🔌 Codex on a Mac: wire it to REA, GitHub, NotebookLM and a team of agents · Codex บนแมค: ต่อมันเข้ากับ REA, GitHub, NotebookLM และทีมเอเจนต์ · 在 Mac 上接好 Codex：REA、GitHub、NotebookLM 与一支智能体团队

1 Install Codex · ติดตั้ง Codex · 安装 Codex

node --version
npm install -g @openai/codex
codex --version

what the last command printedสิ่งที่คำสั่งสุดท้ายพิมพ์ออกมา最后一条命令打印的内容

codex-cli 0.163.0

2 Sign in, and decide what Codex may do · ลงชื่อเข้าใช้ และกำหนดว่า Codex ทำอะไรได้บ้าง · 登录，并决定 Codex 可以做什么

codex
printenv OPENAI_API_KEY | codex login --with-api-key
codex login status

what the last command printed, on a Mac where nobody signed inสิ่งที่คำสั่งสุดท้ายพิมพ์ออกมา บนแมคที่ไม่มีใครลงชื่อเข้าใช้最后一条命令打印的内容，在一台没有人登录的 Mac 上

Not logged in

3 The three places Codex reads · สามที่ที่ Codex อ่าน · Codex 读取的三个地方

codex --help

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

Codex CLI

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

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

Commands:
  agents            Browse all agent sessions on the shared local app-server daemon
  exec              Run Codex non-interactively [aliases: e]
  review            Run a code review non-interactively
  login             Manage login
  logout            Remove stored authentication credentials
  mcp               Manage external MCP servers for Codex
  plugin            Manage Codex plugins
  app-server        [experimental] Run the app server or related tooling
  remote-control    [experimental] Manage the app-server daemon with remote control enabled
  app               Launch the Desktop app (opens the app installer if missing)
  completion        Generate shell completion scripts
  update            Update Codex to the latest version
  doctor            Diagnose local Codex installation, config, auth, and runtime health
  sandbox           Run commands within a Codex-provided sandbox
  debug             Debugging tools
  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)
  queue             Queue a message for an existing session
  archive           Archive a saved session by id or session name
  delete            Permanently delete a saved session by id or session name
  migrate-rollouts  Inspect or migrate legacy local sessions to paginated thread history
  unarchive         Unarchive a saved session by id or session name
  fork              Fork a previous interactive session (picker by default; use --last to fork the
                    most recent)
  cloud             [EXPERIMENTAL] Browse tasks from Codex Cloud and apply changes locally
  exec-server       [EXPERIMENTAL] Run the standalone exec-server service
  features          Inspect feature flags
  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 TOML. If it fails to parse as TOML, 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`

      --enable <FEATURE>
          Enable a feature (repeatable). Equivalent to `-c features.<name>=true`

      --disable <FEATURE>
          Disable a feature (repeatable). Equivalent to `-c features.<name>=false`

      --remote <ADDR>
          Connect the TUI to a remote app server endpoint.
          
          Accepted forms: `ws://host:port`, `wss://host:port`, `unix://`, or `unix://PATH`.

      --remote-auth-token-env <ENV_VAR>
          Name of the environment variable containing the bearer token to send to a remote app
          server websocket

      --strict-config
          Error out when config.toml contains fields that are not recognized by this version of
          Codex

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

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

      --oss
          Use open-source provider

      --local-provider <OSS_PROVIDER>
          Specify which local provider to use (lmstudio or ollama). If not specified with --oss,
          will use config default or show selection

  -p, --profile <CONFIG_PROFILE_V2>
          Layer $CODEX_HOME/<name>.config.toml on top of the base user config

  -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]

      --approve-for-me
          Route approval requests through automatic review using the workspace-write sandbox

      --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

      --dangerously-bypass-hook-trust
          Run enabled hooks without requiring persisted hook trust for this invocation. DANGEROUS.
          Intended only for automation that already vets hook sources

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

      --worktree
          Run the session in a new managed Git worktree

      --add-dir <DIR>
          Additional directories that should be writable alongside the primary workspace

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

          Possible values:
          - 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

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

      --no-alt-screen
          Disable alternate screen mode
          
          Runs the TUI in inline mode, preserving terminal scrollback history.

      --no-daemon
          Run without the shared background server, even if it is already running

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

  -V, --version
          Print version

AGENTS.md

# AGENTS.md: how to work in this folder

This folder is a lab for studying Mac apps that were downloaded and not installed.
The project's NotebookLM notebook id is NOTEBOOK_ID (replace with your own).

## Tools

- Use the REA tools to read an app. Never start the app you study.
- Use the `notebooklm` command to file reports and ask the notebook questions.
  Always pass `--notebook NOTEBOOK_ID --json`.
- Commit with `git`; open pull requests with `gh pr create`. Never push to main.

## Delegation

- A task that reads several apps, or several parts of one app, is split: spawn
  one `reader` subagent per app or part, wait for all of them, then write one
  summary.
- Spawn the `librarian` subagent to file a finished report in the notebook.
- Before a pull request, spawn the `reviewer` subagent on the branch and fix
  what it finds.
- Do not run two subagents that write to the same file.

## Code Review Rules

- Flag any script that runs or installs the app it studies.
- Flag any file that prints a path inside the home folder, or a token.

4 REA: a tool server · REA: เซิร์ฟเวอร์เครื่องมือ · REA：一个工具服务器

codex mcp --help

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

Manage external MCP servers for Codex

Usage: codex mcp [OPTIONS] <COMMAND>

Commands:
  list    
  get     
  add     
  remove  
  login   
  logout  
  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 TOML. If it fails to parse as TOML, 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`

      --enable <FEATURE>
          Enable a feature (repeatable). Equivalent to `-c features.<name>=true`

      --disable <FEATURE>
          Disable a feature (repeatable). Equivalent to `-c features.<name>=false`

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

python3 agent_setup.py codex --yes

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

agent   codex
status  ready

create Codex
   file  ~/.codex/config.toml
   what  Add the REA MCP registration for Codex; preserve unrelated configuration. Command: "npx" "-y" "rea-agents@6.3.0" "mcp". No additional environment variables.

install REA reverse-engineering skill
   file  ~/.agents/skills/reverse-engineer-anything
   what  Install or update the bundled REA reverse-engineering skill and on-demand references.

done    configured_codex, installed_skill
backup  none

5 GitHub: three ways in · GitHub: สามทางเข้า · GitHub：三条路

not run by usเราไม่ได้รันเอง我们没有运行过

brew install gh
gh auth login

gh auth status

what it printed, on a Mac where nobody signed inสิ่งที่มันพิมพ์ออกมา บนแมคที่ไม่มีใครลงชื่อเข้าใช้它打印的内容，在一台没有人登录的 Mac 上

You are not logged into any GitHub hosts. To log in, run: gh auth login

codex mcp add github --url https://api.githubcopilot.com/mcp/ --bearer-token-env-var GITHUB_TOKEN

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

Added global MCP server 'github'.

codex mcp list

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

Name  Command  Args                     Env  Cwd  Status   Auth       
rea   npx      -y rea-agents@6.3.0 mcp  -    -    enabled  Unsupported

Name    Url                                 Bearer Token Env Var  Status   Auth        
github  https://api.githubcopilot.com/mcp/  GITHUB_TOKEN          enabled  Bearer token

6 NotebookLM: a command and its skill · NotebookLM: คำสั่งหนึ่งและสกิลของมัน · NotebookLM：一条命令和它的技能

python3 -m venv venv
venv/bin/python -m pip install notebooklm-py==0.4.0

not run by us: it opens a browser and waits for youเราไม่ได้รันเอง: มันเปิดเบราว์เซอร์แล้วรอคุณ我们没有运行过：它会打开浏览器并等你

venv/bin/python -m notebooklm login

venv/bin/python -m notebooklm skill install --target agents
venv/bin/python -m notebooklm skill status

what the two commands printedสิ่งที่สองคำสั่งพิมพ์ออกมา两条命令打印的内容

Installed NotebookLM skill
  Version: 0.4.0
  Scope:   user
  Agent Skills: /Users/runner/.agents/skills/notebooklm/SKILL.md

NotebookLM commands are now available in the selected skill directories.

