Universal Conductor Prompts

Measure twice, code once.

Universal, agent-agnostic prompts that bring Conductor's Context-Driven Development workflow to any AI coding agent — Claude Code, Codex, GitHub Copilot, Windsurf, Antigravity.

Philosophy

Instead of just asking an AI to write code, these prompts enforce a consistent, high-quality lifecycle for every task:

Context → Spec & Plan → Implement → Review

By treating project context as a managed artifact alongside your code, you transform your repository into a single source of truth that drives every AI interaction with deep, persistent project awareness.

Prompts

| Prompt | Purpose | When to Use | | -------- | --------- | ------------- | | setup | Analyze project, define product/tech/workflow context | Once per project (or to update context) | | planning | Create spec + implementation plan for a feature/bug | Starting new work | | implement | Execute tasks from a plan with TDD workflow | Active development | | review | Review completed work against standards | After implementation | | status | Show progress overview | Anytime during development |

Workflow

1. Initialize Your Project (Run Once)

Use the setup prompt to set up your project context. This creates the conductor/ directory with:

conductor/
├── index.md                 # Master index of all context files
├── product.md               # Product definition and goals
├── product-guidelines.md    # Design/UX/prose standards
├── tech-stack.md            # Languages, frameworks, tools
├── workflow.md              # Development lifecycle (TDD, commits, etc.)
├── tracks.md                # Registry of all tracks (features/bugs)
└── tracks/                  # Track-specific artifacts

2. Plan a Feature or Bug Fix

Use the planning prompt to create a new track. The agent will:

• Ask clarifying questions about your feature/bug
• Generate a detailed specification (spec.md)
• Generate an actionable implementation plan (plan.md)
• Create track artifacts in conductor/tracks/<track_id>/

3. Implement the Track

Use the implement prompt to execute the plan. The agent will:

• Select the next pending task from plan.md
• Follow your workflow (TDD by default: Write Tests → Implement → Verify)
• Commit code with conventional commit messages
• Update progress in plan.md
• Run phase verification checkpoints

4. Review the Work

Use the review prompt to verify completed work. The agent will:

• Check compliance with the spec, plan, and style guides
• Run the test suite
• Scan for security issues
• Output a structured review report with severity-rated findings

5. Check Progress

Use the status prompt anytime to get a dashboard view of:

• Track progress (completed/in-progress/pending)
• Current phase and task
• Recommended next action

Installation

A built-in CLI syncs all prompts to the right location for each platform.

# Install to all platforms
pnpx @kurokeita/conductor-for-all --all

# Install to specific platform(s)
pnpx @kurokeita/conductor-for-all --platform copilotAgents
pnpx @kurokeita/conductor-for-all --platform copilotAgents,antigravity

# Preview without writing anything
pnpx @kurokeita/conductor-for-all --platform copilotAgents --dry-run

# Remove installed prompts
pnpx @kurokeita/conductor-for-all --platform copilotAgents --uninstall

# Interactive mode (no flags)
pnpx @kurokeita/conductor-for-all

Works with any package manager: npx @kurokeita/conductor-for-all, bunx @kurokeita/conductor-for-all

| Platform | Prompts installed to | | :--- | :--- | | Antigravity | ~/.gemini/antigravity/global_workflows | | Claude Code | ~/.claude/skills/<name>/SKILL.md | | Codex Skills | ~/.codex/skills/<name>/SKILL.md | | GitHub Copilot IDE Agents | .github/agents | | GitHub Copilot IDE Prompts | .github/prompts/*.prompt.md | | GitHub Copilot CLI Agents | ~/.copilot/agents | | GitHub Copilot CLI Skills | ~/.copilot/skills/<name>/SKILL.md | | Gemini CLI | ~/.gemini/commands/conductor (as .toml) | | Windsurf | ~/.codeium/windsurf/global_workflows |

Or clone and run locally:

git clone https://github.com/kurokeita/conductor-for-all
cd conductor-for-all
pnpm install
pnpm dev           # interactive mode
pnpm dev -- --platform copilotAgents --dry-run

Usage by Agent

GitHub Copilot (VS Code / CLI)

After running conductor-for-all --platform copilotPrompts, prompts are installed to .github/prompts/ and discoverable by VS Code. Reference them in chat:

# In VS Code chat
@workspace Use the setup prompt to set up this project

For GitHub Copilot CLI, use conductor-for-all --platform copilotCliSkills to install as skills.

Claude Code

Run conductor-for-all --platform claudeCode to install each Conductor command as a Claude Code skill in ~/.claude/skills/<name>/SKILL.md.

Codex

Run conductor-for-all --platform codexSkills to install each Conductor command as a Codex skill in ~/.codex/skills/<name>/SKILL.md.

Windsurf

Copy or symlink the prompt files into your Windsurf configuration, or paste the prompt content directly into a conversation.

Any Other Agent

These are plain Markdown files. You can:

1. Paste the content directly into a chat session
2. Reference the file if the agent supports file references (e.g., @file:commands/conductor-planning.md)
3. Include in system instructions if the agent supports custom instructions

Resume Capability

All prompts support automatic resume if a session is interrupted (crash, timeout, context loss, switching agents):

| Prompt | Resume mechanism | State tracked in | | -------- | ----------------- | ------------------ | | setup | Tracks which setup step was last completed and skips ahead | conductor/setup_state.json | | planning | Detects partially created tracks (spec exists but no plan, etc.) | conductor/tracks/<id>/metadata.json (status: "planning") | | implement | Detects [~] markers in tracks.md and plan.md to find interrupted tasks; checks for uncommitted changes | conductor/tracks.md + plan.md markers + git status | | review | Detects tracks with reviewing status and incomplete review fix tasks | conductor/tracks/<id>/metadata.json (status: "reviewing") | | status | Stateless — always reads current state fresh | N/A |

The resume check runs automatically as Step 0 of each prompt. If interrupted work is found, the agent will ask whether to resume or start fresh.

Generated Artifacts

| File | Created By | Purpose | | ------ | ----------- | --------- | | conductor/product.md | setup | Product vision, users, goals | | conductor/product-guidelines.md | setup | Design/UX/prose standards | | conductor/tech-stack.md | setup | Technical decisions | | conductor/workflow.md | setup | Development lifecycle rules | | conductor/tracks.md | planning | Registry of all tracks | | conductor/tracks/<id>/spec.md | planning | Track specification | | conductor/tracks/<id>/plan.md | planning | Track implementation plan | | conductor/setup_state.json | setup | Resume state for setup process | | conductor/tracks/<id>/metadata.json | planning | Track metadata and resume state |

Credits

Inspired by Conductor by Google — a Gemini CLI extension for Context-Driven Development.

These prompts distill Conductor's methodology into portable, agent-agnostic Markdown that works with any AI coding assistant.