jeo

JEO - Integrated Agent Orchestration

Safety Notice

This listing is imported from skills.sh public index metadata. Review upstream SKILL.md and repository scripts before running.

Copy this and send it to your AI assistant to learn

Install skill "jeo" with this command: npx skills add jeo-tech-ai/oh-my-unity3d/jeo-tech-ai-oh-my-unity3d-jeo

JEO - Integrated Agent Orchestration

Keyword: jeo | Platforms: Claude Code, Codex CLI, Gemini CLI, OpenCode

Workflow: Plan (ralph

  • plannotator ) -> Execute (team or bmad ) -> Verify (agent-browser ) -> Cleanup (worktree-cleanup )

JEO is the release-oriented orchestration skill package shipped in this repository. It standardizes one path through planning, implementation, browser verification, and worktree cleanup across the supported AI coding tools.

agentation , annotate , and agentui are intentionally removed in v2.0.0 . UI review is handled with agent-browser snapshots and normal edit loops only.

  1. Execution Contract

Run the phases in order. Do not skip PLAN. Do not enter EXECUTE without an approved plan.

Unity3D 모드 (unity-mcp 감지)

unity-mcp 서버가 실행 중인 경우 (localhost:8080 연결 확인 시) Unity3D 검증 루프가 활성화됩니다.

STEP 0: Bootstrap

Create the working state directories:

mkdir -p .omc/state .omc/plans .omc/logs

If .omc/state/jeo-state.json does not exist, create it with:

{ "mode": "jeo", "phase": "plan", "task": "<detected task>", "plan_approved": false, "plan_gate_status": "pending", "team_available": null, "retry_count": 0, "last_error": null, "checkpoint": null, "created_at": "<ISO 8601>", "updated_at": "<ISO 8601>" }

User-facing activation message:

JEO activated. Phase: PLAN.

STEP 0.1: Recovery Rules

  • Update checkpoint whenever a phase starts.

  • Update last_error and increment retry_count before failing pre-flight.

  • Resume from .omc/state/jeo-state.json if a prior run already exists.

  • If retry_count >= 3 , stop and ask the user whether to continue.

  1. PLAN

PLAN is mandatory in every JEO run.

Required output

Write plan.md with:

  • goal

  • implementation steps

  • risks

  • completion criteria

Required plan gate

Run the blocking plannotator loop:

bash scripts/plannotator-plan-loop.sh plan.md /tmp/plannotator_feedback.txt 3

Rules:

  • Never use & .

  • If plannotator is missing, auto-run bash scripts/ensure-plannotator.sh first.

  • Proceed only when the result contains approved=true .

  • If the plan is rejected, update plan.md and run the loop again.

  • If the loop exits with 32 , use manual conversation approval and do not execute until the user explicitly approves.

When approved, update the state file:

{ "phase": "execute", "plan_approved": true, "plan_gate_status": "approved" }

  1. EXECUTE

JEO supports two execution paths.

Claude Code

Use team execution:

/omc:team 3:executor "<task>"

JEO expects CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 .

Codex, Gemini CLI, OpenCode

Use BMAD as the fallback orchestration path:

/workflow-init /workflow-status

Execution rule:

  • keep the implementation aligned to the approved plan

  • update phase to execute

  • return to PLAN only if the plan is invalidated by new information

  1. VERIFY

Use browser verification when the task includes UI or browser behavior.

Primary command:

agent-browser snapshot http://localhost:3000

Useful follow-ups:

agent-browser screenshot http://localhost:3000 -o verify.png playwriter --help

Verification rule:

  • confirm the changed behavior, not only page load

  • capture evidence before cleanup when the task is browser-facing

  • keep the state file in phase = "verify" until verification is complete

STEP 3: VERIFY — Unity3D 검증 루프

Unity3D 모드가 활성화된 경우 브라우저 검증 대신 아래 루프를 실행합니다.

① unity-mcp: run_tests → pass/fail 집계 ② unity-mcp: read_console → Error/Exception 패턴 탐지 ③ unity-mcp: editor_state → 씬 로드 상태 확인 ④ unity-mcp: find_gameobjects → 필수 오브젝트 확인

검증 결과:

  • 모두 통과 → CLEANUP

  • 실패 → Fix 루프 (code-refactoring 또는 unity-mcp: validate_script → 재검증, 최대 3회)

  • 3회 초과 → 사용자 확인 요청

jeo-state.json 업데이트:

{ "unity_verify": { "tests_passed": true, "console_errors": 0, "retry_count": 0 } }

  1. CLEANUP

After verification:

bash scripts/worktree-cleanup.sh || git worktree prune

Then set:

{ "phase": "done", "checkpoint": "cleanup" }

Cleanup rule:

  • warn before cleanup if there are uncommitted changes

  • clean extra worktrees only

  • never delete unrelated user work

  1. Unity3D 워크플로우 예제

Workflow 1: 게임 기획 → 씬 프로토타이핑

jeo "씬 프로토타이핑: <게임명>" [PLAN] bmad-gds-brainstorm-game → bmad-gds-gdd [EXECUTE] unity-mcp: manage_scene → manage_gameobject → manage_probuilder [VERIFY] unity-mcp: run_tests → read_console → editor_state → Fix 루프 (max 3) [CLEANUP]

Workflow 2: 스프린트 스토리 → C# 구현

jeo "스토리 구현: <스토리명>" [PLAN] bmad-gds-sprint-planning → bmad-gds-create-story [EXECUTE] bmad-gds-dev-story → unity-mcp: create_script → validate_script → script_apply_edits [VERIFY] unity-mcp: run_tests → read_console → Fix 루프 → bmad-gds-code-review [CLEANUP]

Workflow 3: 에셋 파이프라인 자동화

jeo "에셋 파이프라인: <에셋 종류>" [PLAN] file-organization [EXECUTE] unity-mcp: manage_asset → manage_texture → manage_material → manage_prefabs → batch_execute [VERIFY] unity-mcp: read_console → run_tests → performance-optimization → Fix 루프 [CLEANUP]

Workflow 4: Unity UI/비주얼 개발

design-system (Unity3D Design Guide 탐색) → ui-component-patterns → unity-mcp: manage_ui → manage_animation → manage_vfx → [VERIFY] unity-mcp: run_tests → read_console

Workflow 5: 성능 최적화 & 디버깅

jeo "성능 최적화: <증상>" [PLAN] log-analysis (unity-mcp: read_console) → find_gameobjects → codebase-search [EXECUTE] performance-optimization → unity-mcp: manage_components → batch_execute [VERIFY] unity-mcp: run_tests → read_console → bmad-gds-performance-test → Fix 루프 [CLEANUP]

  1. Quick Start

Install dependencies and helpers

bash scripts/install.sh --all

Check current environment

bash scripts/check-status.sh

Configure each platform

bash scripts/setup-claude.sh bash scripts/setup-codex.sh bash scripts/setup-gemini.sh bash scripts/setup-opencode.sh

  1. Installed Components

Tool Purpose

omc

Claude Code team orchestration

omx

OpenCode orchestration

ohmg

Gemini multi-agent support

bmad

Fallback workflow orchestration

ralph

specification-first planning loop

plannotator

blocking plan review gate

agent-browser

browser verification

playwriter

optional browser automation helper

  1. Platform Notes

Claude Code

  • setup-claude.sh enables team mode and plannotator review hooks.

  • JEO does not fall back to single-agent execution in Claude Code.

Codex CLI

  • setup-codex.sh writes developer_instructions , /prompts:jeo , and a PLAN_READY notify hook.

Gemini CLI

  • setup-gemini.sh installs a plannotator-oriented AfterAgent helper and appends JEO guidance to GEMINI.md .

OpenCode

  • setup-opencode.sh registers plugins and slash commands for plan, execute, verify, and cleanup.
  1. State File

Path:

.omc/state/jeo-state.json

Example:

{ "mode": "jeo", "phase": "verify", "task": "Ship the JEO release package", "plan_approved": true, "plan_gate_status": "approved", "team_available": false, "retry_count": 0, "last_error": null, "checkpoint": "verify", "created_at": "2026-03-09T00:00:00Z", "updated_at": "2026-03-09T00:00:00Z" }

  1. Troubleshooting

Problem Action

plannotator unavailable Run bash scripts/ensure-plannotator.sh or bash scripts/install.sh --with-plannotator

Claude team execution unavailable Run bash scripts/setup-claude.sh , restart Claude Code, confirm CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1

Codex prompt not found Re-run bash scripts/setup-codex.sh

Gemini hook missing Re-run bash scripts/setup-gemini.sh

Browser verification unavailable Install agent-browser with bash scripts/install.sh --with-browser

Worktrees remain after completion Run bash scripts/worktree-cleanup.sh and then git worktree prune

  1. Release Notes

v2.0.0

  • removed agentation integration from the skill contract

  • removed annotate and agentui workflow branches

  • simplified verification around agent-browser

  • aligned setup scripts and docs to the reduced release surface

Source Transparency

This detail page is rendered from real SKILL.md content. Trust labels are metadata-based hints, not a safety guarantee.

Related Skills

Related by shared tags or category signals.

Automation

git-workflow

No summary provided by upstream source.

Repository SourceNeeds Review
General

unity-mcp

No summary provided by upstream source.

Repository SourceNeeds Review
General

api-documentation

No summary provided by upstream source.

Repository SourceNeeds Review
General

plannotator

No summary provided by upstream source.

Repository SourceNeeds Review