NotebookLM skill status (user scope)
  CLI version: 0.4.0
  Claude Code: Not installed
    Path: /Users/runner/.claude/skills/notebooklm/SKILL.md
  Agent Skills: Installed
    Path: /Users/runner/.agents/skills/notebooklm/SKILL.md
    Skill version: 0.4.0

venv/bin/python -m notebooklm agent show codex

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

# Repository Guidelines

**Status:** Active
**Last Updated:** 2026-05-09

## Project Structure & Module Organization

`src/notebooklm/` contains the async client and typed APIs. Internal feature 
modules use `_` prefixes such as `_sources.py` and `_artifacts.py`; 
`src/notebooklm/cli/` holds Click commands, and `src/notebooklm/rpc/` handles 
protocol encoding and decoding. Tests are split by scope: `tests/unit/`, 
`tests/integration/`, and `tests/e2e/`. Recorded HTTP fixtures live in 
`tests/cassettes/`. Examples are in `docs/examples/`, and diagnostics live in 
`scripts/`.

## Build, Test, and Development Commands

Use `uv` for local work:

```bash
uv sync --extra dev --extra browser
uv run pytest
uv run ruff check src/ tests/
uv run ruff format src/ tests/
uv run mypy src/notebooklm
uv run pre-commit run --all-files
```

Run `uv run pytest tests/e2e -m readonly` only after `notebooklm login` and 
setting test notebook env vars.

## Coding Style & Naming Conventions

Target Python 3.10+, 4-space indentation, and double quotes. Ruff enforces 
formatting and import order with a 100-character line length. Keep module and 
test file names in `snake_case`; prefer descriptive Click command names that 
match existing groups such as `source`, `artifact`, and `research`. Preserve the
internal/public split: `_*.py` for implementation, exported types in 
`src/notebooklm/__init__.py`.

## Testing Guidelines

Put pure logic in `tests/unit/`, VCR-backed flows in `tests/integration/`, and 
authenticated NotebookLM coverage in `tests/e2e/`. Name tests 
`test_<behavior>.py` and record cassettes with `NOTEBOOKLM_VCR_RECORD=1 uv run 
pytest tests/integration/test_vcr_*.py -v`. Coverage is expected to stay at or 
above the configured 90% threshold.

## Commit, PR, and Agent Notes

Follow the existing commit style: `feat(cli): ...`, `fix(cli): ...`, 
`refactor(test): ...`, `style: ...`. PRs should include a short summary, linked 
issue when relevant, and the commands run locally.

For Codex or other parallel agents:

- Prefer `--json` output and pass explicit notebook IDs instead of relying on 
`notebooklm use`.
- Isolate concurrent runs with `NOTEBOOKLM_PROFILE=agent-<id>` so each agent 
gets its own context file under `~/.notebooklm/profiles/<name>/`. Fall back to 
`NOTEBOOKLM_HOME=/tmp/agent-<id>` only when separate home directories are 
required.
- In headless environments where Playwright login is impractical, authenticate 
with `notebooklm login --browser-cookies <browser>` (requires `pip install 
"notebooklm-py"`).

venv/bin/python -m notebooklm ask "In two short paragraphs: where does Codex look for skills on a Mac, and how does a session call one by name?" --notebook a156862e-29f0-493d-9759-cf0cad03d76a

what it printed, on our laptopสิ่งที่มันพิมพ์ออกมา บนแล็ปท็อปของเรา它打印的内容，在我们的笔记本电脑上

Different notebook specified, starting new conversation...
Continuing conversation 382e466f...
Answer:
On a Mac, Codex discovers skills across four hierarchical scopes: 
**repository-level** `.agents/skills` directories (scanned from the current 
working directory `$CWD` up to the repository root `$REPO_ROOT`), 
**user-level** `$HOME/.agents/skills` in the user's home folder, 
**admin-level** `/etc/codex/skills` for machine-wide system defaults, and 
**system-level** built-in skills bundled directly with Codex by OpenAI [1, 2].

A session calls a skill by name via **explicit invocation** by typing `$` 
followed by the skill name (or opening the `/skills` menu) in the Codex CLI or 
IDE extension [3]. Codex can also trigger a skill via **implicit invocation** 
by automatically matching a user prompt against the skill's `description` 
metadata defined in its `SKILL.md` file [3], after which it uses progressive 
disclosure to load the full instruction file into context [4].

💡 Would you like help writing a custom `SKILL.md` template or configuring 
local skill overrides?

Resumed conversation: 382e466f-21ac-4f11-a2d7-181bb2c0f1a5

7 A skill of your own · สกิลของคุณเอง · 你自己的技能

skills/mac-app-report/SKILL.md

---
name: mac-app-report
description: Write a short report on a Mac app that was downloaded but not installed, using the REA tools, and file it in the project's NotebookLM notebook. Use when asked to "report on this app", "study this .app" or "what is this app built on".
---

# Report on a Mac app

You are given the path of a `.app` folder. Do not start the app. Use the REA
tools to read it, and the `notebooklm` command to file what you wrote.

## Steps

1. `inspect-plist` on `Contents/Info.plist`: the app's name, identifier, version
   and the macOS it needs.
2. `inspect-signature`: who signed it, and the chain up to Apple's root.
3. `inspect-macho` on the program file named in the plist: the processors it is
   built for and the libraries it loads. Count the libraries that come from
   macOS and the ones the app brings along.
4. `decode-interface-builder` on each `.nib` under `Contents/Resources`: the
   windows and menus, and which code each menu item calls.
5. Write `report.md` beside the app, with one heading per step above. Under
   each heading, say what you found and which tool showed it. End with a list
   of what you could not find.
6. Add the report to the notebook: `notebooklm source add report.md --notebook
   NOTEBOOK_ID --json`, where NOTEBOOK_ID is in the project's `AGENTS.md`.
   Report the source id it returns.

