「OpenAI Codex CLI」を試す

brew install codex

codex --version

codex-cli 0.40.0

codex --help

Codex CLI

If no subcommand is specified, options will be forwarded to the interactive CLI.

Usage: codex [OPTIONS] [PROMPT]
       codex [OPTIONS] [PROMPT] <COMMAND>

Commands:
  exec        Run Codex non-interactively [aliases: e]
  login       Manage login
  logout      Remove stored authentication credentials
  mcp         [experimental] Run Codex as an MCP server and manage MCP servers
  proto       Run the Protocol stream via stdin/stdout [aliases: p]
  completion  Generate shell completion scripts
  debug       Internal debugging commands
  apply       Apply the latest diff produced by Codex agent as a `git apply` to your local working tree [aliases: a]
  resume      Resume a previous interactive session (picker by default; use --last to continue the most recent)
  help        Print this message or the help of the given subcommand(s)

Arguments:
  [PROMPT]
          Optional user prompt to start the session

Options:
  -c, --config <key=value>
          Override a configuration value that would otherwise be loaded from `~/.codex/config.toml`. Use a dotted path
          (`foo.bar.baz`) to override nested values. The `value` portion is parsed as JSON. If it fails to parse as
          JSON, the raw string is used as a literal.

          Examples: - `-c model="o3"` - `-c 'sandbox_permissions=["disk-full-read-access"]'` - `-c
          shell_environment_policy.inherit=all`

  -i, --image <FILE>...
          Optional image(s) to attach to the initial prompt

  -m, --model <MODEL>
          Model the agent should use

      --oss
          Convenience flag to select the local open source model provider. Equivalent to -c model_provider=oss; verifies
          a local Ollama server is running

  -p, --profile <CONFIG_PROFILE>
          Configuration profile from config.toml to specify default options

  -s, --sandbox <SANDBOX_MODE>
          Select the sandbox policy to use when executing model-generated shell commands

          [possible values: read-only, workspace-write, danger-full-access]

  -a, --ask-for-approval <APPROVAL_POLICY>
          Configure when the model requires human approval before executing a command

          Possible values:
          - untrusted:  Only run "trusted" commands (e.g. ls, cat, sed) without asking for user approval. Will escalate
            to the user if the model proposes a command that is not in the "trusted" set
          - on-failure: Run all commands without asking for user approval. Only asks for approval if a command fails to
            execute, in which case it will escalate to the user to ask for un-sandboxed execution
          - on-request: The model decides when to ask the user for approval
          - never:      Never ask for user approval Execution failures are immediately returned to the model

      --full-auto
          Convenience alias for low-friction sandboxed automatic execution (-a on-failure, --sandbox workspace-write)

      --dangerously-bypass-approvals-and-sandbox
          Skip all confirmation prompts and execute commands without sandboxing. EXTREMELY DANGEROUS. Intended solely
          for running in environments that are externally sandboxed

  -C, --cd <DIR>
          Tell the agent to use the specified directory as its working root

      --search
          Enable web search (off by default). When enabled, the native Responses `web_search` tool is available to the
          model (no per‑call approval)

  -h, --help
          Print help (see a summary with '-h')

  -V, --version
          Print version

codex

tree -a ~/.codex

/Users/kun432/.codex
├── auth.json
├── log
│   └── codex-tui.log
├── sessions
│   └── 2025
│       └── 09
│           └── 24
│               └── rollout-2025-09-24T19-13-22-01997b36-a4ec-7532-adcc-0ca4d269172a.jsonl
└── version.json

git clone https://github.com/streamlit/streamlit-example && cd streamlit-example

codex

codex "このレポジトリの内容について日本語で説明して"

codex "このレポジトリの内容について日本語で説明して"

