This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine
What I Built
The Séance of the Dead Repositories is an atmospheric, CLI-based developer tool that summons the "spirits" of abandoned, archived, or forgotten open-source codebases.
Every year, thousands of brilliant open-source projects die due to dependency drift, language shifts, or developer burnout. They become digital ghosts lingering on GitHub.
When you feed a dead repository URL or local path into The Séance, a local open-weight AI takes on the persona of the codebase's lingering spirit inside a haunting, terminal-based séance chamber. It delivers three things:
The Post-Mortem: An eerie retrospective on why the project was abandoned and what bit rot killed it.
The Spectral Diagnosis: An Abstract Syntax Tree (AST) scan that pinpoints modern runtime errors (e.g., Python 2→3 syntax, deprecated API calls, broken build targets).
The Resurrection: Auto-generated, tested patches applied directly to bring the legacy software back to life on your modern machine.
Demo
Terminal Output Example:
Plaintext
🕯️ The candles flicker. The spirit of 'py-retro-canvas-2015' speaks...
💀 "I fell silent eight winters ago when Python 2 was banished to the shadow realm.
My print statements lacked parentheses; my imports withered away..."
⚡ [SCANNING BIT ROT]: Found 14 syntax breaking points and 3 dead dependencies.
🪄 [EXORCISING ERRORS]: Generating migration patches and applying sandboxed tests...
✨ [RESURRECTION COMPLETE]: Repo patched and running on Python 3.12!
How I Built It
The entire project is engineered around a 100% open-source local AI pipeline:
Open-Weight Code Model: Qwen2.5-Coder-7B-Instruct running locally via Ollama. It handles both the atmospheric persona formatting and the high-precision code migration tasks.
Agent Harness & Framework: Built using LangGraph (Python) to orchestrate stateful multi-step workflows (Repository Cloning → AST Parsing → Local LLM Analysis → Sandboxed Patch Generation → Test Verification Loop).
Terminal UI: Built with Rich and Textual to create a dark, candle-lit CLI complete with digital marigolds, ASCII spirit boards, and ritualistic progress indicators.
Sandboxed Execution: Uses local Docker containers to safely run test suites and verify patch stability without compromising the host environment.
Why Does Open Innovation Matter?
Open innovation isn't just an implementation detail for this project—it is the entire point.
Digital Preservation Without API Dependency: Using a closed API to preserve abandoned code creates a paradox: if the API vendor changes pricing, deprecates models, or goes out of business, the resurrection tool itself dies. Open-weight models guarantee that anyone can bring old software back to life 20 years from now—completely offline.
Total Code Privacy: Analyzing legacy repositories often involves inspecting obsolete security configurations, historical API keys, or private developer notes. Local inference ensures that zero bytes of raw source code ever touch a third-party server.
Zero-Cost Iteration Loops: Auto-fixing code requires dozens of trial-and-error reasoning passes per file. Running an open-weight model locally on consumer GPU/CPU hardware makes complex agentic loops free to execute indefinitely.
The Spirit of It
During Día de los Muertos, we build ofrendas to remember those who came before us and honor the stories they left behind. In open-source software, code is the story. Millions of lines of functional, creative software lie forgotten simply because runtime environments moved on. The Séance isn't just about fixing broken scripts—it's a digital ofrenda that honors the open-source ancestors whose foundational work built the modern web.
Prize Categories
Primary Category: Ghost in the Machine (Overall) Local / Air-Gapped Innovator: 100% local inference using Ollama with zero internet required once models are pulled.