feat: Agent skill for @casys/mcp-server documentation

[superWorldSavior]

Motivation

Developers using @casys/mcp-server with AI coding agents (Claude Code, Codex, Cursor, etc.) need the framework docs available in context — not behind an API call.

A skill file (SKILL.md + reference docs) is the most direct path: the agent loads it, has the full API surface, and can build with it immediately. Zero setup, zero latency, zero network dependency.

Proposal

1. Skill package (@casys/mcp-server-skill or published on ClawHub)

casys-mcp-server-skill/
├── SKILL.md              # Quick start, core concepts, common patterns
└── references/
    ├── api-server.md     # ConcurrentMCPServer API reference
    ├── api-middleware.md  # Middleware pipeline, built-in middlewares
    ├── api-auth.md       # OAuth2/OIDC presets, scope-based auth
    ├── api-transport.md  # HTTP, SSE, STDIO transport options
    ├── api-observability.md  # Prometheus, OpenTelemetry
    └── examples.md       # Real-world patterns, recipes

2. SKILL.md content

• When to use (vs official SDK, vs FastMCP)
• Install (npm + deno add)
• Core pattern: register tool → add middleware → start server
• Quick reference for middleware pipeline order
• Link to reference files for deep dives

3. Distribution

• ClawHub: agents with OpenClaw can install via clawhub install
• In-repo: /skill directory in this repo for anyone to copy
• llms.txt: generate from skill content for LLM-friendly web access

Use cases

• Agent builds an MCP server from scratch using the skill as context
• Agent adds auth/rate-limiting to an existing server by reading the middleware reference
• Agent troubleshoots issues with the correct API patterns loaded

Scope

• SKILL.md — concise, opinionated quick reference
• references/ — detailed API docs per module
• Publish to ClawHub
• Add /skill directory to repo
• (Optional) Generate llms.txt from skill content for web