Skip to content

Latest commit

 

History

History
600 lines (472 loc) · 19.5 KB

File metadata and controls

600 lines (472 loc) · 19.5 KB
name agent-implementation-plan
description Zapabob/Codex Custom Skills Implementation Plan - VRChat, Blender, YukkuriMovieMaker, YOLO integration

Zapabob/Codex Custom Skills Implementation Plan

Zapabob/Codex カスタムスキル実装計画

Overview | 概要

This document outlines the comprehensive implementation plan for adding new custom skills to the Zapabob/Codex fork. The plan covers skill conversion from existing agents, new skill creation, slash command integration, and MCP server development.

本文書はZapabob/Codexフォークに新しいカスタムスキルを追加するための包括的な実装計画を概述する。既存エージェントからのスキル変換、新しいスキル作成、スラッシュコマンド統合、MCPサーバー開発をカバーする。


Current State | 現状

Completed | 完了済み

Item Status Date
Upstream merge (openai/codex) ✅ Complete 2026-02-08
Custom features preservation ✅ Verified 2026-02-08
Git merge conflict resolution ✅ Complete 2026-02-08
Deep research (VRChat/Blender/YMM4/OpenClaw) ✅ Complete 2026-02-08
Slash Commands Implementation ✅ Complete 2026-02-08
MCP Server Configuration ✅ Complete 2026-02-08
MCP VRChat Server ✅ Complete 2026-02-08
MCP Blender Server ✅ Complete 2026-02-08
MCP YMM4 Server ✅ Complete 2026-02-08
MCP YOLO Server ✅ Complete 2026-02-08

Existing Custom Features | 既存の独自機能

Feature Type Files
/qc Slash Command slash_command.rs:23, chatwidget.rs:2883
/dev-mode Slash Command slash_command.rs:24, chatwidget.rs:2886
/git4d Slash Command slash_command.rs:50, chatwidget.rs:3084
/vr Slash Command slash_command.rs:51, chatwidget.rs:3086
/ar Slash Command slash_command.rs:52, chatwidget.rs:3087
QC System Code codex-rs/core/src/qc/
Git4D System Code codex-rs/core/src/git4d_accelerated.rs
Dev Mode Code codex-rs/core/src/orchestration/
Custom Agents Config .codex/agents/*.yaml (9 agents)

Implementation Plan | 実装計画

Phase 1: Agent to Skill Conversion (Week 1-2)

1.1 Convert Existing Agents to Skills

Agent Target Skill Priority Complexity
qc-optimizer.yaml qc-optimizer High Medium
git4d-runtime-checker.yaml git4d-runtime High Medium
git4d-schema-auditor.yaml git4d-schema High Medium
performance-analyst.yaml performance-analyst Medium Medium
security-audit.yaml sec-audit Medium Medium

1.2 Skill Structure

.codex/skills/{skill-name}/
├── SKILL.md              # Main skill definition
├── agents/
│   └── openai.yaml       # Agent configuration
└── scripts/
    └── run_{skill-name}.py  # Execution script

.cursor/skills/{skill-name}/
├── SKILL.md
├── agents/
│   └── openai.yaml
└── scripts/
    └── run_{skill-name}.py

Phase 2: New Skill Creation (Week 2-3)

2.1 New Skills to Create

Skill Technology Priority Complexity Description
vrchat-dev VRChat SDK3 High 8/10 VRChat world/avatar development
blender-cad Blender Python Medium 6/10 Blender CAD modeling automation
yukkuri-movie YMM4 Medium 7/10 ゆっくりMovieMaker video production
yolo-auto OpenClaw Medium 5/10 YOLO full-stack automation

2.2 Technology Stack Research

VRChat SDK3 (Complexity: 8/10)
VRChat SDK3.10.1 (Dec 2025)
├── Udon (VRChat's VM)
│   ├── Udon Graph (visual programming)
│   └── UdonSharp (C#-like syntax)
├── modularavatar (avatar configuration)
├── liltoon (shader)
├── Poiyomi (shader)
└── VRClightvolume (lighting)

Key MCP Tools Needed:

  • UdonSharp compilation
  • modularavatar integration
  • PhysBones configuration
  • Contact system setup
  • World upload automation
Blender Python (Complexity: 6/10)
Blender 4.0+
├── bpy (Python API)
├── Breaking API changes (3.x → 4.0)
├── STEP/IGES import
├── Geometry Nodes
└── USD export

Key MCP Tools Needed:

  • Geometry creation/modification
  • Material assignment
  • Render setting automation
  • Export to various formats
YukkuriMovieMaker (Complexity: 7/10)
YMM4 v4.49.0.2
├── Plugin Architecture
│   ├── [AudioEffect]
│   ├── [VideoEffect]
│   └── IMovieMakerPlugin
├── MIDI Integration
└── IPC Communication

Key MCP Tools Needed:

  • Scene management
  • Character animation
  • Audio effect control
  • Video rendering
  • Project file manipulation
OpenClaw (Complexity: 5/10)
OpenClaw Latest
├── MCP Protocol
├── Claude Code Integration
├── GPU Model Selection
│   ├── claude-3-opus
│   ├── gpt-4
│   └── gpu-5.3-codex (mentioned by user)
└── Multi-Agent Orchestration

Key MCP Tools Needed:

  • GPU model selection
  • Task distribution
  • Result aggregation

Phase 3: Slash Command Integration (Week 3-4)

3.1 New Slash Commands

Add to codex-rs/tui/src/slash_command.rs:

pub enum SlashCommand {
    // ... existing ...
    VrChat,    // NEW - VRChat development
    Blender,   // NEW - Blender CAD automation
    Yukkuri,   // NEW - YukkuriMovieMaker production
    Yolo,      // NEW - YOLO full automation
}

impl SlashCommand {
    pub fn description(self) -> &'static str {
        match self {
            // ... existing ...
            SlashCommand::VrChat => "VRChat world and avatar development",
            SlashCommand::Blender => "Blender CAD modeling automation",
            SlashCommand::Yukkuri => "ゆっくりMovieMaker video production",
            SlashCommand::Yolo => "YOLO full-stack automation with OpenClaw",
        }
    }
}

3.2 Handler Registration

Add handlers in codex-rs/tui/src/chatwidget.rs:

  • Line 2883: /vr handler (existing)
  • Line 2886: /ar handler (existing)
  • New: /VRChat, /Blender, /Yukkuri, /YOLO

Phase 4: MCP Server Development (Week 4-6)

4.1 New MCP Servers

mcp-servers/
├── mcp-vrchat-server/
│   └── src/
│       └── main.rs
├── mcp-blender-server/
│   └── src/
│       └── main.rs
└── mcp-ymm4-server/  # yukkuri-movie-maker
    └── src/
        └── main.rs

4.2 MCP Server Structure

// Example: mcp-vrchat-server/src/main.rs
use mcp_server::prelude::*;

#[tokio::main]
async fn main() -> Result<()> {
    let server = McpServer::new("vrchat", "VRChat MCP Server")
        .with_tool(UdonCompiler::tool())
        .with_tool(WorldUploader::tool())
        .with_tool(AvatarConfigurator::tool());

    server.run().await
}

Timeline | タイムライン

Week Phase Tasks Deliverables
1 Phase 1 Convert qc-optimizer agent to skill SKILL.md, agents/openai.yaml
1 Phase 1 Convert git4d agents to skills SKILL.md × 2, agents/openai.yaml × 2
2 Phase 1 Convert remaining agents SKILL.md × 3, agents/openai.yaml × 3
2 Phase 2 Create vrchat-dev skill SKILL.md, agents/openai.yaml, MCP server design
3 Phase 2 Create blender-cad skill SKILL.md, agents/openai.yaml, MCP server design
3 Phase 2 Create yukkuri-movie skill SKILL.md, agents/openai.yaml, MCP server design
3 Phase 2 Create yolo-auto skill SKILL.md, agents/openai.yaml, MCP server design
4 Phase 3 Add slash commands Rust code changes, handler integration
4-5 Phase 4 Build vrchat MCP server Working MCP server with tools
5-6 Phase 4 Build blender MCP server Working MCP server with tools
6 Phase 4 Build yukkuri MCP server Working MCP server with tools

File Structure | ファイル構造

Skills Directory Structure

.codex/
├── skills/
│   ├── qc-optimizer/
│   │   ├── SKILL.md
│   │   ├── agents/
│   │   │   └── openai.yaml
│   │   └── scripts/
│   │       └── run_qc-optimizer.py
│   ├── git4d-runtime/
│   │   ├── SKILL.md
│   │   ├── agents/
│   │   │   └── openai.yaml
│   │   └── scripts/
│   │       └── run_git4d-runtime.py
│   ├── git4d-schema/
│   │   ├── SKILL.md
│   │   ├── agents/
│   │   │   └── openai.yaml
│   │   └── scripts/
│   │       └── run_git4d-schema.py
│   ├── vrchat-dev/
│   │   ├── SKILL.md
│   │   ├── agents/
│   │   │   └── openai.yaml
│   │   └── scripts/
│   │       └── run_vrchat-dev.py
│   ├── blender-cad/
│   │   ├── SKILL.md
│   │   ├── agents/
│   │   │   └── openai.yaml
│   │   └── scripts/
│   │       └── run_blender-cad.py
│   ├── yukkuri-movie/
│   │   ├── SKILL.md
│   │   ├── agents/
│   │   │   └── openai.yaml
│   │   └── scripts/
│   │       └── run_yukkuri-movie.py
│   └── yolo-auto/
│       ├── SKILL.md
│       ├── agents/
│       │   └── openai.yaml
│       └── scripts/
│           └── run_yolo-auto.py

.cursor/
├── skills/
│   ├── qc-optimizer/
│   │   └── SKILL.md
│   ├── git4d-runtime/
│   │   └── SKILL.md
│   ├── git4d-schema/
│   │   └── SKILL.md
│   ├── vrchat-dev/
│   │   └── SKILL.md
│   ├── blender-cad/
│   │   └── SKILL.md
│   ├── yukkuri-movie/
│   │   └── SKILL.md
│   └── yolo-auto/
│       └── SKILL.md

MCP Servers Directory Structure

mcp-servers/
├── mcp-vrchat-server/
│   ├── Cargo.toml
│   └── src/
│       └── main.rs
├── mcp-blender-server/
│   ├── Cargo.toml
│   └── src/
│       └── main.rs
└── mcp-ymm4-server/
    ├── Cargo.toml
    └── src/
        └── main.rs

Success Criteria | 成功基準

Functional Requirements

  1. Agent to Skill Conversion

    • All 9 existing agents converted to skills
    • Skills auto-load when agent executes
    • Dual placement (.codex/skills/ and .cursor/skills/)
  2. New Skills

    • vrchat-dev skill functional
    • blender-cad skill functional
    • yukkuri-movie skill functional
    • yolo-auto skill functional
  3. Slash Commands

    • /VRChat command registered
    • /Blender command registered
    • /Yukkuri command registered
    • /YOLO command registered
  4. MCP Servers

    • vrchat MCP server builds and runs
    • blender MCP server builds and runs
    • yukkuri MCP server builds and runs

Quality Requirements

  1. Code Quality

    • Clippy warnings: 0
    • Test coverage: 65%+
    • Documentation: 100%
  2. Integration

    • All skills integrate with existing /qc, /git4d commands
    • Slash commands work in TUI
    • MCP tools available in Codex

Configuration | 設定

Update .codex/mcp-servers.yaml

mcpServers:
  # ... existing servers ...
  vrchat:
    command: uvx
    args:
      - "--from"
      - "mcp-vrchat-server"
      - "mcp-vrchat-server"
    env: {}
  blender:
    command: uvx
    args:
      - "--from"
      - "mcp-blender-server"
      - "mcp-blender-server"
    env: {}
  yukkuri:
    command: uvx
    args:
      - "--from"
      - "mcp-ymm4-server"
      - "mcp-ymm4-server"
    env: {}

References | 参考文献



Meta Prompt for Next Agent | 次のエージェントへのメタ指示

Current Status | 現在のステータス

You are continuing the Zapabob/Codex Custom Skills Implementation Project.

実装状況: All core infrastructure has been implemented.

You are powered by the model named minimax-m2.1-free. The exact model ID is opencode/minimax-m2.1-free.

Completed Items | 完了済み項目

Item Status
Slash Commands: /VRChat, /Blender, /Yukkuri, /YOLO ✅ Complete
MCP Servers: vrchat, blender, yukkuri, yolo-auto ✅ Complete
Skills: vrchat-dev, blender-cad, yukkuri-movie, yolo-auto, git4d-runtime, git4d-schema ✅ Complete
Implementation Logs in _docs/ ✅ Complete

What Did We Do | 做了什么

1. Planning Document: Created AGENT.md with comprehensive implementation plan for VRChat SDK3, Blender Python, YukkuriMovieMaker, OpenClaw integration.

2. Slash Commands: Added VrChat, Blender, Yukkuri, Yolo to slash_command.rs and handlers in chatwidget.rs.

3. MCP Server Configuration: Updated .codex/mcp-servers.yaml with 4 new servers.

4. MCP Servers Created:

Server Tools
mcp-vrchat-server vrchat_compile_udon, vrchat_upload_world, vrchat_configure_avatar, vrchat_setup_physbones, vrchat_create_contact
mcp-blender-server blender_create_geometry, blender_assign_material, blender_render, blender_export, blender_geometry_nodes
mcp-ymm4-server ymm4_create_scene, ymm4_add_character, ymm4_audio_effect, ymm4_video_effect, ymm4_render
mcp-yolo-server yolo_select_gpu, yolo_distribute_task, yolo_execute_workflow, yolo_aggregate_results, yolo_monitor_progress

5. Skill Definitions: Created SKILL.md in .codex/skills/ and .cursor/skills/.

6. Implementation Logs: _docs/2026-02-08_Skill実装計画{main}.md and _docs/2026-02-08_Skill実装{main}.md.


Next Tasks | 次のタスク

High Priority | 高優先度

  1. Agent YAML Conversion

    • Convert remaining .codex/agents/*.yaml to skill format
    • Files: performance-analyst.yaml, security-audit.yaml, etc.
  2. Build Verification

    cd mcp-servers/mcp-vrchat-server && cargo build
    cd mcp-servers/mcp-blender-server && cargo build
    cd mcp-servers/mcp-ymm4-server && cargo build
    cd mcp-servers/mcp-yolo-server && cargo build
  3. Test Implementation

    • Add unit tests for MCP servers
    • Add integration tests for slash commands

Medium Priority | 中優先度

  1. CI/CD Integration - Add MCP server builds to GitHub Actions
  2. Documentation - Add usage examples and tutorials
  3. Complete Agent Mapping - Finish agent_servers configuration

Key Files | 主要ファイル

File Purpose
codex-rs/tui/src/slash_command.rs Slash command definitions
codex-rs/tui/src/chatwidget.rs Command handlers
.codex/mcp-servers.yaml MCP server configuration
.codex/skills/*/SKILL.md Skill definitions
.codex/skills/*/agents/openai.yaml Agent configurations
mcp-servers/*/src/main.rs MCP server implementations
_docs/2026-02-08_Skill実装{main}.md Implementation details
_docs/2026-02-08_Skill実装計画{main}.md Planning log

Version Information | バージョン情報

  • Codex CLI: v2.14.1
  • Rust: 2024 Edition
  • Edition: zapabob/codex custom fork
  • Model: opencode/minimax-m2.1-free

Continue Implementation | 実装の継続

  1. Read AGENT.md (this file) for context
  2. Read _docs/2026-02-08_Skill実装{main}.md for implementation details
  3. Start with Agent YAML Conversion (highest priority)
  4. Run build verification after code changes

Success Criteria Checklist | 成功基準チェックリスト

  • All 9 existing agents converted to skills
  • vrchat-dev skill functional
  • blender-cad skill functional
  • yukkuri-movie skill functional
  • yolo-auto skill functional
  • /VRChat command registered
  • /Blender command registered
  • /Yukkuri command registered
  • /YOLO command registered
  • vrchat MCP server builds
  • blender MCP server builds
  • yukkuri MCP server builds
  • yolo MCP server builds
  • Clippy warnings: 0
  • Test coverage: 65%+

Japanese Prompt | 日本語プロンプト

Zapabob/Codexカスタムスキル実装の継続:

背景:
- VRChat SDK3、Blender Python、ゆっくりMovieMaker、OpenClaw統合のカスタムスキル
- バージョン:v2.14.1、Rust 2024 Edition
- 全てのコアインフラ実装済み

完了済み:
- ✅ スラッシュコマンド:/VRChat、/Blender、/Yukkuri、/YOLO
- ✅ MCPサーバー設定(.codex/mcp-servers.yaml)
- ✅ 4つのMCPサーバー:vrchat、blender、yukkuri、yolo-auto
- ✅ スキル定義(.codex/skills/、.cursor/skills/)
- ✅ 実装ログ(_docs/)

次のタスク(優先順):
1. 残りのagent YAMLファイルをスキル形式に変換
2. MCPサーバーのビルド検証
3. ユニットテスト追加
4. CI/CD統合

主要ファイル:
- AGENT.md(このファイル)
- _docs/2026-02-08_Skill実装{main}.md(実装詳細)
- codex-rs/tui/src/slash_command.rs(スラッシュコマンド)
- codex-rs/tui/src/chatwidget.rs(ハンドラー)
- .codex/mcp-servers.yaml(MCP設定)
- mcp-servers/*/src/main.rs(MCP実装)
- .codex/skills/*/SKILL.md(スキル定義)

最初はAGENT.mdを読んで文脈を理解してから、
Agent YAML変換(最高優先度)から始めてください。

Document Version: 2.0 Last Updated: 2026-02-08 Author: Zapabob/Codex Implementation Team For: Next agent continuing the implementation