[DX] Curated example prompts directory

Goal

Teach users what to ask, not just what's possible.

What to ship

examples/prompts/ directory with 10-15 prompts. Each in its own .md file with:

• The exact prompt to type.
• Which tool(s) it will exercise.
• Expected response shape (or example).
• Optional: notes on what a misbehaving LLM might do instead.

Coverage:

• Discovery: list databases, what collections exist in fleet
• Read: find all vehicles where status is 'maintenance'
• Aggregate: count vehicles by region, what's the average age per fleet
• Schema: what fields does the vehicles collection have
• Diagnostic: why is this query slow
• Safe deny: prompts that should be denied with current default config (demonstrates safety)

Acceptance criteria

• README links to examples/prompts/ from the quickstart.
• Each prompt has been tested end-to-end against a real server.
• Includes at least one prompt designed to hit each tool tier (read / write / management).

Why

LLMs are bad at discovering capability. A human-curated prompt list is the fastest way for a new user to feel productive.

Start by reading the README quickstart and inspecting the existing project structure, then create examples/prompts/ with separate Markdown files covering the listed discovery, read, aggregate, schema, diagnostic, and denied prompts. Test each prompt end-to-end against a real server, cover read, write, and management tool tiers, and link the directory from the quickstart when all examples meet the stated format.