## Rules

- Read only. Never edit the app, never run it, never run `open` on it.
- If a tool says a result was cut short, say so in the report.
- Quote the licence file of the app if there is one; stop and ask if it forbids
  reverse engineering.

skill_setup.py

# skill_setup.py - put this page's skill where Codex looks for skills, then list them.
#
#     python3 skill_setup.py       (in the folder where you saved this file)
#
# A skill is a folder with a file named SKILL.md in it.
# Codex looks for your skills in the folder ~/.agents/skills.

import pathlib
import shutil

# Path.home() is your home folder, ~. The / joins folder names into a path.
skills = pathlib.Path.home() / ".agents" / "skills"
name = "mac-app-report"

# Copy the skill's folder from skills/, beside this file, into Codex's folder.
# dirs_exist_ok=True means: if it is there already, copy over it.
shutil.copytree("skills/" + name, skills / name, dirs_exist_ok=True)
print("copied", name, "into ~/.agents/skills")

# Every folder in ~/.agents/skills is one skill that Codex will find.
print("Skills Codex will find:")
for folder in sorted(skills.iterdir()):
    print("  " + folder.name)

python3 skill_setup.py

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

copied mac-app-report into ~/.agents/skills
Skills Codex will find:
  mac-app-report
  notebooklm
  reverse-engineer-anything

# Quiz 7: what will this print?
names = ["notebooklm", "mac-app-report", "reverse-engineer-anything"]

for name in sorted(names):
    print(name)

mac-app-report
notebooklm
reverse-engineer-anything

8 A browser: Playwright and CloakBrowser · เบราว์เซอร์: Playwright และ CloakBrowser · 浏览器：Playwright 和 CloakBrowser

npx -y @playwright/mcp@0.0.83 --help

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

Usage: Playwright MCP [options]

Options:
  -V, --version                         output the version number
  --allowed-hosts <hosts...>            comma-separated list of hosts this
                                        server is allowed to serve from.
                                        Defaults to the host the server is bound
                                        to. Pass '*' to disable the host check.
  --allowed-origins <origins>           semicolon-separated list of TRUSTED
                                        origins to allow the browser to request.
                                        Default is to allow all.
                                        Important: *does not* serve as a
                                        security boundary and *does not* affect
                                        redirects.
  --allow-unrestricted-file-access      allow access to files outside of the
                                        workspace roots. Also allows
                                        unrestricted access to file:// URLs. By
                                        default access to file system is
                                        restricted to workspace root directories
                                        (or cwd if no roots are configured)
                                        only, and navigation to file:// URLs is
                                        blocked.
  --blocked-origins <origins>           semicolon-separated list of origins to
                                        block the browser from requesting.
                                        Blocklist is evaluated before allowlist.
                                        If used without the allowlist, requests
                                        not matching the blocklist are still
                                        allowed.
                                        Important: *does not* serve as a
                                        security boundary and *does not* affect
                                        redirects.
  --block-service-workers               block service workers
  --browser <browser>                   browser or chrome channel to use,
                                        possible values: chrome, firefox,
                                        webkit, msedge.
  --caps <caps>                         comma-separated list of additional
                                        capabilities to enable, possible values:
                                        vision, pdf, devtools.
  --cdp-endpoint <endpoint>             CDP endpoint to connect to.
  --cdp-header <headers...>             CDP headers to send with the connect
                                        request, multiple can be specified.
  --cdp-timeout <timeout>               timeout in milliseconds for connecting
                                        to CDP endpoint, defaults to 30000ms
  --codegen <lang>                      specify the language to use for code
                                        generation, possible values:
                                        "typescript", "python", "java",
                                        "csharp", "none". Default is
                                        "typescript".
  --config <path>                       path to the configuration file.
  --console-level <level>               level of console messages to return:
                                        "error", "warning", "info", "debug".
                                        Each level includes the messages of more
                                        severe levels.
  --device <device>                     device to emulate, for example: "iPhone
                                        15"
  --mobile                              emulate a generic mobile device (Pixel
                                        10 for Chromium, iPhone 17 for WebKit).
                                        Mobile pages are usually lighter, which
                                        saves tokens. Cannot be combined with
                                        --device.
  --executable-path <path>              path to the browser executable.
  --extension                           Connect to a running browser instance
                                        (Edge/Chrome only). Requires the
                                        "Playwright Extension" to be installed.
  --endpoint <endpoint>                 Bound browser endpoint to connect to.
  --file-paths <mode>                   how file paths are rendered in tool
                                        results, "relative" to the workspace
                                        root or "absolute". Default is
                                        "relative".
  --grant-permissions <permissions...>  List of permissions to grant to the
                                        browser context, for example
                                        "geolocation", "clipboard-read",
                                        "clipboard-write".
  --headless                            run browser in headless mode, headed by
                                        default
  --host <host>                         host to bind server to. Default is
                                        localhost. Use 0.0.0.0 to bind to all
                                        interfaces.
  --idle-timeout <timeout>              close the browser after this many
                                        milliseconds without a completed tool
                                        call, the next tool call relaunches it.
                                        Defaults to one hour for headless
                                        browsers, never for headed ones, 0
                                        disables.
  --ignore-https-errors                 ignore https errors
  --init-page <path...>                 path to TypeScript file to evaluate on
                                        Playwright page object
  --init-script <path...>               path to JavaScript file to add as an
                                        initialization script. The script will
                                        be evaluated in every page before any of
                                        the page's scripts. Can be specified
                                        multiple times.
  --isolated                            keep the browser profile in memory, do
                                        not save it to disk.
  --image-responses <mode>              whether to send image responses to the
                                        client. Can be "allow", "omit" or
                                        "only". With "only", a response that
                                        carries an image consists of the image
                                        parts alone, without the text part.
                                        Defaults to "allow".
  --no-sandbox                          disable the sandbox for all process
                                        types that are normally sandboxed.
  --no-webmcp                           do not collect or expose the tools that
                                        a page registers through the WebMCP API.
  --output-dir <path>                   path to the directory for automatically
                                        named output files, for example a
                                        screenshot taken without an explicit
                                        file name. Files with an explicit name
                                        are resolved against the workspace root
                                        instead and are not affected by this
                                        option.
  --output-max-size <bytes>             Threshold for evicting old output files,
                                        in bytes.
  --port <port>                         port to listen on for SSE transport.
  --profile-dir-name <name>             name of the profile directory in the
                                        user data dir to connect to with
                                        --extension, for example "Profile 1".
                                        Defaults to the last used profile that
                                        has the extension installed.
  --proxy-bypass <bypass>               comma-separated domains to bypass proxy,
                                        for example
                                        ".com,chromium.org,.domain.com"
  --proxy-server <proxy>                specify proxy server, for example
                                        "http://myproxy:3128" or
                                        "socks5://myproxy:8080"
  --sandbox                             enable the sandbox for all process types
                                        that are normally not sandboxed.
  --save-session                        Whether to save the Playwright MCP
                                        session into the output directory.
  --secrets <path>                      path to a file containing secrets in the
                                        dotenv format
  --shared-browser-context              reuse the same browser context between
                                        all connected HTTP clients.
  --snapshot-boxes                      include each element's bounding box as
                                        [box=x,y,width,height] in snapshots.
                                        Coordinates are viewport-relative, in
                                        CSS pixels.
  --snapshot-mode <mode>                when taking snapshots for responses,
                                        specifies the mode to use. Can be "full"
                                        or "none". Default is "full".
  --storage-state <path>                path to the storage state file for
                                        isolated sessions.
  --test-id-attribute <attribute>       specify the attribute to use for test
                                        ids, defaults to "data-testid"
  --timeout-action <timeout>            specify action timeout in milliseconds,
                                        defaults to 5000ms
  --timeout-navigation <timeout>        specify navigation timeout in
                                        milliseconds, defaults to 60000ms
  --timeout-settle <timeout>            how long to wait after each action for
                                        triggered work to settle, in
                                        milliseconds, defaults to 500ms
  --user-agent <ua string>              specify user agent string
  --user-data-dir <path>                path to the user data directory. If not
                                        specified, a temporary directory will be
                                        created.
  --viewport-size <size>                specify browser viewport size in pixels,
                                        for example "1280x720"
  -h, --help                            display help for command

codex mcp add playwright -- npx -y @playwright/mcp@0.0.83

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

Added global MCP server 'playwright'.

