Skip to content
View jghiringhelli's full-sized avatar

Block or report jghiringhelli

Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
jghiringhelli/README.md

JC Ghiringhelli

I build tools and methodologies for AI-assisted software engineering. Independent researcher. Based in the US, originally from Uruguay.

In 2026 I published Generative Specification — a programming discipline for the stateless reader. The core argument: the dominant failure mode of AI-assisted development is not incorrect code, it is architectural drift. The fix is a discipline that constrains what a stateless reader can derive from the artifacts alone, across sessions that share no persistent context. I built and tested it across six production projects before writing the paper.

Paper (Zenodo): doi.org/10.5281/zenodo.19073543Methodology hub: genspec.devExperiments + quality gates: github.com/jghiringhelli/generative-specification

The language-layer implementation of these ideas is Loom — an AI-native formal language where every construct traces to a published theorem.


Projects

The methodology, reproducible experiments, and a community quality gate library — 17 gates mapped to the seven GS properties, open for contribution. If you want to challenge a claim, reproduce an experiment, or propose a gate, this is the place.

Production-grade engineering standards for AI coding assistants. Generates tailored instruction files (CLAUDE.md, .cursor/rules/, Copilot instructions) from curated template blocks matched to your stack. Supports Claude, Cursor, Copilot, Windsurf, Cline, and Aider.

npm

npx forgecraft-mcp setup .

Persistent, tiered AI memory that survives context resets — for Claude, GitHub Copilot, Cursor, and any MCP-compatible assistant. Five cognitive memory types (Semantic, Episodic, Procedural, Architectural, Preference), automatic tier promotion, decay, triggers, and optional Railway Postgres sync for cross-machine memory.

npm

npx -y chronicle-mcp

An AI-native formal language. Write a specification; Loom compiles it to Rust, TypeScript, Python, OpenAPI, and Terraform simultaneously — each output carrying Hoare contracts, type-level security labels, session type protocols, and differential privacy annotations. Every construct in the language traces to a published theorem between 350 BCE and 2011. The annotation burden that blocked the formal tradition for decades is gone. The tradition activates.

Graph-powered code intelligence for AI assistants. Builds a knowledge graph of your codebase — imports, calls, class hierarchies — so Claude, Copilot, and Cursor understand how your code actually connects, not just what files contain.

npm


Tech: TypeScript · Node.js · MCP · PostgreSQL · Knowledge Graphs · LLMs

Popular repositories Loading

  1. codeseeker codeseeker Public

    MCP server with semantic search + knowledge graph for Claude Code, Cursor, and Copilot

    TypeScript 18 2

  2. forgecraft-mcp forgecraft-mcp Public

    MCP server that gives AI coding assistants production-grade engineering standards — SOLID, testing, architecture, CI/CD — for Claude, Cursor, Copilot, Windsurf, Cline & Aider

    TypeScript 3 1

  3. generative-specification generative-specification Public

    Generative Specification: experiments, evidence, white paper, and community quality gates flywheel

    TypeScript 1

  4. scholaris-mcp scholaris-mcp Public

    Academic paper review MCP server — compression audit, structured review, citation checking, plagiarism detection, venue fit, and diff audit for researchers

    TypeScript 1

  5. idfx-solr-signals idfx-solr-signals Public

  6. homebrew-codeseeker homebrew-codeseeker Public

    Official Homebrew tap for CodeSeeker - Graph-powered code intelligence for Claude Code and AI assistants

    Ruby