Getting Started

This guide takes you from “I found this repository” to a productive AI-assisted automotive session in about five minutes.

What these skills are (and aren’t)

Each skill in skills/ is a prompt framework: a documented way of working with an AI assistant on a specific automotive task. There is nothing to install or run. The value is in:

• knowing what context to provide (the difference between a generic answer and a useful one),
• following a workflow that mirrors how good technicians actually diagnose and work, and
• understanding the limitations, so you know when to trust the output and when to verify.

They are not a replacement for factory service manuals, professional judgment, or physical inspection.

Step 1 — Pick your skill

Use the selection guide. The most common entry points:

• Warning light or scanner code → OBD-II Diagnostic Assistant
• Noise, leak, smell, vibration → Vehicle Troubleshooting
• “What service does my car need?” → Vehicle Maintenance Planner

Step 2 — Gather your vehicle context

Almost every skill needs this baseline. Have it ready before you start:

Vehicle: [year] [make] [model] [trim]
Engine:  [displacement, cylinders, turbo/NA, fuel type]
Transmission: [auto/manual/CVT]
Mileage: [current]
Recent work: [anything replaced or serviced recently]

The more precise this is, the better the answers. “My Civic” and “my 2015 Civic 1.8 CVT with 130k miles” get very different quality of advice.

Step 3 — Start with an example prompt

Every skill README has an Example Prompts section. Copy one, replace the details with yours, and paste it into your AI assistant (Claude, ChatGPT, or any capable model).

For a more persistent setup, paste the entire skill README into:

• a Claude Project’s instructions,
• a Custom GPT’s system prompt, or
• your agent framework’s system prompt.

This primes the assistant with the skill’s workflow and limitations for the whole conversation.

Step 4 — Work the conversation, not the question

The skills are designed for iteration:

1. Describe the situation with full context.
2. Answer the assistant’s follow-up questions — they discriminate between causes.
3. Do the suggested checks in the real world.
4. Report what you found, including measurements.
5. Repeat until the diagnosis is confirmed or the plan is complete.

The single biggest usage mistake is treating the assistant as a search engine (“P0420 fix”) instead of a diagnostic partner.

Step 5 — Verify before you act

Before applying any safety-relevant advice:

• Cross-check specs (torque values, capacities, clearances) against the factory service manual.
• Confirm the diagnosis with a physical test before buying parts.
• Defer to a professional for brakes, steering, SRS/airbags, fuel systems, and anything high-voltage — the skills will tell you when you have reached that boundary.

Read the full safety disclaimer.

Try a data-reading tool without a car

The companion obd2-mcp demo prints simulated trouble codes, freeze frame and live data. It is an optional tool integration; the prompt frameworks above work without it.

Next steps

• Browse the annotated example conversations to see complete sessions.
• Read the FAQ.
• Want to improve or add a skill? See the Skill Authoring Guide and CONTRIBUTING.md.

Start without choosing a skill

The Vehicle Brief Builder selects a workflow for your goal and attaches reviewed model references when the scope matches. Copy the generated prompt into your preferred AI, or save/print it for a workshop conversation. No account or API key is needed. Read its quality limits.