nc://channel-chatwork

Chatwork: talk to Nodecode from a Chatwork room. A Nodecode rewrite of TakeshiTGAL/hermes-plugin-chatwork by TakeshiTGAL.

Chatwork is the business chat most Japanese companies live in: rooms, no threads, and a message body that is its own markup. This cell puts a bot account in it — mention it with Chatwork's own To button, or reply to something it wrote, and it answers under your message the way the Reply button does, so you get the usual notification. A poll of Chatwork's own API brings the messages in, so your machine needs no public address and no open port.

Get it

Tell the agent "add the chatwork cell", or run nodecode add nc://channel-chatwork.

The channel kit that ships with Nodecode comes with it, and does the rooms, lanes, cards and delivery; this folder contributes Chatwork's API and notation.

Set it up

1. Use an account of its own for the bot — not yours. Log in to Chatwork as it and issue an API token from https://www.chatwork.com/service/packages/chatwork/subpackages/api/token.php.
2. Keep the token in a file (mode 600) or an environment variable, never in the config.
3. Invite that account to every room it should answer in, with permission to write.
4. Add a channels.chatwork section to ~/.nodecode/config.jsonc; config.example.jsonc has every key. The least that works:

"channels": {
  "chatwork": {
    "token_file": "~/.nodecode/secrets/chatwork-token",
    "allowed_rooms": ["123456789"],
    "allowed_users": ["987654"],
    "owner": ["987654"]
  }
}

At least one of allowed_rooms / allowed_users must be set: a message runs with full access to your machine, so the bot never starts open to everyone.

5. Say "set up chatwork" to Nodecode, or call (nck:probe "chatwork") yourself: it names the account the token belongs to and the rooms it can see, by id and name, so you can fill allowed_rooms from that instead of guessing ids. The room id is the number after rid in the room's URL.

Use it

• Group rooms: press Chatwork's To button on the bot — [To:the bot] — or reply to one of its messages, and it answers with a reply of its own.
• My Chat and direct chats: every message is an ask, mention or not.
• Free-response rooms: list a room in free_response_rooms and every message there is answered; set require_mention false to do that everywhere.
• Commands: a line starting with / is a Nodecode command (/help, /new, /model), answered in the room; with owner set, the commands that act on your machine are the operator's alone.
• Writing: answers are plain text with Chatwork's own notation — a code block becomes a Chatwork code box, a heading becomes 【...】, and a notification tag the model writes by itself is defused so it cannot ping a room.

The lane checks Chatwork every few seconds (poll_seconds, default 5), and only re-reads a room whose metadata moved, so it stays inside Chatwork's 300 calls per five minutes. Messages said before the bot started are marked, not answered; a restart does not replay them. A token that belongs to another account starts fresh in every room.

What it does not do

Chatwork has no message deletion, so a status card stays where it was written and the answer posts under it. There is no typing indicator (the beat is one cheap call a day), no reaction, and no file upload. Threads do not exist in Chatwork, so every answer lands in the room it was asked in.

License

MIT, see LICENSE. Rewritten from the plugin named in NOTICE.