skills.rest
daymade
daymadeCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

kimi-use

Drive the Kimi desktop app via computer-use to query its built-in data plugins.

1.4k|216|Updated Oct 22, 2025
Search Tags:#financial data#data verification#plugins#computer-use#kimi#gui automation#MacOS
One-click install
npx skills add https://github.com/daymade/claude-code-skills --skill kimi-use
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: kimi-use
Source: https://github.com/daymade/claude-code-skills/tree/main/kimi-use
Command: npx skills add https://github.com/daymade/claude-code-skills --skill kimi-use

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Many valuable data sources — 天眼查 company records, 同花顺 iFinD financials, 标普全球市场财智, SEC, IMF, academic and legal databases — exist only as plugins inside the Kimi desktop app with no standalone API. This Skill turns the user's own logged-in Kimi session into a zero-credential data gateway by driving the app GUI through computer-use, while enforcing the verification discipline that screen-transcribed data requires.

Core Features & Use Cases

  • Dual-host GUI driving: Step-by-step operation flows for both Claude Code (computer-use MCP with screenshots and coordinates) and Codex (computer plugin with accessibility tree and element_index), including authorization, exclusive-lock handling, and provider prerequisites.
  • Query prompt patterns: Templates that name the target plugin, force source-labeled honest answers, anchor entities by full name or stock code, and pin time calibers — plus probe queries for first-time capability checks.
  • Verification discipline: Rules learned from real failures — screen transcription corrupts Chinese proper names, displayed counts are truncations, and two agreeing channels may share the wrong caliber — requiring cross-checks against authoritative sources before data enters deliverables.
  • Use Case: Ask the agent to fetch a company's shareholder list via the Kimi 天眼查 plugin; it drives the app in Work/Agent mode with K3, extracts the full Markdown result, cross-checks names against an authoritative source, and saves the data with source, caliber, and fetch-date labels.

Quick Start

Use the Kimi app to query 贵州茅台's latest financial statements through the 恒生聚源 plugin, label every field with its source, and cross-check the key figures against an authoritative channel before saving.

Frequently Asked Questions about kimi-use

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I query data sources that have no API using the Kimi desktop app?▼

Drive the Kimi.app GUI with computer-use tools: request access, open the app, switch to Work/Agent mode with the K3 model, then send a query that names the target plugin and demands source-labeled answers. Extract results via the copy button to clipboard with pbpaste.

How to control the Kimi app from Claude Code with computer-use?▼

Computer-use tools are deferred in Claude Code, so first run ToolSearch for "computer-use" to load them, then call request_access for the Kimi app. Note the tools are unavailable in Kimi k3 provider sessions and authorization is an exclusive per-machine lock.

Does this work with the kimi.com web version in a browser?▼

No, browser automation of kimi.com is handled by the separate kimi-webbridge skill, which uses a local daemon instead of GUI driving. This skill only covers the Kimi desktop application on macOS.

Why does Kimi say a plugin is not callable when it is installed?▼

Negative capability claims are only trustworthy in Work/Agent mode with the K3 model; Chat mode and the K2.6 fast model have falsely reported installed plugins as uncallable. Re-run a probe query under the correct mode and model before accepting any negative conclusion.

Can I trust data copied from the Kimi app screen directly?▼

No. Screen transcription has mis-transcribed Chinese proper names and treated truncated lists as complete, and Kimi itself can state wrong figures. Load-bearing names and numbers must be cross-checked against an independent authoritative source before entering deliverables.

What are the limitations of using Kimi plugins as a data channel?▼

The flow is verified only on macOS, requires the user's logged-in session, and installed does not mean callable for every plugin. Interface-reported counts reflect only what the plugin indexed, so coverage tasks need a second channel and union of results.

Related Skills

Explore More Data & Analytics Skills→
ThinkInAIXYZThinkInAIXYZ
6.3k

computer-use

Automate native desktop applications through accessibility trees, screenshots, and verified UI actions.

Official
Advanced
yusufcmgyusufcmg

market-research

Produce source-attributed market research reports with explicit sources and recommendations.

Community
Advanced
browserbasebrowserbase
3.7k

company-research

Discover and research target companies matching an ICP using Browserbase Search API.

Official
Advanced
Z5Jonathan-makerZ5Jonathan-maker

kimi-webbridge

Automate authenticated browser navigation and interaction via the Kimi WebBridge daemon.

Community
Intermediate