EdgeOne Pages Skills

Official Agent Skills for developing and deploying projects on EdgeOne Pages.

Installation

npx skills add TencentEdgeOne/edgeone-pages-skills

After installation, your AI coding agent will automatically detect when you want to develop or deploy and use the right skill.

Usage

The skills are automatically available once installed. The agent will use them when relevant tasks are detected.

Deployment examples:

Deploy my project to EdgeOne Pages

Publish this React app to EdgeOne Pages China site

Deploy this Next.js project and give me the preview URL

Development examples:

Create an API for user registration

Add WebSocket support to my project

Write middleware to protect my /api routes with auth

Set up Edge Functions with KV storage for a page view counter

Create a Go API with Gin framework

Build a Python backend with FastAPI

Skills

Skill 1: edgeone-pages-deploy

Deploys frontend and full-stack projects to EdgeOne Pages.

Triggers: "deploy my app", "publish this site", "push this live", "create a preview deployment", "deploy to EdgeOne", "ship to production", "上线", "发布", "发一版", "重新部署"

What it does:

• Installs the EdgeOne CLI (edgeone) if not present
• Authenticates via browser login (preferred) or API token (headless/CI)
• Supports both China and Global sites
• Deploys with automatic framework detection and build
• Returns the live preview URL and console link

Skill 2: edgeone-pages-dev

Guides development of full-stack features on EdgeOne Pages.

Triggers: "create an API", "add a serverless function", "write middleware", "build a full-stack app", "add WebSocket support", "set up edge functions", "use KV storage", "create a Go API", "build a Python backend", "use Flask/FastAPI/Gin on EdgeOne Pages"

What it does:

• Helps choose the right runtime (Edge Functions vs Cloud Functions vs Middleware)
• Provides correct project structure and file-based routing patterns
• Guides Edge Functions development (KV Storage, Web APIs)
• Guides Cloud Functions development:
  • Node.js — npm, database, Express/Koa, WebSocket
  • Go — Gin, Echo, Chi, Fiber, standard net/http
  • Python — Flask, FastAPI, Django, Sanic, Handler class
• Guides Middleware development (request interception, auth, redirects, A/B testing)
• Covers local dev setup, environment variables, and debugging

Skill Structure

skills/
├── edgeone-pages-deploy/
│   ├── SKILL.md                        # Deployment flow, CLI setup, login, token management
│   └── references/
│       └── command-reference.md        # CLI commands, env vars, token management
└── edgeone-pages-dev/
    ├── SKILL.md                        # Entry point — decision trees & routing table
    └── references/
        ├── edge-functions.md           # Edge Functions (V8 runtime, Web APIs)
        ├── kv-storage.md              # KV Storage setup & API reference
        ├── node-functions.md          # Cloud Functions — Node.js (npm, Express/Koa, WebSocket)
        ├── go-functions.md            # Cloud Functions — Go (Gin, Echo, Chi, Fiber, net/http)
        ├── python-functions.md        # Cloud Functions — Python (Flask, FastAPI, Django, Sanic)
        ├── middleware.md              # Middleware (auth, redirects, A/B testing)
        ├── recipes.md                 # Project structure templates & common recipes
        └── troubleshooting.md         # Debugging & troubleshooting guide

Each skill follows the skill-creator standard:

• SKILL.md — YAML frontmatter (name + description) + core instructions
• references/ — detailed reference docs loaded on demand, routed from SKILL.md

Trigger Evaluation

Automated test suite to verify skill trigger accuracy. Uses Claude API as a "skill router" to batch-test queries and compute Precision / Recall / F1.

# Run full evaluation
ANTHROPIC_API_KEY=sk-xxx node eval/run-eval.mjs

# Verbose mode — print model's reasoning for each query
ANTHROPIC_API_KEY=sk-xxx node eval/run-eval.mjs --verbose

# Use a different model
ANTHROPIC_API_KEY=sk-xxx node eval/run-eval.mjs --model=claude-opus-4-20250514

Pass criteria: Precision ≥ 0.90, Recall ≥ 0.80, F1 ≥ 0.85. Results are saved to eval/results.json.

See eval/trigger-tests.md for the full test case design and boundary analysis.

Requirements

• Node.js ≥ 16

• npm (for CLI installation)

• An EdgeOne Pages account

  China site | Global site

License

MIT