Glonce

Nietzschian Debugger is a CLI that only asks questions, never gives the fix

toolsThe Glonce editorsHacker News🔥54Glonce rating54Impact25Novelty35Relevance75Surprise45Players10Credibility65
Nietzschian Debugger is a CLI that only asks questions, never gives the fix

Nietzschian Debugger is a GitHub project that bills itself as "the anti-AI". Its pitch is that every other tool gives you answers, while this one confronts you with questions. You paste in a bug and it does not fix it. Per the README, it challenges your assumptions about what is happening, forces you to confront the evidence you are avoiding, tears apart weak reasoning, and tries to guide you to the "aha!" moment, which the README calls your moment and not the AI's.

The README's motivation is a blunt claim: AI is making developers weaker. It describes a habit of copy-pasting from ChatGPT, shipping, and repeating, and says nobody understands their own code anymore. The tool aims to build stronger developers "through struggle, not shortcuts". The README offers no evidence for the claim that AI makes developers weaker.

The tool is a command-line program. It can be run from the published npm release (v0.1.0) with npx nietzschian-debugger, which needs Node.js 20 or newer. The current source is a separate Python path, installed with pip from the GitHub repository and run as nietzschian debug "your bug"; it needs Python 3.10 or newer. Either way you need your own Anthropic API key (the README calls this BYOK, bring your own key), set as the ANTHROPIC_API_KEY environment variable. The README gives setup lines for Linux and macOS, Windows PowerShell and Windows CMD.

There are three intensity modes. The default, nietzsche, is the standard mode. Socrates is the gentle mode, "for when you need guidance, not punishment". Zarathustra is maximum intensity, "for seniors who want pain". You can also point the tool at specific code by naming a file path in the bug description, for example a file under ./src/middleware/, and the README says it then reads your actual code rather than giving generic examples.

The README includes a sample session. A user says their login endpoint returns 403 for some users. The tool pushes back on the word "some" as vague and asks what separates users who fail from those who do not. When the user says it seems random, the tool answers that "random" is the word we use when we have not looked hard enough, and tells the user to check the server logs. The user finds that all the failing users have expired JWT tokens. After another question about what changed in the auth system, the user realises the token TTL was changed the week before without invalidating existing sessions. The tool closes with the line that the user did not need to be told, they needed to be forced to look. The session ends with the summary "4 questions to root cause (Solved)" and a "Debugging Profile" with bar charts for assumption-checking, evidence-gathering and root cause speed. This is an illustrative sample, not a measured result, and the source does not say how the Growth Score is computed.

Other listed features: every response is a question and the tool "will NEVER tell you the fix"; a "contextual philosophy" that uses Nietzsche when you are avoiding, Seneca when you are overwhelmed and Sun Tzu when you need strategy; a Growth Score that tracks debugging skills across sessions; session history saved locally in .nietzschian/sessions/ for trend tracking; and a sliding context in which older turns of long sessions are summarized so they do not crash.

The README names its audience: junior developers who want to learn rather than copy-paste, senior developers who miss the struggle that made them good, engineering managers who want their teams thinking and not just prompting, bootcamp students, and anyone who believes the best developer is a self-reliant one.

For contributors, the README shows a clone, pip install -e ".[dev]", and a test run with pytest covering 121 tests. The stack is Python 3.10 or newer, the Claude API through the Anthropic SDK, Click for the CLI, Rich for the terminal UI, and pytest. The project is MIT licensed.

Key facts

  • Nietzschian Debugger is a CLI that responds to a pasted bug only with questions; the README says it "will NEVER tell you the fix".
  • It needs the user's own Anthropic API key (BYOK). The npm release v0.1.0 needs Node.js 20 or newer; the current Python source needs Python 3.10 or newer.
  • Three intensity modes: nietzsche (default), socrates (gentle) and zarathustra (maximum).
  • Sessions are saved locally in .nietzschian/sessions/, and a Growth Score tracks debugging skills across sessions.
  • The repository has 121 tests and an MIT licence. The README's claim that AI makes developers weaker comes with no evidence.

Why it matters

Most developer tools compete on how quickly they hand over a fix. This project takes the opposite position: it withholds the answer on purpose, and the README argues that copy-pasting from ChatGPT is leaving developers who do not understand their own code. It is a small, opinionated experiment in using an LLM as a questioner instead of an answer machine. The README's premise is the author's own assertion, with no evidence given.

Who it affects

The README targets junior developers who want to learn, senior developers who miss the struggle, engineering managers who want their teams thinking, and bootcamp students. In practice it matters to anyone with an Anthropic API key who wants to try a Socratic style of debugging.

How to use it

Set ANTHROPIC_API_KEY, then run npx nietzschian-debugger for the npm release (v0.1.0, Node.js 20 or newer). For the current source, install with pip install git+https://github.com/DilawarShafiq/nietzschian-debugger.git (Python 3.10 or newer) and run nietzschian debug "your bug". Add --intensity socrates for a gentler mode or --intensity zarathustra for maximum. Mention a file path in the bug description and the tool reads that code. The project is MIT licensed. Usage runs on your own Anthropic API key.

How solid is it

The material is the project's own README, so everything about behaviour is self-description. The 4-question session and the Growth Score bars are an illustrative sample, not measured results. No evidence, user study or benchmark shows the tool improves debugging skills or that AI tools make developers weaker. The text does not say which Claude model the tool uses. No release date, star count, user count or reception is given. The repository does list a test suite of 121 tests.

Risks and caveats

The "never gives answers" promise is the maker's claim and is not independently tested here. The npm release (v0.1.0) and the current Python source are two separate install paths, so what you get depends on which you pick. Each session calls the Anthropic API on your own key. The tool's premise, that AI makes developers weaker, is asserted without support, and a question-only approach may suit some bugs and some users better than others.

“You paste your bug. It doesn't fix it.”

— Nietzschian Debugger README