spec-kit-conduct-ext

Conduct wrapper for Spec Kit that executes each spec-kit phase with sub-agents to stabilize behavior and reduce context pollution in the main agent.

Commands

Command	What it does
speckit.conduct.run	Run one phase (specify, plan, tasks, or implement) through step-by-step sub-agent orchestration

Installation

Install from the latest release (requires a tagged release to exist):

specify extension add --from https://github.com/twbrandon7/spec-kit-conduct-ext/archive/refs/tags/v1.0.0.zip

Or for local development:

specify extension add --dev /path/to/spec-kit-conduct-ext

Usage

Phase-by-Phase Execution

Run one phase at a time. Each invocation delegates that phase's steps to sub-agents and keeps the main agent context clean.

/speckit.conduct.run <phase> <feature name or requirements>

Supported phase values:

specify
plan
tasks
implement

Run each phase separately:

/speckit.conduct.run specify Build a pomodoro app
/speckit.conduct.run plan
/speckit.conduct.run tasks
/speckit.conduct.run implement

The command does not run all phases in one invocation. You should review outputs between phases and then trigger the next phase explicitly.

Framework Configuration

If you want to explicitly specify the AI coding tool for orchestration, set framework in .specify/extensions/conduct/conduct-config.yml:

framework: "copilot"

This file is created automatically when the extension is installed (using conduct-config.template.yml as the template). load.sh returns that value in its JSON output as framework. Supported values match the agent identifiers used by the loader: copilot, claude, gemini, cursor-agent, qwen, opencode, codex, windsurf, kilocode, auggie, roo, codebuddy, amp, shai, kiro-cli, bob, qodercli, tabnine, kimi, generic.

Requirements

• Spec Kit >=0.3.1

License

MIT