This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine
What I Built
Every software repository eventually becomes a cemetery.
As projects grow under tight sprint deadlines, they accumulate supernatural baggage:
- Modules created during a 2:41 AM panic that no one imports anymore.
- Commits titled
"temporary fix (DO NOT TOUCH)"that silently crystallized into critical production architecture. - Features that were ceremoniously pruned from the codebase, only to be reincarnated three weeks later under aliases like
feature-v2.ts. - Unit tests martyred and commented out just to appease the CI pipeline.
I built Commit Cemetery 🪦 — an offline AI code archaeologist and forensic autopsy engine for Git repositories.
Commit Cemetery digs through your local git history, reflogs, and AST dependencies to exhume the spirits lingering in your codebase:
-
☠️ Crime Scenes (Suspicious Commits): Flags desperation incantations (
"pls work","final-final-v2"), broken promises ("temporary fix","clean up later"), test desecration ("commenting out tests so CI passes"), and commits pushed during the witching hour (12:00 AM – 5:00 AM). - 👻 Ghost File Autopsy: Identifies unreferenced, dormant files that have rested undisturbed in the catacombs for months, calculating a Necrotic Risk Score (0–100) and generating personalized tombstone epitaphs.
- 💀 Resurrections & Reverts: Traces code that refused to stay buried — detecting deleted features reborn under altered names and "reverts of reverts" that re-summoned previously banished bugs.
- 🧙 Gravedigger Archetypes: Awards historical developer titles (The Midnight Gravedigger, The Chaos Necromancer, The Ancient Caretaker) based on nocturnal push frequencies and git triage habits.
- 🔮 The Séance Chamber: An interactive terminal where developers can commune directly with the spirit of the codebase in natural language ("Who broke production on Friday?", "Is it safe to delete legacyAuth.ts?").
- 🎧 Native Crypt Drone: A browser-synthesized ambient drone created with the Web Audio API (zero audio files, 100% offline) that sets an eerie atmosphere while inspecting dead code.
Demo
- Live Interactive Demo: https://youtu.be/g1HbFIYjVBo
- GitHub Repository: https://github.com/dnano-more/commit-cemetery
Exhumation Snapshot
🪦 FORENSIC AUTOPSY REPORT: zenith-form
Condition: F (Possessed)
Epitaph: "Here lies Zenith Form: where temporary fixes became sacred dogma."
🧠 Gemma AI Forensic Verdict:
"This repository appears to contain the residual haunting of an authentication overhaul
abandoned mid-ritual. Multiple commits died violent deaths at 2:00 AM, only to be
resurrected under aliases like 'fix-final-final-v2'."
☠️ Most Suspicious Commit:
- Hash: c8f491b
- Author: Dnyaneshwar (02:41 AM)
- Message: "temporary fix (DO NOT TOUCH)"
- Evidence: 47 files touched in an unreviewed midnight push; reverted 48h later.
👻 Ghost Files:
- src/utils/legacyAuth.ts (256 days dormant | Risk: 92/100)
* Epitaph: "R.I.P. legacyAuth.ts — Abandoned during the Great JWT Migration of February."
Code
🪦 COMMIT CEMETERY
Every repository has a past. Dig through yours.
An offline, local-first AI code archaeologist and forensic autopsy engine for Git repositories.
Powered by Next.js 16, TypeScript, and local open-weight AI (Gemma via Ollama).
The Problem • Features • Architecture • Why Gemma? • Quick Start • The Séance • Sample Report
💀 The Problem
Over months and years, every software repository turns into a haunted cemetery:
- Ghost Files: Unreferenced utility files and deprecated scripts from 2 years ago that survive 14 refactors because nobody dares delete them.