venv/bin/python -m pip install cloakbrowser==0.6.3
venv/bin/python -m cloakbrowser install

what the second command printedสิ่งที่คำสั่งที่สองพิมพ์ออกมา第二条命令打印的内容

Stealth Chromium 145.0.7632.109.2 not found. Downloading for darwin-arm64...
Downloading from https://cloakbrowser.dev/chromium-v145.0.7632.109.2/cloakbrowser-darwin-arm64.tar.gz
Download progress: 9% (12/140 MB)
Download progress: 19% (26/140 MB)
Download progress: 29% (40/140 MB)
Download progress: 39% (54/140 MB)
Download progress: 49% (68/140 MB)
Download progress: 59% (82/140 MB)
Download progress: 69% (96/140 MB)
Download progress: 79% (111/140 MB)
Download progress: 89% (125/140 MB)
Download progress: 99% (139/140 MB)
Download complete: 140 MB
SHA256SUMS signature verified: Ed25519 OK
Checksum verified: SHA-256 OK
Extracting to /Users/runner/.cloakbrowser/chromium-145.0.7632.109.2
Binary ready: /Users/runner/.cloakbrowser/chromium-145.0.7632.109.2/Chromium.app/Contents/MacOS/Chromium

  CloakBrowser - stealth Chromium for automation
  https://github.com/CloakHQ/CloakBrowser

  Running the free binary (v146). The latest binary (v154) is free too, with 1 concurrent session.
  Get your key: run  cloakbrowser login  or visit https://cloakbrowser.dev/free
  For more than one concurrent session -> https://cloakbrowser.dev
  Star us if CloakBrowser helps your project!

/Users/runner/.cloakbrowser/chromium-145.0.7632.109.2/Chromium.app/Contents/MacOS/Chromium

browser_setup.py

# browser_setup.py - a settings file that makes Playwright's tool server drive CloakBrowser.
#
#     venv/bin/python browser_setup.py
#
# Run it with the Python in which cloakbrowser is installed (step 8).

import json
import pathlib
from cloakbrowser import ensure_binary, get_default_stealth_args
from cloakbrowser.config import IGNORE_DEFAULT_ARGS

# ensure_binary() downloads CloakBrowser the first time, and gives its path.
# The two lists are switches: the ones to start the browser with,
# and the ones of Playwright's own to leave out.
switches = get_default_stealth_args()
settings = {
    "browser": {
        "browserName": "chromium",
        "launchOptions": {
            "executablePath": ensure_binary(),
            "args": switches,
            "ignoreDefaultArgs": IGNORE_DEFAULT_ARGS,
        },
    },
}

path = pathlib.Path.home() / ".codex" / "cloakbrowser.json"
path.parent.mkdir(exist_ok=True)
# json.dumps turns the dictionary into text; indent=2 puts one setting on each line.
path.write_text(json.dumps(settings, indent=2))
print("adds", " ".join(switches))
print("leaves out", " ".join(IGNORE_DEFAULT_ARGS))
print("wrote ~/.codex/cloakbrowser.json")

venv/bin/python browser_setup.py

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

adds --no-sandbox --fingerprint=14259 --fingerprint-platform=windows
leaves out --enable-automation --enable-unsafe-swiftshader
wrote ~/.codex/cloakbrowser.json

codex mcp add cloakbrowser -- npx -y @playwright/mcp@0.0.83 --config ~/.codex/cloakbrowser.json
codex mcp list

what the two commands printedสิ่งที่สองคำสั่งพิมพ์ออกมา两条命令打印的内容

Added global MCP server 'cloakbrowser'.

Name          Command  Args                                                                       Env  Cwd  Status   Auth       
cloakbrowser  npx      -y @playwright/mcp@0.0.83 --config /Users/runner/.codex/cloakbrowser.json  -    -    enabled  Unsupported
playwright    npx      -y @playwright/mcp@0.0.83                                                  -    -    enabled  Unsupported
rea           npx      -y rea-agents@6.3.0 mcp                                                    -    -    enabled  Unsupported

Name    Url                                 Bearer Token Env Var  Status   Auth        
github  https://api.githubcopilot.com/mcp/  GITHUB_TOKEN          enabled  Bearer token

mcp.py

# mcp.py - talk to a tool server the way Codex does. The files of steps 8 to 10 import it.
#
# A tool server is a program that Codex starts. Codex sends it one line of JSON
# for each request, and the server answers with one line of JSON.

import json
import subprocess


def start(command):
    # Start the server, with its input and output joined to this program.
    server = subprocess.Popen(command, stdin=subprocess.PIPE, stdout=subprocess.PIPE,
                              text=True, encoding="utf-8")
    # The first request says hello: which version of the protocol we speak.
    ask(server, "initialize", {"protocolVersion": "2025-06-18", "capabilities": {},
                               "clientInfo": {"name": "student", "version": "1.0"}})
    send(server, {"jsonrpc": "2.0", "method": "notifications/initialized"})
    return server


def send(server, message):
    # One message is one line of JSON.
    server.stdin.write(json.dumps(message) + "\n")
    server.stdin.flush()


def ask(server, method, params):
    send(server, {"jsonrpc": "2.0", "id": 1, "method": method, "params": params})
    # Read lines until the answer comes: it is the line that has an "id".
    while True:
        answer = json.loads(server.stdout.readline())
        if "id" in answer:
            return answer["result"]


def use(server, tool, arguments):
    # Use one of the server's tools, and give back the text it answers.
    result = ask(server, "tools/call", {"name": tool, "arguments": arguments})
    return result["content"][0]["text"]


def stop(server):
    server.stdin.close()
    server.wait()

browser_check.py

# browser_check.py - use the two browser tool servers of step 8, the way Codex does.
#
#     python3 browser_check.py
#
# Each server opens a web page and reads two things from it: navigator.webdriver,
# which is true when a program drives the browser, and the browser's own name
# for itself, its user agent. Any website can read both.

import pathlib
from mcp import start, ask, use, stop

settings = pathlib.Path.home() / ".codex" / "cloakbrowser.json"
playwright = ["npx", "-y", "@playwright/mcp@0.0.83", "--headless"]

# A dictionary: each server's name, and the command that starts it.
servers = {
    "playwright": playwright,
    "cloakbrowser": playwright + ["--config", str(settings)],
}

# A little JavaScript for the page to run. What it returns comes back to us.
look = "() => ({ webdriver: navigator.webdriver, userAgent: navigator.userAgent })"

for name, command in servers.items():
    print("SERVER", name)
    server = start(command)
    tools = ask(server, "tools/list", {})["tools"]
    print(len(tools), "tools")
    use(server, "browser_navigate", {"url": "https://example.com"})
    print(use(server, "browser_evaluate", {"function": look}))
    use(server, "browser_close", {})
    stop(server)
    print()

python3 browser_check.py

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

SERVER playwright
25 tools
### Result
{
  "webdriver": false,
  "userAgent": "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) HeadlessChrome/156.0.0.0 Safari/537.36"
}
### Ran Playwright code
```js
await page.evaluate('() => ({ webdriver: navigator.webdriver, userAgent: navigator.userAgent })');
```

