More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    LINE Dev

    1

    Teaches AI assistants LINE development: Messaging API bots, rich menus, LIFF and LINE Login, checked against the live docs, plus dry-run-first API operations.

    Free

    Security scanned
    LINE Dev

    LINE Dev

    Example session with this skill installed

    We're building a bot for our company's LINE group chat (a Taiwan LINE Official Account, on the free plan for now). Before we start coding, can you confirm a few things?

    1. How long can the bot use the reply token after a group message? Our LLM can take 30+ seconds, so can we still answer with a reply? And can we show a "typing..." animation while it thinks?
    2. If we push into the group instead, how many messages does one push count as? How many messages does the free plan include per month?
    3. How do we get the display name of whoever posted in the group, even if they never added the bot as a friend? Can we list all group members?
    4. When the bot answers, how do we @mention the person who asked?

    Please cite the official docs, and say so where something isn't documented.

    • Read your context and instructions
    • Compiled the line dev

    Summary

    A 30-second LLM still fits in the reply window, but only if the webhook is acknowledged first and the LLM runs in the background. A push to a group costs one message per member, member names work without friendship, listing every member needs a verified or premium account, and mentions need a textV2 message.

    QuestionShort answerSource
    Reply token lifetimeSingle use, within one minute of receiving the webhook. LINE says not to rely on the exact limitSend reply message
    Loading animation in a groupNot possible: one-on-one chats onlyDisplay a loading animation
    One push to a groupCounts once per member (30 members = 30 messages). Replies are freePricing: how messages are counted
    Free quota in TaiwanCheck the Taiwan plan page; the table in LINE's docs is Japan's examplePricing plans by region
    Poster's nameGroup member profile endpoint, works for non-friendsGet group chat member profile
    Full member listVerified or premium accounts onlyGet group chat member user IDs
    @mentiontextV2 with a substitution mention objectText message (v2)

    1. The reply token and a 30-second LLM

    • The token is single use and must be used within one minute after the webhook arrives. LINE warns the limit can change without notice, so use it as soon as you can.
    • Your server also has to answer the webhook itself within 2 seconds, or LINE records a request_timeout (webhook error statistics). So: verify the signature, return 200, run the LLM in the background, then reply with the stored token.
    • Put a timeout on the LLM call well inside the minute (for example 45 s) with a fallback reply, so the token is never wasted. Past that point only a push works, and a push costs quota (see 2).
    • There's no loading animation in groups. A quick "Working on it..." reply would use up the only reply token, so the real answer would then have to be a push.

    2. What a group push costs

    LINE counts messages by recipient: a push with four message objects to a chat of five people counts as five. Push, multicast, broadcast and narrowcast count; replies don't.

    The free quota depends on the country and plan. For a Taiwan account, check LINE Official Account (Taiwan) and its pricing FAQ. The plan table on developers.line.biz is Japan's, so don't budget with those numbers.

    Not documented: whether the bot itself, or members who blocked the account, count toward a group push. Measure it with the delivery statistics after a pilot instead of assuming.

    3. Names and the member list

    • For the sender's name, call GET https://api.line.me/v2/bot/group/{groupId}/member/{userId} with source.groupId and source.userId from the event. It works whether or not the user added or blocked the account.
    • Don't use GET /v2/bot/profile/{userId} for group members: it only covers friends and recent one-on-one senders, so it returns 404 for most of the group.
    • source.userId can be missing for users who never consented to profile access, so handle undefined.
    • "Get group chat member user IDs" works only on verified or premium accounts. On an unverified account, build the roster from memberJoined events, senders and mentions; silent members stay unknown.

    4. Mentioning the person who asked

    {
      "type": "textV2",
      "text": "{asker} Here's the summary you asked for.",
      "substitution": {
        "asker": {
          "type": "mention",
          "mentionee": { "type": "user", "userId": "U1234567890abcdef1234567890abcdef" }
        }
      }
    }
    
    • Mentions work only in reply and push messages. The bot must be in the group, and every mentioned user must be a member. Up to 20 mentions per message; keys use 0-9a-zA-Z_, up to 20 characters; you can't mention a bot.
    • Escape { and } in LLM or user text as {{ and }}, or the message fails or turns that text into placeholders.

    Next steps

    1. Build the webhook as verify, return 200, then a background job that replies within the minute, with a timeout fallback.
    2. Look up your Taiwan plan's quota before adding any scheduled pushes to the group.
    3. Use the group member profile endpoint for names, and cache them.
    4. Build mentions as textV2 from typed segments, escaping braces in any text you insert.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Playbook for LINE (LY Corporation) platform development. Instead of bundling a copy of the docs, it looks every limit, quota, price and field up in LINE's live official docs and cites the developers.line.biz page. It covers three jobs:

    • Build: write, review and debug Messaging API webhooks, reply/push messages, Flex Messages, rich menus, group-chat bots, LIFF apps and LINE Login with the smallest fix that works, in your project's existing stack (Node/TypeScript, Python, Deno, serverless and more). Fixes can be checked without a phone: the bundled webhook tool signs realistic events and sends them to your endpoint.
    • Answer spec questions: limits, quotas, pricing, message counting, error codes and event behaviour, looked up in the live docs with links, and flagged when they depend on region, plan or version.
    • Operate: call the Messaging API for you: check the monthly quota, send a test message, deploy a rich menu, set or test the webhook URL. Every write is a dry run until you confirm.