-
Deceptive Commits: Commits titled
"temporary fix (DO NOT TOUCH)"pushed at 2:41 AM that quietly hardened into critical production dependencies. -
Reincarnated Code: Features that were ceremoniously pruned from the master branch, only to be resurrected under aliases like
feature-v2.tsthree weeks later. - Test Desecration: Assertions commented out in panic just to pass…
The repository is built with Next.js 16 (App Router & Turbopack), TypeScript, Tailwind CSS v4, and native Git CLI integration.
How I Built It
Commit Cemetery is deliberately not a naive LLM wrapper. Pumping an entire git history into a prompt window is wasteful, hallucination-prone, and expensive.
Instead, I implemented a two-stage pipeline: Deterministic Forensic Extraction followed by Local AI Archaeological Interpretation.
Local Git Repository
│
▼
┌─────────────────────────────────┐
│ DETERMINISTIC FORENSIC ENGINE │
│ • git log --all --pretty │
│ • git diff & revert detection │
│ • dormant file age analysis │
│ • nocturnal hour heuristics │
└────────────────┬────────────────┘
│
▼ Structured Forensic Evidence
┌─────────────────────────────────┐
│ LOCAL OPEN-WEIGHT AI (GEMMA) │ ◄── 100% Offline via Ollama
│ • Evaluates crime severity │ (gemma:2b / gemma:7b)
│ • Synthesizes Gothic verdicts │
│ • Powers interactive Séance │
└────────────────┬────────────────┘
│
▼
┌─────────────────────────────────┐
│ CEMETERY FRONTEND │
│ • Interactive Tombstones │
│ • Web Audio API Crypt Drone │
│ • Markdown Autopsy Exporter │
└─────────────────────────────────┘
The Tech Stack
-
AI Core: Google's open-weight Gemma (
gemma:2b) running locally via Ollama. -
Deterministic Heuristics Engine: Node.js child processes executing native
git log,git ls-files, and AST dependency checks to build structured crime manifests. - Frontend & API: Next.js 16 with React 19, TypeScript, and Tailwind CSS v4.
- Atmospheric Audio: Procedural oscillator synthesis using browser Web Audio API.
- Fallback Resilience: An intelligent built-in archaeological engine ensures the app functions seamlessly out of the box even when Ollama is not active.
Why Does Open Innovation Matter?
For developer tooling that touches source code, open-source and open-weight AI is not just a nice-to-have — it is an engineering necessity.
- Absolute Source Code Privacy: Git logs and reflogs contain confidential architectural decisions, proprietary algorithms, author metadata, and occasionally leaked API credentials. Sending raw git logs to a third-party closed SaaS API violates corporate security policies. Because Gemma runs completely locally on the developer's laptop, the code never leaves localhost.
- Infinite Offline Scans at Zero Cost: Forensic analysis often requires parsing hundreds of commits and files. With local open-weight inference, there are no rate limits, token meters, or subscription tiers.
- Control & Independence: Developers are not locked into closed API deprecation schedules. An open-weight model like Gemma can be inspected, customized, or swapped freely.
My Agent Session
This project was built with the assistance of an agentic pair-programming workflow, systematically breaking down the challenge rules, structuring the deterministic git parser, designing custom SVG vectors, and refining the local inference pipeline.
The Spirit of It
During development, I tested Commit Cemetery on an old private side-project from two years ago.
Running the exhumation yielded an immediate chill: it surfaced a commit titled "quick fix - will clean up before launch" timestamped at 3:14 AM on a Sunday. Directly beneath it was a ghost file: an entire payments helper module containing hardcoded sandbox test keys that had sat completely untouched and unreferenced for 620 days.
We often think of git history as dry data. But in reality, every commit message is a fossilized human emotion: confidence, exhaustion, panic, relief. Commit Cemetery turns those digital ghosts into an entertaining yet genuinely useful tool that helps teams clean up dead code and remember where they came from.
Prize Categories
- Primary: Overall Challenge Entry (Theme: Ghost in the Machine)
- Featured Category: Best Use of Gemma (Gemma open-weight model deployed via Ollama for local-first forensic archaeology).
Top comments (0)