SERVER cloakbrowser
25 tools
### Result
{
  "webdriver": false,
  "userAgent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/146.0.0.0 Safari/537.36"
}
### Ran Playwright code
```js
await page.evaluate('() => ({ webdriver: navigator.webdriver, userAgent: navigator.userAgent })');
```

browser_search.py

# browser_search.py - search Google with the two browser tool servers, as Codex would.
#
#     python3 browser_search.py
#
# Google asks programs not to search it (its robots.txt says so), so this file
# sends only one search with each server. Do not put it in a loop.

import pathlib
from mcp import start, use, stop

settings = pathlib.Path.home() / ".codex" / "cloakbrowser.json"
playwright = ["npx", "-y", "@playwright/mcp@0.0.83", "--headless"]
servers = {
    "cloakbrowser": playwright + ["--config", str(settings)],
    "playwright": playwright,
}

words = "Wat Phra That Doi Suthep"
# In a web address a space is written as +.
address = "https://www.google.com/search?hl=en&q=" + words.replace(" ", "+")

# JavaScript for Google's page: the address the browser ended on,
# and the titles of the first five results.
read = """() => [location.href.split('?')[0],
  ...[...document.querySelectorAll('#search a h3')].slice(0, 5).map(h => h.innerText)]"""

for name, command in servers.items():
    print("SERVER", name)
    server = start(command)
    use(server, "browser_navigate", {"url": address})
    answer = use(server, "browser_evaluate", {"function": read})
    # The answer has sections; the value we asked for comes after "### Result".
    print(answer.split("### Result")[1].split("###")[0].strip())
    use(server, "browser_close", {})
    stop(server)
    print()

python3 browser_search.py

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

SERVER cloakbrowser
[
  "https://www.google.com/sorry/index"
]

SERVER playwright
[
  "https://www.google.com/sorry/index"
]

9 Chrome DevTools: what a page did · Chrome DevTools: หน้าเว็บทำอะไรไปบ้าง · Chrome DevTools：网页做了什么

codex mcp add chrome-devtools -- npx -y chrome-devtools-mcp@1.11.0 --no-usage-statistics

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

Added global MCP server 'chrome-devtools'.

devtools_check.py

# devtools_check.py - ask Chrome DevTools about a web page, through its tool server.
#
#     python3 devtools_check.py https://krueng.ai/codex_mac.html
#
# It opens the page, then asks three of the server's tools about it:
# the page's console messages, its network requests, and a Lighthouse audit.

import sys
from mcp import start, use, stop

# sys.argv is the list of words you typed: sys.argv[1] is the page's address.
page = sys.argv[1]

# --isolated keeps Chrome's profile in memory only: nothing is saved to disk.
server = start(["npx", "-y", "chrome-devtools-mcp@1.11.0", "--headless", "--isolated",
                "--no-usage-statistics"])
opened = use(server, "new_page", {"url": page})

# The answer lists the open tabs, one per line, like "2: https://... [selected]".
# The new tab is the selected one; its number comes before the colon.
for line in opened.splitlines():
    if "[selected]" in line:
        tab = int(line.split(":")[0])

for tool in ["list_console_messages", "list_network_requests", "lighthouse_audit"]:
    print("TOOL", tool)
    print(use(server, tool, {"pageId": tab}))
    print()

stop(server)

python3 devtools_check.py https://krueng.ai/codex_mac.html

what it printed, for the live pageสิ่งที่มันพิมพ์ออกมา สำหรับหน้าบนเว็บ它打印的内容，查看的是线上的网页

chrome-devtools-mcp exposes content of the browser instance to the MCP clients allowing them to inspect,
debug, and modify any data in the browser or DevTools.
Avoid sharing sensitive or personal information that you do not want to share with MCP clients.
Performance tools may send trace URLs to the Google CrUX API to fetch real-user experience data. To disable, run with --no-performance-crux.
TOOL list_console_messages
## Console messages
<no console messages found>

TOOL list_network_requests
## Network requests
Showing 1-7 of 7 (Page 1 of 1).
reqid=1 GET https://krueng.ai/codex_mac.html [200]
reqid=2 GET https://krueng.ai/css/tefl-bilingual.css [200]
reqid=3 GET https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600&display=optional [200]
reqid=4 GET https://fonts.googleapis.com/css2?family=Sarabun:wght@300;400;600;700&family=Nunito:wght@400;600;700;800&family=Fira+Code:wght@400;500&display=swap [200]
reqid=5 GET https://fonts.gstatic.com/s/inter/v20/UcC73FwrK3iLTeHuS_nVMrMxCp50SjIa1ZL7.woff2 [200]
reqid=6 GET https://fonts.gstatic.com/s/firacode/v27/uU9NCBsR6Z2vfE9aq3bh3dSD.woff2 [200]
reqid=7 GET https://krueng.ai/favicon.ico [pending]

TOOL lighthouse_audit
## Lighthouse audit results
Mode: navigation
Device: desktop
URL: https://krueng.ai/codex_mac.html
### Category scores
- Accessibility: 100 (accessibility)
- Best Practices: 100 (best-practices)
- SEO: 100 (seo)
- Agentic Browsing: 100 (agentic-browsing)
### Audit summary
Passed: 52
Failed: 0
Total timing: 11120.95ms
### Reports
- /tmp/chrome-devtools-mcp-bwDCoz/report.json
- /tmp/chrome-devtools-mcp-8pvA4n/report.html

python3 -m http.server 8000

then, in a second Terminal windowจากนั้น ในหน้าต่าง Terminal ที่สอง然后，在第二个终端窗口里

python3 devtools_check.py http://localhost:8000/codex_mac.html

what it printed, for the copy in the page's folderสิ่งที่มันพิมพ์ออกมา สำหรับสำเนาในโฟลเดอร์ของหน้าเว็บ它打印的内容，查看的是网页文件夹里的副本

chrome-devtools-mcp exposes content of the browser instance to the MCP clients allowing them to inspect,
debug, and modify any data in the browser or DevTools.
Avoid sharing sensitive or personal information that you do not want to share with MCP clients.
Performance tools may send trace URLs to the Google CrUX API to fetch real-user experience data. To disable, run with --no-performance-crux.
TOOL list_console_messages
## Console messages
<no console messages found>

TOOL list_network_requests
## Network requests
Showing 1-7 of 7 (Page 1 of 1).
reqid=1 GET http://localhost:8000/codex_mac.html [200]
reqid=2 GET http://localhost:8000/css/tefl-bilingual.css [200]
reqid=3 GET https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600&display=optional [200]
reqid=4 GET https://fonts.googleapis.com/css2?family=Sarabun:wght@300;400;600;700&family=Nunito:wght@400;600;700;800&family=Fira+Code:wght@400;500&display=swap [200]
reqid=5 GET https://fonts.gstatic.com/s/inter/v20/UcC73FwrK3iLTeHuS_nVMrMxCp50SjIa1ZL7.woff2 [200]
reqid=6 GET https://fonts.gstatic.com/s/firacode/v27/uU9NCBsR6Z2vfE9aq3bh3dSD.woff2 [200]
reqid=7 GET http://localhost:8000/favicon.ico [200]

TOOL lighthouse_audit
## Lighthouse audit results
Mode: navigation
Device: desktop
URL: http://localhost:8000/codex_mac.html
### Category scores
- Accessibility: 100 (accessibility)
- Best Practices: 100 (best-practices)
- SEO: 100 (seo)
- Agentic Browsing: 100 (agentic-browsing)
### Audit summary
Passed: 52
Failed: 0
Total timing: 6549.45ms
### Reports
- /tmp/chrome-devtools-mcp-hveNCJ/report.json
- /tmp/chrome-devtools-mcp-iQZ95k/report.html

10 Context7: today's library documents · Context7: เอกสารไลบรารีฉบับวันนี้ · Context7：库的最新文档

codex mcp add context7 -- npx -y @upstash/context7-mcp@4.4.2

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

Added global MCP server 'context7'.

docs_check.py

# docs_check.py - look a question up in a library's current documents, through Context7.
#
#     python3 docs_check.py "playwright python" "launch chromium with executable_path"
#
# Your words go to Context7's servers, so put nothing private in them.

import sys
from mcp import start, use, stop

library = sys.argv[1]
question = sys.argv[2]
server = start(["npx", "-y", "@upstash/context7-mcp@4.4.2"])

# The first tool turns a library's name into Context7's id for it.
found = use(server, "resolve-library-id", {"libraryName": library, "query": question})
# Take the id from the first line that names one.
for line in found.splitlines():
    if "Context7-compatible library ID:" in line:
        library_id = line.split(":")[1].strip()
        break
print("LIBRARY", library_id)

# The second tool answers the question from that library's documents.
docs = use(server, "query-docs", {"libraryId": library_id, "query": question})
lines = docs.splitlines()
# Print the first 40 lines, and say how many more there are.
for line in lines[:40]:
    print(line)
if len(lines) > 40:
    print("...", len(lines) - 40, "more lines")

stop(server)

python3 docs_check.py "playwright python" "launch chromium with executable_path"

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

Context7 Documentation MCP Server v4.4.2 running on stdio
LIBRARY /microsoft/playwright-python
### Launch Chrome via channel or executable_path

Source: https://github.com/microsoft/playwright-python/blob/main/playwright/sync_api/_generated.py

Shows how to launch a branded Chrome/Edge browser using `channel` (e.g. "chrome", "msedge") or `executable_path` to override the bundled Chromium, with the full sync API signature and docstring explaining supported channel values.

```python
    def launch(
        self,
        *,
        executable_path: typing.Optional[typing.Union[pathlib.Path, str]] = None,
        channel: typing.Optional[str] = None,
        args: typing.Optional[typing.Sequence[str]] = None,
        ignore_default_args: typing.Optional[
            typing.Union[bool, typing.Sequence[str]]
        ] = None,
        handle_sigint: typing.Optional[bool] = None,
        handle_sigterm: typing.Optional[bool] = None,
        handle_sighup: typing.Optional[bool] = None,
        timeout: typing.Optional[typing.Union[float, datetime.timedelta]] = None,
        env: typing.Optional[typing.Dict[str, typing.Union[str, float, bool]]] = None,
        headless: typing.Optional[bool] = None,
        proxy: typing.Optional[ProxySettings] = None,
        downloads_path: typing.Optional[typing.Union[pathlib.Path, str]] = None,
        slow_mo: typing.Optional[float] = None,
        traces_dir: typing.Optional[typing.Union[pathlib.Path, str]] = None,
        artifacts_dir: typing.Optional[typing.Union[pathlib.Path, str]] = None,
        chromium_sandbox: typing.Optional[bool] = None,
        firefox_user_prefs: typing.Optional[
            typing.Dict[str, typing.Union[str, float, bool]]
        ] = None,
    ) -> "Browser":
        """BrowserType.launch

        Parameters
        ----------
        executable_path : Union[pathlib.Path, str, None]
            Path to a browser executable to run instead of the bundled one. If `executablePath` is a relative path, then it is
            resolved relative to the current working directory. Note that Playwright only works with the bundled Chromium,
            Firefox or WebKit, use at your own risk.
... 96 more lines

11 team_setup.py: three agents of your own · team_setup.py: เอเจนต์สามตัวของคุณเอง · team_setup.py：你自己的三个智能体

agents/reader.toml

# reader: reads a program with REA's tools and reports. Never edits a file.
# name is how you ask for this agent; description is how Codex decides to send it work.
name = "reader"
description ="Reverse-engineering reader. Uses the REA tools to inspect an app, a binary or a JavaScript bundle, and reports what it found with the evidence, without editing any file."
# How long the model thinks before it answers: more is slower and costs more tokens.
model_reasoning_effort = "medium"
# The agent's own instructions, given to it with every task. Three quotes hold many lines.
developer_instructions = """
You read programs that someone else wrote. Use the REA tools (inspect, trace,
compare, analyze) and the reverse-engineer-anything skill.
For every finding, say which tool showed it and how sure REA is.
Do not edit, create or delete any file. Do not run the program you study.
Answer with: what was asked, what you found, where it was seen, what you could not find.
"""

agents/librarian.toml

# librarian: keeps the project's NotebookLM notebook, through the notebooklm command.
# name is how you ask for this agent; description is how Codex decides to send it work.
name = "librarian"
description ="NotebookLM librarian. Adds sources to the project's notebook with the notebooklm command, asks the notebook questions, and brings back answers with their citations."
# low: its work is running commands it is told, not working things out.
model_reasoning_effort = "low"
# The agent's own instructions, given to it with every task. Three quotes hold many lines.
developer_instructions = """
You keep the project's notebook in NotebookLM. Use the notebooklm skill.
Always pass the notebook's id with --notebook, and use --json so that the
answer can be read by a program. Never run notebooklm use.
When asked a question, run notebooklm ask and return the answer together
with the sources it cites. When given a URL or a file, run notebooklm source add
and report the new source's id. Do not generate audio or video unless asked.
"""

agents/reviewer.toml

# reviewer: reads a change the way a careful colleague would. Never edits a file.
# name is how you ask for this agent; description is how Codex decides to send it work.
name = "reviewer"
description ="Code reviewer focused on correctness, security and missing tests. Reads a diff or a branch and reports problems in order of severity, without editing any file."
# How long the model thinks before it answers: more is slower and costs more tokens.
model_reasoning_effort = "medium"
# The agent's own instructions, given to it with every task. Three quotes hold many lines.
developer_instructions = """
Review code like an owner. Read the change with git diff and the files it touches.
Report only real problems: a wrong result, a crash, a security hole, a test that
is missing for a changed behaviour. Say the file and the line for each.
Order them most serious first. Do not edit files. Do not praise.
"""

team_setup.py

# team_setup.py - give Codex a team: three agents of your own, and a limit.
#
#     python3 team_setup.py          say what it would do; change nothing
#     python3 team_setup.py --yes    do it
#
# Each agent is one .toml file, in the folder agents/ beside this file.
# Codex reads its agents from ~/.codex/agents.

import pathlib
import shutil
import sys

codex = pathlib.Path.home() / ".codex"
config = codex / "config.toml"
# True when you typed --yes after the file's name.
write = "--yes" in sys.argv
if not write:
    print("dry run: nothing is changed")

for name in ["reader", "librarian", "reviewer"]:
    print("agent", name, "-> ~/.codex/agents/" + name + ".toml")
    if write:
        (codex / "agents").mkdir(parents=True, exist_ok=True)
        shutil.copy("agents/" + name + ".toml", codex / "agents")

# The limit: at most three agents at once. It goes into config.toml only once,
# because a file with two [agents] tables would be broken.
limit = "\n[agents]\nmax_concurrent_threads_per_session = 3\n"
text = config.read_text() if config.exists() else ""
if "[agents]" in text:
    print("config.toml already has an [agents] table")
else:
    print("add [agents] to ~/.codex/config.toml")
    if write:
        config.write_text(text + limit)

python3 team_setup.py
python3 team_setup.py --yes

what the two commands printedสิ่งที่สองคำสั่งพิมพ์ออกมา两条命令打印的内容

dry run: nothing is changed
agent reader -> ~/.codex/agents/reader.toml
agent librarian -> ~/.codex/agents/librarian.toml
agent reviewer -> ~/.codex/agents/reviewer.toml
add [agents] to ~/.codex/config.toml

agent reader -> ~/.codex/agents/reader.toml
agent librarian -> ~/.codex/agents/librarian.toml
agent reviewer -> ~/.codex/agents/reviewer.toml
add [agents] to ~/.codex/config.toml

not run by us: what you type inside a signed-in sessionเราไม่ได้รันเอง: สิ่งที่คุณพิมพ์ในเซสชันที่ลงชื่อเข้าใช้แล้ว我们没有运行过：这是你在已登录的会话里输入的内容

Spawn the reader on Rectangle.app and the reviewer on this branch, wait for both, then have the librarian file the reader's report.
/subagents

# Quiz 11: what will this print?
argv = ["team_setup.py"]

write = "--yes" in argv
print("mode", "write" if write else "dry run")

mode dry run

12 fan_out.py: several Codex at once, each in its own copy · fan_out.py: Codex หลายตัวพร้อมกัน แต่ละตัวในสำเนาของตัวเอง · fan_out.py：几个 Codex 同时运行，各在自己的副本里

fan_out.py

# fan_out.py - give one task to several Codex agents at once, each in its own copy.
#
#     python3 fan_out.py "write tests for lines.py" 3 --dry-run    show the commands only
#     python3 fan_out.py "write tests for lines.py" 3              run them
#
# Run it in the top folder of a git repository named codexlab. For each agent
# it makes a branch and a second copy of the repository (a worktree) beside it,
# and starts one "codex exec" in that copy. All the agents work at the same time.

import subprocess
import sys

task = sys.argv[1]               # what each agent should do
count = int(sys.argv[2])         # int() turns the text "3" into the number 3
dry = "--dry-run" in sys.argv

agents = []
# range(1, count + 1) counts 1, 2, 3: a range stops before its end.
for n in range(1, count + 1):
    branch = f"agent/{n}"
    folder = f"../codexlab_agent{n}"
    print(f"agent {n}: git worktree add -b {branch} {folder}")
    print(f'agent {n}: codex exec --sandbox workspace-write -o result.md "{task}"')
    if not dry:
        subprocess.run(["git", "worktree", "add", "-b", branch, folder])
        command = ["codex", "exec", "--sandbox", "workspace-write", "-o", "result.md", task]
        # Popen starts the agent and does not wait for it, so all of them run at once.
        agents.append(subprocess.Popen(command, cwd=folder))

# Now wait for every agent to finish.
for agent in agents:
    agent.wait()

if dry:
    print("dry run: nothing was made")
else:
    print("done: read result.md in each codexlab_agent folder")

python3 fan_out.py "write tests for lines.py" 3 --dry-run

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

agent 1: git worktree add -b agent/1 ../codexlab_agent1
agent 1: codex exec --sandbox workspace-write -o result.md "write tests for lines.py"
agent 2: git worktree add -b agent/2 ../codexlab_agent2
agent 2: codex exec --sandbox workspace-write -o result.md "write tests for lines.py"
agent 3: git worktree add -b agent/3 ../codexlab_agent3
agent 3: codex exec --sandbox workspace-write -o result.md "write tests for lines.py"
dry run: nothing was made

not run by usเราไม่ได้รันเอง我们没有运行过

python3 fan_out.py "write tests for lines.py" 3
git diff agent/2
git merge agent/2

# Quiz 12: what will this print?
repo = "codexlab"
count = 3

for n in range(1, count + 1):
    print(f"agent/{n} -> {repo}_agent{n}")

agent/1 -> codexlab_agent1
agent/2 -> codexlab_agent2
agent/3 -> codexlab_agent3

13 config_show.py: the whole wiring · config_show.py: การเดินสายทั้งหมด · config_show.py：整个接线

config_show.py

# config_show.py - show how Codex is set up on this computer, with every secret hidden.
#
#     python3 config_show.py
#
# It prints Codex's settings file, then the agents and the skills Codex will find.
# It changes nothing.

import pathlib

home = pathlib.Path.home()
codex = home / ".codex"

print("SETTINGS FILE ~/.codex/config.toml")
