Skip to content

dvelton/instrucgen

Repository files navigation

instrucgen

Enter a GitHub username, get a comprehensive prompt that generates personalized AI custom instructions based on their public profile and activity.

Try it live

What this does

Most AI custom instructions files are either generic boilerplate or require you to manually describe your tech stack and preferences. instrucgen takes a different approach: it looks at what you actually do on GitHub and builds a detailed prompt from that.

Enter any public GitHub handle. The app fetches the user's profile, repositories, dependency files, and recent activity, then analyzes all of it to construct a structured prompt you can paste into your AI tool of choice. The AI does the final generation -- instrucgen does the hard work of gathering, organizing, and contextualizing the data so the output is grounded in reality rather than generic filler.

How it works

The app makes 15-25 GitHub API calls per analysis, pulling data from multiple sources:

  • User profile -- bio, company, location, account age
  • Repositories -- up to 500 repos, sorted by recent activity, with topics and descriptions
  • Language stats -- byte-level language breakdowns for the top 8 repos
  • Dependency files -- package.json, requirements.txt, Cargo.toml, go.mod, Gemfile, pyproject.toml fetched via raw.githubusercontent.com (no rate limit on this endpoint)
  • Project structure -- src/, tests/, docs/, CI configs, linter configs, Dockerfiles, tsconfig presence
  • Public events -- last 300 events for activity patterns and contribution style

From that raw data, the analyzer produces a structured developer profile covering:

  • Primary and secondary languages weighted by byte volume and repo count, with trend detection
  • Frameworks and libraries detected from actual dependency files, not just repo language tags
  • Build tools, testing frameworks, linters, CI/CD platforms, containerization, infrastructure tooling
  • Project type classification (web frontend, backend, CLI, library, data science, devops, etc.)
  • Coding style signals (TypeScript preference, strict configs, monorepo patterns, commit conventions)
  • Domain expertise inferred from bio, topics, and repo descriptions
  • Activity level, contribution style (solo vs collaborative), and recent focus areas

The prompt builder then wraps all of this in a well-structured prompt with interpretive framing -- not just "this user has 40% TypeScript" but context about what that means for the AI generating their instructions. It also includes anti-boilerplate guidance so the AI doesn't pad the output with "write clean, readable code."

Target formats

Pick your AI tool before copying. The prompt adjusts its meta-instructions for each:

  • GitHub Copilot -- copilot-instructions.md format
  • ChatGPT -- Custom Instructions with the two-field format and character limits
  • Claude -- project-level or account-level custom instructions
  • Cursor -- .cursorrules format
  • Generic -- universal markdown for any AI tool

Running locally

git clone https://github.com/dvelton/instrucgen.git
cd instrucgen
npm install
npm run dev

Rate limits

The GitHub API allows 60 unauthenticated requests per hour. That's enough for a few analyses. If you hit the limit, the app supports an optional GitHub personal access token (stored in localStorage, never sent anywhere except github.com and raw.githubusercontent.com) for 5,000 requests/hour.

Dependency files are fetched via raw.githubusercontent.com, which has no rate limit, so framework detection works regardless.

Tech stack

TypeScript, Vite, Tailwind CSS. No backend, no database, no third-party services. Everything runs in your browser.

About

Enter a GitHub handle, get a comprehensive AI custom instructions prompt based on their public profile and activity

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors