codex_opencode is a Codex-native workflow pack inspired by oh-my-openagent.

It does not build a new runtime or command shell. It adapts the collaboration method into native Codex primitives:

• project-scoped custom agents in .codex/agents/
• project-scoped workflow skills in .codex/skills/
• durable planning and execution state in .sisyphus/

• the original role topology, adapted into Codex custom agents: sisyphus, prometheus, metis, momus, atlas, hephaestus, sisyphus-junior, oracle, explore, librarian, multimodal-looker
• planning and execution separation
• plan files and execution notepads
• explicit workflow entrypoints instead of free-form agent drift
• high-signal multi-turn continuity for long tasks

• slash-command runtime
• background-task shell or tmux pane orchestration
• hook, plugin, and tool-registry internals from oh-my-openagent
• category-driven model routing and provider fallback runtime
• hashline editing, LSP, AST-Grep, and skill-MCP lifecycle management
• the full product shell around OpenCode

• $plan-work Create an execution-ready plan under .sisyphus/plans/
• $start-work Execute an existing plan through atlas
• $ultrawork Action-first execution with dynamic routing
• $review-work Findings-first review for plans, code, and config

• sisyphus Main orchestrator
• prometheus Strategic planner
• metis Pre-planning consultant
• momus Practical plan reviewer
• atlas Plan execution conductor
• hephaestus Autonomous deep worker
• sisyphus-junior Focused executor for small bounded slices
• oracle Architecture and risk consultant
• explore Local codebase explorer
• librarian Docs and OSS researcher
• multimodal-looker Read-only visual analyst

codex_opencode/
├── README.md
├── .codex/
│   ├── config.toml
│   ├── agents/
│   └── skills/
├── .sisyphus/
│   ├── README.md
│   ├── plans/
│   ├── notepads/
│   └── templates/
├── scripts/
│   ├── install-into-repo.zsh
│   ├── render-prompt.zsh
│   ├── smoke-agents.zsh
│   └── target-session.zsh
└── docs/

The default usage is embed-in-repo.

Use this when your goal is:

• copy the workflow into your project
• cd "/path/to/your-project"
• run codex
• use the workflow directly inside that repo

In this mode:

• .codex/ and the .sisyphus/ skeleton are the required pieces
• scripts are optional helpers, not runtime requirements
• plans and notepads live in the current repo .sisyphus/
• named custom-agent prompts are the most stable entry

Minimal startup:

cd "/path/to/your-project"
codex

Or install the core pack into an existing repo in one step:

cd "/Users/bytedance/PycharmProjects/test3/codex_opencode"
./scripts/install-into-repo.zsh "/path/to/your-project"

That installer now also:

• writes a managed .gitignore block so .codex/, .sisyphus/, and docs/codex-opencode/ are not accidentally ignored
• copies the usage docs into docs/codex-opencode/ inside the target repo

Recommended first prompt:

Use the custom agent named prometheus for this task.
Its registration file is .codex/agents/prometheus.toml.

Create an execution-ready .sisyphus plan for a very small, safe improvement in this repository.

Start here:

• docs/embed-in-repo.zh.md
• docs/usage-handbook.zh.md
• docs/operator-cheatsheet.zh.md

Use this only when your goal is:

• audit or operate against many target repos
• keep the active runtime separate from the target repo
• avoid routing drift when the target repo already contains its own workflow pack

In this mode:

• the harness workspace is the active runtime
• target-repo workflow files are usually evidence, not the active runtime
• helper scripts are useful but still optional

Read these docs if you intentionally choose that mode:

• docs/harness-playbook.zh.md
• docs/prompts.md

To verify the pack itself without embedding it elsewhere:

cd "/Users/bytedance/PycharmProjects/test3/codex_opencode"
codex exec --skip-git-repo-check -s read-only "Use the custom agent named prometheus for this task. Its registration file is .codex/agents/prometheus.toml. Explain how this workspace is structured and what it is for."

To verify that all 11 custom agents are actually recognized and spawnable by Codex CLI:

cd "/Users/bytedance/PycharmProjects/test3/codex_opencode"
./scripts/smoke-agents.zsh

The scripts/ directory is optional. It is mainly useful when you keep a separate workflow repo for harness-mode work.

• install-into-repo.zsh copies .codex/, .sisyphus/, writes a managed .gitignore block, and installs usage docs into docs/codex-opencode/
• render-prompt.zsh generates copy-paste prompt text for common harness flows
• target-session.zsh starts a harness Codex session with --add-dir
• smoke-agents.zsh validates agent discoverability in this repo

• docs/architecture.md
• docs/agent-smoke-matrix.md
• docs/agents.md
• docs/skills.md
• docs/prompts.md
• docs/embed-in-repo.zh.md
• docs/project-overview.zh.md
• docs/usage-handbook.zh.md
• docs/usage-guide.zh.md
• docs/operator-cheatsheet.zh.md
• docs/user-manual.md
• docs/harness-playbook.zh.md

This template was derived from:

• /Users/bytedance/PycharmProjects/test3/oh-my-openagent
• Codex custom-agent and skill capabilities available in the current Codex CLI environment