Love VS Code. Love Copilot's inline diff and unified UX. Want Claude Code-level development workflows. BitFrog makes it happen.
You might be in one of these situations:
- Your company requires GitHub Copilot — enterprise policy, security compliance, no alternatives allowed
- You genuinely prefer VS Code + Copilot — the inline diff view, the integrated Chat panel, the slash command UX, the seamless editor integration
- You've seen what Claude Code can do — structured brainstorm → plan → execute → review workflows — and you want that inside your favorite editor
The problem: Copilot's agent mode doesn't have built-in structure. When you need to go from idea to shipped feature, there's no workflow, no discipline, no handoff chain.
BitFrog gives you Claude Code-level structured development workflows, running entirely inside GitHub Copilot.
BitFrog adds 7+1 specialized agents directly inside Copilot Chat. Each handles one phase of development, handing off to each other automatically:
@bitfrog (ask anything — auto-routes to the right agent)
↓
brainstorm → plan → execute → review
↕
debug
mozi (autonomous — give a goal, it does everything)
No new editor. No new subscription. Just install and go.
Agent Plugin (recommended, VS Code 1.110+):
Cmd+Shift+P → Chat: Install Plugin → rainyulei/bitfrog-copilot
VS Code Marketplace:
Search "BitFrog Copilot" in Extensions → Install
BitFrog works best with GPT-5.4 — it is the recommended model for all agents.
To set the model: click the model picker in Copilot Chat → select GPT-5.4.
Other models (Claude Sonnet 4, Gemini 2.5 Pro, etc.) also work, but GPT-5.4 provides the best balance of instruction-following and tool use for BitFrog's structured workflows.
Different agents need different levels of autonomy:
| Agent Phase | Recommended Mode | Why |
|---|---|---|
| @bitfrog (router) | Default | Only reads code and routes — no risk |
| @bitfrog-brainstorm | Default | Explores and asks questions — no code changes |
| @bitfrog-plan | Default | Reads codebase, writes plan doc — minimal risk |
| @bitfrog-execute | Bypass Approvals | Needs to edit files, run tests, commit — frequent tool calls; approving each one breaks flow |
| @bitfrog-debug | Bypass Approvals | Needs to read logs, edit code, run tests freely |
| @bitfrog-review | Default | Reviews code, suggests changes — you decide what to apply |
| @bitfrog-mentor | Default | Read-only guidance |
| @bitfrog-mozi | Bypass Approvals | Autonomous end-to-end execution — needs full tool access without interruption |
How to set: Click the permission level indicator in the Chat view (VS Code 1.111+).
Rule of thumb: Use Default when thinking (brainstorm, plan, review). Use Bypass when doing (execute, debug, mozi).
Open Copilot Chat, select @bitfrog, and describe what you want to do:
- "I want to add a user authentication system" → routes to brainstorm
- "Here's the spec, break it into tasks" → routes to plan
- "Execute the plan above" → routes to execute
- "This API returns 500" → routes to debug
- "Review the changes" → routes to review
- "Just do it — add dark mode support" → routes to mozi
Or select a specific agent directly from the dropdown.
| Agent | What it does | When to use it | Mode |
|---|---|---|---|
| @bitfrog | Auto-routes to the right agent | Don't know where to start? Start here | Default |
| @bitfrog-brainstorm | Explores ideas, challenges assumptions, writes specs | "I want to build..." / "I have an idea..." | Default |
| @bitfrog-plan | Maps dependencies, decomposes into bite-sized TDD tasks | Design is done, need a concrete plan | Default |
| @bitfrog-execute | TDD implementation with parallel sub-agents | Plan is ready, time to code | Bypass |
| @bitfrog-debug | Four-diagnostic-method root cause analysis | Something is broken | Bypass |
| @bitfrog-review | Three-reflection review (spec → code quality → user intent) | Code is done, need quality check | Default |
| @bitfrog-mentor | Guided learning through hints, never gives direct answers | Want to understand, not just get answers | Default |
| @bitfrog-mozi | [BETA] Autonomous deep worker — investigates, plans, executes, and self-verifies until goal is achieved | "Just do it" — give a focused goal, get it done end-to-end | Bypass |
These work behind the scenes — you don't interact with them directly:
| Sub-Agent | Used by | Purpose |
|---|---|---|
bitfrog-spec-reviewer |
brainstorm | Independent design spec review before approval |
bitfrog-code-reviewer |
review | Independent code quality review (peer reflection) |
bitfrog-task-worker |
execute | Parallel execution of independent tasks |
Most AI coding tools use hard rules: "you MUST write tests", "you MUST NOT skip review". BitFrog uses Chinese philosophy thinking models — agents don't follow rules mechanically, they understand why the rules exist.
| Rules-based approach | BitFrog's philosophy approach |
|---|---|
| "You MUST write tests" | Agent understands why tests matter → writes them naturally |
| "You MUST NOT skip root cause analysis" | Agent knows treating symptoms creates more problems |
| Checklist: pass / fail | Reflection: "Is my thinking process right?" |
| Find the optimal solution | Find the appropriate solution for this context |
Five principles drive all agents:
| Principle | Meaning |
|---|---|
| 中庸之道 (Right Measure) | Every action has its appropriate level — too much process is as harmful as too little |
| 格物致知 (Investigate First) | Understand the true nature of the problem before proposing solutions |
| 知行合一 (Unity of Knowing and Doing) | If you know you should do something but skip it, you don't truly understand |
| 辨证论治 (Diagnose Before Treating) | Same symptom, different root causes — classify the problem level first |
| 阴阳互生 + 三省吾身 (Awareness + Reflection) | Stay aware of the whole system; reflect on your thinking, not just the output |
Read the full guide: bitfrog-philosophy.md
There are excellent dedicated design skills available (e.g. Impeccable) that can be installed alongside BitFrog and used directly within your workflow. Rather than building a mediocre built-in design agent, we recommend pairing BitFrog with a specialized design skill of your choice.
Named after the Chinese philosopher 墨子 (Mozi), who believed in hands-on craftsmanship and end-to-end ownership.
Give Mozi a focused goal — it investigates the codebase, aligns with you on the approach, then autonomously executes, self-verifies, and iterates until the goal is achieved. No handoffs, no waiting between phases.
When to use Mozi vs the standard workflow:
| Scenario | Use |
|---|---|
| Exploring a new idea, unsure of direction | brainstorm → plan → execute → review |
| Clear, focused goal you want done end-to-end | @bitfrog-mozi |
| Complex multi-subsystem project | brainstorm → plan → execute → review |
| "Just add dark mode to the settings page" | @bitfrog-mozi |
Recommended mode: Bypass Approvals — Mozi needs full tool access to work autonomously without interruption.
All agents support English and 简体中文 automatically.
If BitFrog helps your workflow, give it a star — it helps other developers in the same situation discover it.
Inspired by Superpowers by Jesse Vincent. BitFrog builds on that foundation by replacing external rule constraints with internal philosophical thinking models.
MIT
BitFrog Copilot is part of the BitFrog product family.