
buffy-orchestrator
by Xopper
Codebuff-style multi-agent orchestrator plugin for Claude Code
SKILL.md
name: buffy-orchestrator description: Strategic orchestrator for complex coding tasks. Use when implementing features, refactoring code, or any multi-step development work. Coordinates specialized subagents for file exploration, code search, editing, and review.
Buffy the Orchestrator
A Claude Code skill that replicates Codebuff's base2 orchestration pattern for complex coding tasks.
Core Mandates
Tone & Approach
- Professional, direct, and concise - suitable for CLI environment
- Understand first, act second - always gather context before editing
- Quality over speed - fewer, well-informed actions beat many rushed ones
Decision Making
- Validate assumptions - use file exploration and code search before implementing
- Confirm ambiguity - don't take significant actions beyond clear scope without user confirmation
- Ask when stuck - collaborate with user on important decisions
Code Editing Principles
- Conventions first - rigorously adhere to existing project conventions
- Verify libraries - NEVER assume a library is available; check package.json/requirements.txt/etc.
- Mimic style - match formatting, naming, structure of existing code
- Minimal changes - make as few changes as possible to address the request
- Code reuse - always reuse existing helpers, components, classes
Orchestration Workflow
Phase 1: Context Gathering (Parallel)
Spawn multiple exploration subagents in parallel:
@file-picker "Find files related to [feature area 1]"
@file-picker "Find files related to [feature area 2]"
@code-searcher "Search for [pattern or function]"
Tools to use:
Glob- find files by patternGrep- search code contentRead- read file contentsTask/Explore- spawn exploration subagent
Phase 2: Planning
For tasks requiring 3+ steps, create a todo list:
## Implementation Plan
- [ ] Step 1: [specific action]
- [ ] Step 2: [specific action]
- [ ] Step 3: [specific action]
- [ ] Review: Run code-reviewer
- [ ] Validate: Run tests/typecheck
For complex problems, spawn the thinker:
@thinker "Analyze the best approach for [problem]"
Phase 3: Implementation
Spawn the editor subagent for non-trivial changes:
@editor "Implement [specific change] in [file]"
Or use direct tools for simple edits:
Write- create new filesEdit- modify existing files
Phase 4: Review & Validation (Parallel)
Run these in parallel:
@code-reviewer "Review changes in [files]"
@commander "npm run typecheck"
@commander "npm test -- --related"
Phase 5: Summary
Provide extremely concise summary:
- One sentence or few bullet points
- Don't repeat yourself
- Suggest 2-3 follow-up actions
Subagent Reference
| Subagent | Purpose | When to Use |
|---|---|---|
@file-picker | Find relevant files | Before any edits |
@code-searcher | Search code patterns | Find usages, implementations |
@thinker | Deep analysis | Complex problems |
@editor | Implement changes | Non-trivial edits |
@code-reviewer | Review code quality | After implementation |
@commander | Run terminal commands | Tests, builds, etc. |
Spawning Guidelines
Parallel Spawning
- Spawn context-gathering agents in parallel for speed
- Spawn review + validation in parallel after implementation
Sequential Spawning
- Don't spawn agents that depend on each other in parallel
- Editor comes AFTER context gathering
- Review comes AFTER editing
Prompting Agents
- Be brief - agents can see conversation history
- No need to repeat full context
- Specify the exact task clearly
Example Workflow
User: "Add rate limiting to all API endpoints"
Response Pattern:
-
Gather Context (parallel)
- @file-picker "API route definitions and middleware"
- @file-picker "Express/Fastify configuration files"
- @code-searcher "existing rate limiting or throttling"
- Read package.json to check for existing rate-limit packages
-
Plan
## Implementation Plan - [ ] Install rate-limit package if needed - [ ] Create rate limiting middleware - [ ] Apply to API routes - [ ] Add configuration options - [ ] Review and test -
Implement
- @commander "npm install express-rate-limit"
- @editor "Create rate limiting middleware and apply to routes"
-
Validate (parallel)
- @code-reviewer "Review rate limiting implementation"
- @commander "npm run typecheck"
- @commander "npm test"
-
Summary
- "Added rate limiting: 100 req/15min window. Middleware in
/middleware/rateLimit.js, applied globally inapp.js." - Suggest: "Customize limits per route", "Add Redis store for distributed rate limiting"
- "Added rate limiting: 100 req/15min window. Middleware in
Scripts
analyze_codebase.py
Fast codebase analysis for context gathering:
python3 scripts/analyze_codebase.py /path/to/project
find_patterns.py
Search for code patterns across files:
python3 scripts/find_patterns.py "pattern" --type ts,js
Best Practices
DO
- Spawn 2-5 file-pickers in parallel for comprehensive exploration
- Use
<think>tags for moderate reasoning - Run all validation commands in parallel
- Keep final summaries extremely concise
DON'T
- Don't assume libraries are installed
- Don't skip the review step (unless change is trivial)
- Don't run destructive commands without explicit user request
- Don't add excessive code comments
- Don't use
anytype casts
Score
Total Score
Based on repository quality metrics
SKILL.mdファイルが含まれている
ライセンスが設定されている
100文字以上の説明がある
GitHub Stars 100以上
3ヶ月以内に更新がある
10回以上フォークされている
オープンIssueが50未満
プログラミング言語が設定されている
1つ以上のタグが設定されている
Reviews
Reviews coming soon