    What it gets right

    • Signatures over the raw body bytes (not re-serialized JSON), the console Verify ping with an empty events array, a fast 2xx with the slow work in the background, and dedupe on webhookEventId with a unique insert
    • Reply vs push: single-use, short-lived reply tokens, and a push to a group counted once per member against the monthly quota
    • Group bots: who can list members, which profile endpoint works for group members, isSelf mentions, and textV2 mentions built with escaped braces
    • Flex Message and rich menu limits checked locally before calling the API (sizes, URIs, image requirements), plus LINE's own validate endpoints
    • LIFF and LINE Login: server-side ID-token verification with the LINE Login channel ID, expiry and re-login without login loops, per-provider user IDs
    • The LIFF SDK v2.20.0–v2.31.0 XSS (CVE-2026-112148) and the external-browser restriction planned around 2026-11-24: how to find the version you actually ship and update it
    • Hosting traps: the Supabase Edge Functions JWT check, background work on Vercel and Cloudflare Workers, LINE Notify's end of life
    • Tokens are read from environment variables or --env-file, never printed, and only sent to api.line.me / api-data.line.me

    Tested

    Five scenarios with 49 grading criteria (debugging a webhook 401, group-bot spec questions, deploying a rich menu, LIFF authentication, a group-bot code review), run on Claude Opus 5.5 and graded blind: 47/49 with the skill, 44/49 without, and 16% faster on average with it. After the skill was revised for the two missed criteria, the two affected scenarios were re-run and passed in full, for 49/49 with the skill. A trigger test (10 prompts that should load the skill and 10 similar ones that shouldn't, 3 runs each) was judged correctly every time. Each scenario ran once per setup, so read the numbers as a direction, not proof. Eval set: https://github.com/Scorpio-meow/LINE-skill/blob/main/evals/evals.json

    Requirements

    • Reviewing and writing code: nothing beyond your agent.
    • Doc lookups: Python 3.8+ (standard library only) and web access to raw.githubusercontent.com or github.com (git optional, for full-text search). Without them, the agent can open the same pages with its own web tool.
    • Operating the API: HTTPS access to api.line.me and a channel access token in LINE_CHANNEL_ACCESS_TOKEN or a .env file. Webhook tests sign with your channel secret from LINE_CHANNEL_SECRET.
    • Installed from Agensi, the bundled scripts aren't saved to your machine; your agent fetches them and runs them from standard input.

    Open source under MIT: https://github.com/Scorpio-meow/LINE-skill

    Not an official LY Corporation project. LINE is a trademark of LY Corporation.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for APIs & Backend.

    • Benchmarks were run on Claude only (Opus 5.5), one run per scenario. Other SKILL.md agents should follow the same flow, but that isn't verified.
    • The scripts need an agent that can run terminal commands with network access. Sandboxes that only reach package registries and github.com (such as the Claude apps by default) can still look up docs through github.com but can't reach api.line.me, so operating the API works best from Claude Code or a similar tool on your own machine.
    • Installed from Agensi, the scripts aren't on disk, so the agent fetches them and feeds them to Python on standard input; in that mode JSON can't be piped in with - (pass a file or inline JSON instead). That path hasn't been benchmarked yet; for regular API operations, install from GitHub instead.
    • Pricing and the free message quota depend on the account's country and plan, so the skill sends you to the regional pricing page instead of quoting numbers.
    • Some endpoints need a verified or premium account (for example listing group members); the skill tells you, but can't work around it.
    • Narrowcast and audiences have no dedicated script commands (the generic call command reaches them); LINE MINI App, LINE Ads and the Conversion API are covered only by pointers in the doc map.
    • The gotcha files are a snapshot checked in 2026-10; numbers in them are marked "at the time of writing", and the skill re-checks the live docs before relying on them.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 13 days ago

    • Free with an Agensi account

    Accounts
    GitHubLine
    Software
    Python 3.8+
    Optional
    Git

    Python runs only the bundled scripts, with no packages to install. Git is only for full-text doc search.

    Permissions

    Network Access
    Environment Variables
    Terminal / Shell
    Read Files
    Write Files

    Allowed Hosts

    raw.githubusercontent.com
    github.com
    developers.line.biz
    api.line.me
    api-data.line.me

    File Scopes

    .env
    message
    and
    rich
    menu
    JSON
    or
    images
    you
    point
    it
    to
    files
    saved
    by
    the
    content
    command
    OS
    temp
    folder
    (line-docs-cache)

    Only the bundled scripts use these; reviewing and writing code needs none of them. line_docs.py fetches LINE's official docs from raw.githubusercontent.com, or makes a shallow git clone from github.com, and caches them for a day in the OS temp folder (line-docs-cache). line_api.py reads the channel access token from LINE_CHANNEL_ACCESS_TOKEN or --env-file, never prints it, and sends it only to api.line.me / api-data.line.me. webhook_tools.py reads LINE_CHANNEL_SECRET, never prints it, and only calls the webhook URL you give it (it sends the signature, not the secret). Push, multicast, broadcast, webhook URL changes, rich menu deploys and deletes, and leaving a group are dry runs until you confirm and the agent adds --yes. The content command saves a file a user sent to the bot into the working folder. When the skill's files aren't on disk (as when loaded from Agensi), the agent feeds a script to Python on standard input without saving it.

    Listed13 days ago
    Updated11 days ago

    What's inside

    Frequently Asked Questions

    Teaches AI assistants LINE development: Messaging API bots, rich menus, LIFF and LINE Login, checked against the live docs, plus dry-run-first API operations.

    This skill uses the open SKILL.md standard. Use it through an MCP-compatible agent connected to Agensi, and check the compatibility information on this page for confirmed agents and required capabilities.

    Connect your AI agent to Agensi through MCP, then ask it to use LINE Dev on your task. Your agent loads the latest approved version each time. Paid skills require a purchase or active subscription. Skill files are not available for download; your licence does not allow you to save, copy or export any part of a skill.

    New in APIs & Backend

    Free