From Extension to Plugin: Why This Matters

If you've been following the spec-driven development (SDD) movement, you already know the pain point: chat logs are ephemeral, and AI-assisted coding without persistent context is a recipe for architectural drift. Google introduced Conductor last year to fix exactly this—pushing project awareness out of throwaway chat sessions and into version-controlled markdown files.

Now Conductor is making a bigger move: it's graduating from a Gemini CLI extension into a full Conductor Plugin. According to the official announcement, this shift unlocks portability across multiple AI tooling ecosystems.

TL;DR — Same spec.md, same plan.md, but now conversational and tool-agnostic.

What Actually Changed

1. Conversational Workflow (Goodbye, Rigid Commands)

The old Conductor required strict command sequences. The plugin version lets you just talk to your AI assistant. It dynamically generates context, specs, and plans as you discuss features—no ceremony required.

2. Cross-Tool Portability

Previously locked to Gemini CLI, Conductor now runs wherever plugins are supported—including Antigravity CLI. Your configuration and development tracks persist seamlessly, so you can start a workflow in one tool and finish in another without losing context.

3. Persistent Artifacts Remain

spec.md and plan.md aren't going anywhere. The AI still manages project state in the background, checking off completed tasks and updating context intelligently.

Developer chatting with AI agent to generate spec.md and plan.md files in Conductor Plugin IT Technology Image

Installing Conductor Plugin

Getting started is a single command. If you're on Antigravity CLI:

# Install the Conductor Plugin from the official GitHub repo
agy plugins install https://github.com/gemini-cli-extensions/conductor

That's it. The plugin will register its skills, rules, MCP servers, and hooks in one package.

What Gets Bundled in the Plugin?

A plugin package can contain:

  • Skills — reusable agent capabilities
  • Rules — project-level constraints and guidelines
  • MCP servers — Model Context Protocol integrations
  • Hooks — lifecycle events that trigger agent behavior

This bundling is what makes Conductor portable. Instead of being a Gemini CLI-specific extension, it's now an ecosystem-wide capability.

# Example spec.md structure the agent maintains
## Project: My API Service
### Goals
- RESTful endpoints with OpenAPI spec
### Constraints
- Must use PostgreSQL 16+
### Tasks
- [x] Define schema
- [ ] Implement auth middleware
- [ ] Write integration tests

The agent reads this, updates it, and keeps it in sync with your actual codebase—no manual bookkeeping needed.

Conductor Plugin architecture showing portable ecosystem across Gemini CLI and Antigravity CLI

Limitations and Caveats

Before you rip out your existing workflow, keep these in mind:

  1. Plugin Ecosystem Maturity — Plugin support is still evolving. Not every CLI tool supports the full plugin spec yet, so portability is aspirational in some cases.
  2. Markdown Drift Risk — If your team doesn't commit spec.md and plan.md regularly, you'll lose the whole benefit. Version control discipline is non-negotiable.
  3. AI Hallucination in Specs — The agent can confidently write incorrect constraints into your spec. Always review generated specs before treating them as source of truth.
  4. Vendor Lock-in (Soft) — While Conductor is open source, it's still tightly coupled to Google's tooling ecosystem.

Rule of thumb: Treat spec.md like a PR—review it, diff it, approve it.

What to Learn Next

  • MCP (Model Context Protocol) — Understanding MCP servers is essential for building your own plugins.
  • Antigravity CLI plugin architecture — Read the docs to understand hook lifecycle.
  • Spec-Driven Development patterns — Explore how teams structure specs for monorepos vs. microservices.

If you're interested in how AI agents can absorb expert knowledge without retraining, the organizational second brain concept is a natural follow-up read. And for teams pushing hardware-adjacent AI workflows, CUDA 13.3's Tile C++ and Python 1.0 support is worth a look.

Terminal window running agy plugins install command for Conductor Plugin setup Software Concept Art

The Bottom Line

Conductor's move from extension to plugin is more than a packaging change—it's a signal that spec-driven development is becoming infrastructure, not a novelty. The conversational interface lowers the barrier, but the real value is still in the artifacts: persistent, version-controlled specs that survive beyond any single chat session.

If your team is still treating AI coding as a series of throwaway prompts, this is your cue to reconsider. Start small: install the plugin, write one spec.md for your next feature, and see how much cleaner your architecture decisions become when they're written down.

Your move, dev.

This content was drafted using AI tools based on reliable sources, and has been reviewed by our editorial team before publication. It is not intended to replace professional advice.