なるほど、ここでディレクトリごとにおそらく設定が行われるのだろうと思われる。逆に常に確認してほしい、みたいな設定はできるのかな？

CLIから対話モードを無効化して自動実行させることもできる。とりあえず先程の`config.toml`を一旦削除。

```shell
rm ~/.codex/config.toml

codex exec "このレポジトリの内容について日本語で説明して"

codex resume

codex resume --last

codex resume <セッションID>

mkdir ../todo-list-app-sample && cd $_

codex --full-auto "シンプルなToDoリストアプリを作成して。"

codex -i ~/Desktop/kobe.jpg "この画像について説明して。"

シンプルなToDoリストアプリを作成して。

ボタンの色を黄緑色にして。

背景をより薄い緑色にして。

cd  # ホームディレクトリに移動
codex --cd ~/work/streamlit-example  # --cd で作業ディレクトリ指定

codex debug --help

Internal debugging commands

Usage: codex debug [OPTIONS] <COMMAND>

Commands:
  seatbelt  Run a command under Seatbelt (macOS only)
  landlock  Run a command under Landlock+seccomp (Linux only)
  help      Print this message or the help of the given subcommand(s)

Options:
  -c, --config <key=value>
          Override a configuration value that would otherwise be loaded from `~/.codex/config.toml`. Use a dotted path
          (`foo.bar.baz`) to override nested values. The `value` portion is parsed as JSON. If it fails to parse as
          JSON, the raw string is used as a literal.

          Examples: - `-c model="o3"` - `-c 'sandbox_permissions=["disk-full-read-access"]'` - `-c
          shell_environment_policy.inherit=all`

  -h, --help
          Print help (see a summary with '-h')

codex login --api-key "<APIキー>"

- name: Update changelog via Codex
  run: |
    npm install -g @openai/codex
    codex login --api-key "${{ secrets.OPENAI_KEY }}"
    codex exec --full-auto "update CHANGELOG for next release"

# 最新のセッションを再開して、プロンプトで実行
codex exec "ボタンに絵文字を付けて" resume --last

# プロンプトを標準入力で渡すこともできる。末尾の '-' は付けない。
echo "ボタンに絵文字を付けて" | codex exec resume --last

# セッションIDで再開
codex exec resume XXXXXXXXX-XXXX-XXXX-XXXXXXXXXX "ボタンに絵文字を付けて"

codex exec --help

Run Codex non-interactively

Usage: codex exec [OPTIONS] [PROMPT] [COMMAND]

Commands:
  resume  Resume a previous session by id or pick the most recent with --last
  help    Print this message or the help of the given subcommand(s)
(snip)

codex exec "ボタンに絵文字を付けて" resume --last

[2025-09-25T06:15:06] codex

- I wasn’t able to write to the project files (macOS sandbox kept returning “Operation not permitted”), so the change is prepared but not applied yet.
- Edit `public/index.html:21` and replace the button labels as follows:

**Next Steps**
1. Update `public/index.html` with the snippet above.
2. Reload `public/index.html` in your browser and confirm all buttons now show the emoji labels. If the sandbox restrictions loosen later, let me know and I can apply the edit directly.

[2025-09-25T06:17:00] OpenAI Codex v0.40.0 (research preview)
--------
workdir: /Users/kun432/work/todo-list-app-sample
model: gpt-5-codex
provider: openai
approval: never
sandbox: read-only
reasoning effort: none
reasoning summaries: auto
--------

/status
📂 Workspace
  • Path: ~/work/todo-list-app-sample
  • Approval Mode: on-request
  • Sandbox: workspace-write
  • AGENTS files: (none)
(snip)

[mcp_servers.github_mcp]
command = "docker"
args = ["run", "-i", "--rm", "-e", "GITHUB_PERSONAL_ACCESS_TOKEN", "ghcr.io/github/github-mcp-server"]
env = { "GITHUB_PERSONAL_ACCESS_TOKEN" = "XXXXXXXXXX" }

codex mcp --help

[experimental] Run Codex as an MCP server and manage MCP servers

Usage: codex mcp [OPTIONS] [COMMAND]

Commands:
  serve   [experimental] Run the Codex MCP server (stdio transport)
  list    [experimental] List configured MCP servers
  get     [experimental] Show details for a configured MCP server
  add     [experimental] Add a global MCP server entry
  remove  [experimental] Remove a global MCP server entry
  help    Print this message or the help of the given subcommand(s)

Options:
  -c, --config <key=value>
          Override a configuration value that would otherwise be loaded from
          `~/.codex/config.toml`. Use a dotted path (`foo.bar.baz`) to override nested values.
          The `value` portion is parsed as JSON. If it fails to parse as JSON, the raw string
          is used as a literal.

          Examples: - `-c model="o3"` - `-c 'sandbox_permissions=["disk-full-read-access"]'` -
          `-c shell_environment_policy.inherit=all`

  -h, --help
          Print help (see a summary with '-h')