AI-native

Connect your AI Agent (Agent Skill)

A Skill is a manual for the AI: install it once and it already knows which endpoint to call, what units the fields use, and how errors behave — instead of you pasting usage notes every time.

How is this different from MCP?

MCP gives the AI tools (it calls them itself); a Skill gives the AI knowledge (how to call, what fields mean, where the traps are). They sit side by side — a client can have both.

MCPSkill
What it isA protocol + serverA folder + Markdown
What you needURL + key (a server)Copy one directory (plain files)
What the AI getsTools (22, auto-invoked)Knowledge (endpoints / fields / errors / SDK)
Client support19 (remote MCP capable)45+ (open standard)
Best forAI looks data up in chatAI writes code / understands fields / unsupported clients

Install (pick one)

1Option 1: download and unzip (easiest)

Download the zip and unzip it into your client’s skills directory (next table). Restart the client after installing.

Download ashareapi-skill.zip~20 KB · 5 files · MIT
2Option 2: one command (Mac / Linux / WSL)

Replace `~/.claude/skills` with your client’s directory (see the table below).

mkdir -p ~/.claude/skills && cd ~/.claude/skills && curl -sL https://ashareapi.com/ashareapi-skill.zip -o s.zip && unzip -oq s.zip && rm s.zip
3Option 3: commit it to your project (team share)

Put `ashareapi/` into your project’s skills directory and commit — everyone who clones the repo gets it.

Client paths

ClientGlobal (all projects)Project (commit to git)
Claude Code~/.claude/skills/.claude/skills/
opencode~/.config/opencode/skills/.opencode/skills/
Cursor~/.claude/skills/.cursor/skills/
Codex (OpenAI)~/.codex/skills/.codex/skills/
Antigravity (Google)~/.gemini/antigravity/skills/.agents/skills/
VS Code / GitHub Copilot~/.copilot/skills/.github/skills/
WorkBuddy—.workbuddy/skills/
DeepSeek Harness (DSH)~/.dsh/skills/.dsh/skills/

`~` is your home directory. Not sure which to use? Either `.claude/skills/` or `.agents/skills/` works — both are "universal" locations read by Claude Code / opencode / Cursor / Antigravity / DeepSeek Harness. Full client list: agentskills.io/clients.

.claude/skills/ · .agents/skills/

More supported clients

How to confirm it works

After restarting, ask an A-share question (you don’t need to mention the skill):

  • What is the price of 600667 right now?
  • Show me the last 5 daily candles for 600519
  • Which sectors gained the most today?
  • Find stocks with PE under 20 and ROE above 15 (needs a key)

The difference: it proactively calls ashareapi endpoints instead of asking you for code — and it knows that "no data" and "fetch failed" are two different things.

What’s inside

Four reference files load on demand (so the AI isn’t flooded up front):

SKILL.md
Decision tree: what you want → which endpoint; four must-know semantics (units / string numbers / dates / response shape)
references/endpoints.md
All 32 endpoints with parameters and return notes
references/fields.md
Field semantics and units (volume in lots / amount in CNY / ratios as percentages…)
references/errors.md
401 / 429 / ok:false / empty results, and PoW quota boost
references/sdk.md
Official SDK usage (Python: pip install ashareapi · Node.js: npm install ashareapi)
Get started

The free endpoints (quote / kline / hot / market overview / breadth) need no key — install and try. Grab a key when you need financials, fund flow or top-trader boards.