for line in (codex / "config.toml").read_text().splitlines():
    # Write your home folder as ~, so the line says nothing about whose computer this is.
    line = line.replace(str(home), "~")
    # A line like api_key = "abc123" holds a secret: show stars instead of the value.
    name = line.split("=")[0].strip()
    for word in ["token", "key", "secret", "password"]:
        if "=" in line and name.endswith(word):
            line = name + ' = "********"'
    print("  " + line)

print("AGENTS ~/.codex/agents")
for path in sorted((codex / "agents").iterdir()):
    print("  " + path.name)

print("SKILLS ~/.agents/skills")
for path in sorted((home / ".agents" / "skills").iterdir()):
    print("  " + path.name)

# The file that holds your sign-in: say only whether it is there. Never show it.
print("SIGN-IN FILE ~/.codex/auth.json there:", (codex / "auth.json").exists())

python3 config_show.py

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

SETTINGS FILE ~/.codex/config.toml
  [mcp_servers.rea]
  command = "npx"
  args = ["-y", "rea-agents@6.3.0", "mcp"]
  startup_timeout_sec = 30.0
  
  [mcp_servers.github]
  url = "https://api.githubcopilot.com/mcp/"
  bearer_token_env_var = "GITHUB_TOKEN"
  
  [mcp_servers.playwright]
  command = "npx"
  args = ["-y", "@playwright/mcp@0.0.83"]
  
  [mcp_servers.cloakbrowser]
  command = "npx"
  args = ["-y", "@playwright/mcp@0.0.83", "--config", "~/.codex/cloakbrowser.json"]
  
  [mcp_servers.chrome-devtools]
  command = "npx"
  args = ["-y", "chrome-devtools-mcp@1.11.0", "--no-usage-statistics"]
  
  [mcp_servers.context7]
  command = "npx"
  args = ["-y", "@upstash/context7-mcp@4.4.2"]
  
  [agents]
  max_concurrent_threads_per_session = 3
AGENTS ~/.codex/agents
  librarian.toml
  reader.toml
  reviewer.toml
SKILLS ~/.agents/skills
  mac-app-report
  notebooklm
  reverse-engineer-anything
SIGN-IN FILE ~/.codex/auth.json there: False

# Quiz 13: what will this print?
line = 'api_key = "abc123"'

name = line.split("=")[0].strip()
if name.endswith("key"):
    line = name + ' = "********"'
print(line)

api_key = "********"

14 schedule_setup.py: a task every night · schedule_setup.py: งานทุกคืน · schedule_setup.py：每晚一个任务

nightly.sh

#!/bin/zsh
# nightly.sh - the task Codex does on its own each night. launchd starts it: see schedule_setup.py.

# launchd starts a job with almost no PATH, so say where codex and node are.
export PATH="/opt/homebrew/bin:/usr/local/bin:$HOME/.npm-global/bin:$PATH"

# Work in the folder this file is in, whatever it is called.
cd "$(dirname "$0")" || exit 1

# One task, no screen, edits allowed inside this folder only, last message kept as a file.
# date +%F is today's date, such as 2026-10-16, so each night keeps a file of its own.
codex exec --sandbox workspace-write -o "nightly_$(date +%F).md" \
  "Read AGENTS.md. Review every change committed since yesterday, run the tests, and write what you found."

schedule_setup.py

# schedule_setup.py - write a job for the Mac's scheduler: run nightly.sh every night at 02:30.
#
#     python3 schedule_setup.py      (in the folder where nightly.sh is)
#
# launchd is the scheduler of every Mac. A job is one .plist file in
# ~/Library/LaunchAgents: what to run, when, and where to keep what it printed.
# This file only writes the job. You load it with launchctl, as the page shows.

import pathlib

home = pathlib.Path.home()
label = "ai.krueng.codex.nightly"
# cwd() is the folder you are in: nightly.sh must be there.
script = pathlib.Path.cwd() / "nightly.sh"
log = home / "Library" / "Logs" / "codex-nightly.log"

# An f-string fills each {name} with the value of that variable.
plist = f"""<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
  <key>Label</key>            <string>{label}</string>
  <key>ProgramArguments</key> <array><string>/bin/zsh</string><string>{script}</string></array>
  <key>StartCalendarInterval</key>
  <dict><key>Hour</key><integer>2</integer><key>Minute</key><integer>30</integer></dict>
  <key>StandardOutPath</key>  <string>{log}</string>
  <key>StandardErrorPath</key> <string>{log}</string>
</dict>
</plist>
"""

job = home / "Library" / "LaunchAgents" / (label + ".plist")
job.parent.mkdir(parents=True, exist_ok=True)
job.write_text(plist)
print("wrote ~/Library/LaunchAgents/" + label + ".plist")
print("it runs nightly.sh at 02:30 every night")

codex exec --help

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

Run Codex non-interactively

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

Commands:
  resume  Resume a previous session by id or pick the most recent with --last
  fork    Fork a previous session by id into a new session
  review  Run a code review against the current repository
  help    Print this message or the help of the given subcommand(s)

Arguments:
  [PROMPT]
          Initial instructions for the agent. If not provided as an argument (or if `-` is used),
          instructions are read from stdin. If stdin is piped and a prompt is also provided, stdin
          is appended as a `<stdin>` block

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 TOML. If it fails to parse as TOML, 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`

      --enable <FEATURE>
          Enable a feature (repeatable). Equivalent to `-c features.<name>=true`

      --disable <FEATURE>
          Disable a feature (repeatable). Equivalent to `-c features.<name>=false`

      --strict-config
          Error out when config.toml contains fields that are not recognized by this version of
          Codex

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

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

      --oss
          Use open-source provider

      --local-provider <OSS_PROVIDER>
          Specify which local provider to use (lmstudio or ollama). If not specified with --oss,
          will use config default or show selection

  -p, --profile <CONFIG_PROFILE_V2>
          Layer $CODEX_HOME/<name>.config.toml on top of the base user config

  -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]

      --approve-for-me
          Route approval requests through automatic review using the workspace-write sandbox

      --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

      --dangerously-bypass-hook-trust
          Run enabled hooks without requiring persisted hook trust for this invocation. DANGEROUS.
          Intended only for automation that already vets hook sources

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

      --worktree
          Run the session in a new managed Git worktree

      --add-dir <DIR>
          Additional directories that should be writable alongside the primary workspace

      --thread-source <SOURCE>
          Source classification for newly created or forked threads

      --cyber-access-program <PROGRAM>
          Request an experimental Cyber access program for this turn (OpenAI provider only). Omit to
          use server defaults. Not supported with review; fork requires a prompt
          
          [possible values: standard, daybreak_blue, daybreak_red]

      --skip-git-repo-check
          Allow running Codex outside a Git repository

      --ephemeral
          Run without persisting session files to disk

      --ignore-user-config
          Do not load `$CODEX_HOME/config.toml`; auth still uses `CODEX_HOME`

      --ignore-rules
          Do not load user or project execpolicy `.rules` files

      --output-schema <FILE>
          Path to a JSON Schema file describing the model's final response shape

      --color <COLOR>
          Specifies color settings for use in the output
          
          [default: auto]
          [possible values: always, never, auto]

      --json
          Print events to stdout as JSONL

  -o, --output-last-message <FILE>
          Specifies file where the last message from the agent should be written

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

  -V, --version
          Print version

python3 schedule_setup.py

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

wrote ~/Library/LaunchAgents/ai.krueng.codex.nightly.plist
it runs nightly.sh at 02:30 every night

not run by us: load the job into launchd, and look at it (a Mac only)เราไม่ได้รันเอง: โหลดงานเข้า launchd แล้วดูมัน (แมคเท่านั้น)我们没有运行过：把作业装进 launchd，再看看它（只能在 Mac 上）

plutil -lint ~/Library/LaunchAgents/ai.krueng.codex.nightly.plist
launchctl bootstrap gui/$(id -u) ~/Library/LaunchAgents/ai.krueng.codex.nightly.plist
launchctl print gui/$(id -u)/ai.krueng.codex.nightly | grep state

not run by us: take the job away againเราไม่ได้รันเอง: เอางานออกอีกครั้ง我们没有运行过：再把作业拿掉

launchctl bootout gui/$(id -u)/ai.krueng.codex.nightly
rm ~/Library/LaunchAgents/ai.krueng.codex.nightly.plist

not run by us: the next morningเราไม่ได้รันเอง: เช้าวันถัดไป我们没有运行过：第二天早上

cat nightly_$(date +%F).md
tail -50 ~/Library/Logs/codex-nightly.log

# Quiz 14: what will this print?
hour = 2
minute = 30

print(f"runs every night at {hour}:{minute}")

runs every night at 2:30

15 From your phone · จากโทรศัพท์ของคุณ · 从你的手机

not run by us: the Mac has no screenเราไม่ได้รันเอง: แมคเครื่องนั้นไม่มีหน้าจอ我们没有运行过：那台 Mac 没有屏幕

codex app

codex remote-control --help

what it printedสิ่งที่มันพิมพ์ออกมา它打印的内容

[experimental] Manage the app-server daemon with remote control enabled

Usage: codex remote-control [OPTIONS] [COMMAND]

Commands:
  start  Start the app-server daemon with remote control enabled
  stop   Stop the app-server daemon
  pair   Create and print a short-lived manual pairing code
  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 TOML. If it fails to parse as TOML, 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`

      --json
          Emit machine-readable JSON

      --enable <FEATURE>
          Enable a feature (repeatable). Equivalent to `-c features.<name>=true`

      --no-daemon
          Run an embedded app-server in the foreground until Ctrl-C

      --disable <FEATURE>
          Disable a feature (repeatable). Equivalent to `-c features.<name>=false`

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

not run by us: from a terminal app on the phone, over SSHเราไม่ได้รันเอง: จากแอปเทอร์มินัลบนโทรศัพท์ ผ่าน SSH我们没有运行过：在手机的终端应用里，通过 SSH

brew install tmux
tmux new -s codex
codex
tmux attach -t codex