DEV Community

Cover image for I built a local AI tool to analyze why I abandon 90% of my GitHub repos (Source Code included)
beatsprom
beatsprom

Posted on

I built a local AI tool to analyze why I abandon 90% of my GitHub repos (Source Code included)

#webdev #programming #javascript #ai

If you look at my GitHub, it looks like a graveyard of "next big things."

I have over 20 private repositories that follow the exact same lifecycle:

  1. I get a massive burst of inspiration.
  2. I spend 3 days over-engineering the perfect architecture.
  3. I hit the first boring/hard part (usually auth or marketing).
  4. I abandon the project and start a new one.

For a long time, I thought the problem was my tech stack or my discipline. But after reading some clinical psychology papers on behavioral sabotage in high-performers, I realized it wasn't a code problem. It was an Ego Threat Response.

So, being a developer, I did what developers do: instead of going to therapy, I built a local JavaScript profiling engine to diagnose my exact behavioral flaw, and then engineered an AI System Prompt to act as a clinical intervention.

Here is how I built it, and how you can test it on yourself.

The Architecture of Sabotage

I needed a way to map specific developer behaviors to clinical psychological profiles. I started by building a scoring engine in Vanilla JS.

Instead of asking generic personality questions, I mapped specific failure states. Here is a snippet from my app.js logic:

const questions = [
    {
        text: "Think about the last meaningful goal you genuinely cared about but never finished. What actually happened?",
        options: [
            { text: "I kept refining it, but it never felt ready enough to show anyone.", profile: "perfectionist" },
            { text: "I had a clear vision, but the gap between where I was and where I needed to be felt too large.", profile: "visionary" },
            { text: "I started strong, got excited, then quietly drifted toward something new.", profile: "dopamine" },
            { text: "I needed more research before I felt genuinely ready to begin.", profile: "researcher" },
            { text: "I finished it. I just never really showed it to anyone.", profile: "invisible" }
        ]
    }
    // ... more questions mapping to clinical archetypes
];
Enter fullscreen mode Exit fullscreen mode

The engine tracks the user's responses and tallies them against 7 specific "Builder Archetypes".

My personal diagnosis came back as The Dopamine Chaser (also known as the First-Mover Fugitive). It turns out I don't abandon projects because I'm lazy. I abandon them because the moment the "novelty" wears off and the project gets close to being public, my brain perceives public evaluation as a threat. Fleeing to a new project is a defense mechanism to avoid the reality of finishing one.

The Intervention: Engineering the AI System Prompt

Knowing the problem is only half the battle. I needed an intervention.

I decided to leverage LLMs, but a standard ChatGPT prompt like "Act as a therapist" gives terrible, generic advice. To get clinical-grade results, I had to architect a highly structured System Prompt based on ACT (Acceptance and Commitment Therapy) and IFS (Internal Family Systems).

I structured the PROMPT.html payload to force the AI into a strict state machine:

<!-- Snippet from the core engine -->
<p><strong>CRITICAL RULE: Ask exactly ONE question at a time. Never list questions. Never move forward without a real answer. If the answer is vague or defensive, use a single follow-up before proceeding.</strong></p>

<h2>PHASE 2 — CLINICAL PROFILE</h2>
<p>Synthesize everything you've heard. Identify the pattern with precision. Use the following format:</p>
<ul>
  <li>The primary intervention for this pattern is: [specific technique]</li>
  <li>When you notice [specific avoidance behavior], I will name it immediately</li>
</ul>
Enter fullscreen mode Exit fullscreen mode

By forcing the LLM to ask exactly one question at a time and evaluating the user against the 7 archetypes, it creates a deeply uncomfortable, highly accurate clinical mirror. It doesn't motivate you; it exposes the exact mechanism you use to hide.

The Results

When I ran myself through the engine, the AI didn't tell me to "work harder." It told me: "You are using technical complexity to hide from the vulnerability of shipping. Stop building. Publish what you have right now."

It hit so hard that I actually shipped the tool.

Try it on your own brain

If you have a graveyard of half-finished Next.js/React projects and you want to know the actual psychological reason why you keep abandoning them, you can run your brain through the engine I built.

I deployed the JS quiz to a static site so you don't have to run it locally.

🧠 Take the diagnostic test here: psychoprompt.netlify.app

It takes about 2 minutes. At the end, it will give you your specific clinical archetype and the exact System Prompt you can paste into ChatGPT to start debugging your own behavioral blocks.

Let me know what archetype you get in the comments. Be honest. We all have that one repo we are hiding from.

Top comments (0)