Build a Code Onboarding Assistant with AI

I want to set up MCP Roots so you have controlled, read-only access to a specific local project folder for a code onboarding session.

The folder path is: [REPLACE WITH YOUR ACTUAL FOLDER PATH, e.g. /Users/yourname/projects/my-app]

Please do the following:
1. Generate the correct MCP Roots JSON configuration block I need to add to my Claude MCP settings file to give you scoped access to only this folder.
2. Show me exactly where and how to add this configuration (which file to edit, what the full entry should look like).
3. Explain in 2-3 sentences why scoping access to one root is safer than giving access to my whole file system.
4. Tell me what to say to confirm you can see the folder once the config is applied.

You now have MCP Roots access to my project folder. I want you to act as the orchestrating agent and delegate this codebase analysis across focused subagents. For each subtask below, treat it as a separate delegated analysis unit and report the findings clearly labeled by task name.

Subtask 1 — File Map: List every file and folder in the project. Group them by type (config, source, tests, assets, docs). Flag any file that looks like a main entry point.

Subtask 2 — Architecture Summary: Read the key source files and describe the overall architecture in plain English. What kind of project is this? What pattern does it follow (MVC, component-based, script-based, etc.)?

Subtask 3 — Dependency Audit: Find any package.json, requirements.txt, Gemfile, go.mod, or equivalent. List the main dependencies and explain in one sentence what each major one does.

Subtask 4 — Data Flow Trace: Identify how data enters the application (user input, API call, file read, etc.), how it moves through the code, and where it ends up (output, database write, rendered UI, etc.).

After all four subtasks are complete, write a combined "Onboarding Summary" section that synthesizes the findings into a 200-300 word overview a new developer could read in under 2 minutes. Use plain English. Avoid jargon unless you define it.

Before I rely on your onboarding summary, I want to run an Explain It Back check. Please do the following:

1. In your own words, explain back to me what this codebase does — as if you were describing it to a teammate who has never seen it. Do NOT copy your previous summary. Rephrase it from memory.

2. List any parts of the codebase where you made an assumption or filled in a gap because the code was unclear, missing, or ambiguous. Be honest — if you weren't 100% sure about something in your analysis, flag it here.

3. Tell me: what is the ONE thing about this codebase you are least confident about, and why?

4. Based on your re-explanation, do you want to revise or correct anything from your earlier Onboarding Summary? If yes, write the corrected version of just that section.

I need you to write two custom instruction blocks for ChatGPT — one for the "What would you like ChatGPT to know about you?" field, and one for the "How would you like ChatGPT to respond?" field.

Context: I'm setting up ChatGPT as a code onboarding Q&A assistant. I'll paste a structured codebase summary into the start of each new chat, and then ask questions about the code. I'm an intermediate developer — I understand basic programming concepts but I'm not familiar with this specific codebase yet.

For the first field ("About You"), write instructions that tell ChatGPT:
- I'm a developer who is onboarding to an unfamiliar codebase
- I will always paste a codebase summary at the start of the conversation
- I want it to treat that summary as the source of truth for the session
- I may ask about specific files, functions, patterns, or concepts from the project

For the second field ("How to Respond"), write instructions that tell ChatGPT:
- Always anchor answers to the specific codebase summary I provided, not general assumptions
- If something isn't covered in the summary, say so explicitly instead of guessing
- Use clear, plain English explanations with short code examples when helpful
- Structure longer answers with numbered steps or labeled sections
- End each answer with one follow-up question I might want to ask next

Keep each block under 300 words. Write them ready to copy-paste.

I just completed a full AI-powered code onboarding session using a 5-step workflow. I want you to package this into a reusable template I can save and use on any future codebase.

Here's what the workflow covered:
1. MCP Roots config to give Claude scoped, safe read access to a project folder
2. Claude Code subagent delegation to analyze file structure, architecture, dependencies, and data flow
3. An Explain It Back verification step to check for gaps and assumptions in Claude's analysis
4. ChatGPT custom instructions to set up a consistent Q&A onboarding interface
5. A Q&A session using a pasted codebase summary as the source of truth

Please create a reusable "Code Onboarding Template" document with the following sections:

**Section 1: Pre-Flight Checklist** — A short checklist of things to confirm before starting (MCP Roots set up, folder scoped, Claude Code open, ChatGPT custom instructions saved).

**Section 2: The 4 Claude Code Prompts** — Placeholder versions of the subagent analysis prompt and the Explain It Back prompt, with [BRACKETS] wherever the user needs to fill in their specific project details.

**Section 3: The ChatGPT Session Starter** — A template opening message to paste at the start of every new ChatGPT onboarding chat (includes where to paste the summary and how to frame the first question).

**Section 4: 10 Go-To Onboarding Questions** — A list of 10 great questions to ask ChatGPT once the summary is loaded, covering architecture, logic, debugging, and "gotchas".

Format this as a clean Markdown document I can save as ONBOARDING-TEMPLATE.md.