How to use the lark-apps skill for Feishu Miaoda app building

What is an agent skill and what does lark-apps add?

An agent skill is a folder of instructions and reference files that an AI agent loads when a task matches the skill's description. It does not add new software by itself; it tells the agent which tools to call and how.

The lark-apps skill covers the Miaoda platform. Its SKILL.md is mainly a routing guide: it maps user intents such as creating an app, checking logs, setting environment variables or deploying to the matching lark-cli commands, and points to detailed reference files for each area.

• Create, list, look up and rename Miaoda apps
• Local development (pull source, edit locally, push with git) and cloud development (Miaoda AI generates and iterates)
• Creative design: HTML pages, UI mockups, interactive prototypes, wireframes, landing pages, dashboards and slide decks
• Deploy and release, and check release status and history
• Databases, file storage, environment variables and runtime cache
• Logs, traces, request and error metrics, latency and PV/UV analytics
• Collaborators, app roles, role members and visibility scope
• Open API keys and automation triggers (scheduled, record change, webhook, Feishu approval)
• Plugin install and user ID conversion

Install the lark-apps skill from the ZIP

Download the lark-apps ZIP from this page. The archive contains SKILL.md and the creative-design folder with its reference files, so keep the folder structure intact after unzipping.

For a personal install, unzip it into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. In the Claude apps you can instead upload the ZIP under Settings → Capabilities → Skills.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside your project folder
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Set up lark-cli and log in

The skill runs commands through the lark-cli tool, so lark-cli must be installed and available to the agent. This guide does not cover how to install lark-cli; the skill's metadata lists it as the required binary.

Miaoda apps belong to the user, so the skill uses user identity (--as user). Only log in when the CLI reports you are not logged in or lacks the needed scope. The skill asks for a single login command for this domain:

• Miaoda apps are personal assets: commands run with --as user
• Log in only when needed, because an unnecessary login can interrupt the task and trigger a device authorization
• Missing-permission errors follow the general rules in the lark-shared skill, using the apps domain

How to trigger the skill with natural requests

You do not need to name the skill. The description lists the kinds of requests that should load it, so plain requests work. Mention Miaoda, Spark or a Miaoda link when you want to be explicit.

Requests that match the skill include building a system, tool or app, deploying or publishing one, making an HTML page or site for Miaoda, designing a mockup or prototype, making a slide deck, and asking about Miaoda logs, metrics, environment variables or automations.

• Build a small internal tool on Miaoda and give me a shareable link
• Make a clickable prototype of a checkout flow and publish it
• Show me yesterday's error rate and p99 latency for app app_xxx
• Set the environment variable API_BASE for my Miaoda app
• Set up a scheduled automation that runs my Miaoda app every morning

Examples of common tasks

For metrics, the skill says to resolve the app ID first with a name search, then query the metrics. Request volume and error volume use the requests metric; latency uses the latency metric. Without a --series flag, requests returns total and error values, and latency returns p50 and p99.

For page views and active users, use the analytics command, not the metrics command. For a deployment, the skill's local development path requires your change to be committed and pushed with git before the release command runs.

• Find an app: lark-cli apps +list --keyword "<app name>" --as user
• Request and error volume: lark-cli apps +metric-list --app-id <app_id> --metric requests --since <range> --as user
• Latency: use --metric latency with the same app ID
• Page views and active users: use +analytics-list

Safety notes: scripts, network and live resources

The skill runs scripts and uses the network. Its commands read and write real Miaoda resources, including deployments, database changes, environment variables, access settings and API keys. Treat those actions as acting on production-facing systems, not as a sandbox.

Some secrets are shown only once. Open API key creation and reset return the key once, and automation webhook URLs and tokens are shown once and not saved to disk. Save them somewhere safe when they appear.

For local development, git commit, git push and your git credentials are involved. Review what will be pushed and deployed before confirming.

• Review database changes, deployments and key management before running them
• Keep a copy of one-time keys and webhook tokens when they are shown
• Do not paste credentials into chat if you can avoid it
• Expect scripts and network calls when the skill runs

Tips for getting reliable results

Name the app or give its app ID. The skill looks up an app ID from a name when you only give a name, but a link or ID avoids guesswork. Miaoda app links look like /app/app_xxx or /page/<token>.

For database sync from a Base, the skill says to create one single-table configuration at a time, and that batch or import tasks are one-time tasks that cannot be re-enabled. Plan large syncs in steps.

Use the reference file the skill points to for detail. The main SKILL.md routes you there, so ask the agent to read the relevant reference before a risky operation.

• Give the app name or app_id to avoid ambiguity
• Keep local source in git so changes are visible before release
• Expect one-time tasks to stay one-time; use streaming tasks for continuous sync
• Ask the agent to read the matching reference file for database, release or automation work