

• Conductor is a Gemini CLI extension for context-driven improvement.: That is an open supply Apache 2.0 licensed extension that runs inside the Gemini CLI and drives the AI ​​agent from a repository-local Markdown context quite than an ad-hoc immediate.
• The mission context is saved as versioned Markdown. conductor/: recordsdata like product.md, tech-stack.md, workflow.mdand code type guides outline the product targets, technical selections, and workflow guidelines that the agent reads each time it runs.
• Work is organized into vans. spec.md and plan.md: /conductor:newTrack Create a monitor listing containing: spec.md, plan.mdand metadata.jsonmaking necessities and execution plans express, reviewable, and tied to Git.
• Implementation is managed by /conductor:implement Observe-aware operation: The agent performs duties based on: plan.mdreplace progress tracks.mdassist /conductor:standing, /conductor:evaluateand /conductor:revert For progress inspection and rollback utilizing Git.

Please test lipo and technical details. Please be happy to comply with us too Twitter Do not forget to affix us 100,000+ ML subreddits and subscribe our newsletter. grasp on! Are you on telegram? You can now also participate by telegram.