contextify-ai

AI tools read your code. They can't read your mind. Auto-generate .context.md files at commit time so both humans and AI tools understand your components.

What it does

contextify-ai hooks into your git commit workflow, analyzes changed components using AST parsing, asks what you intended, and generates a structured .context.md file next to each component.

One file. Two audiences.

• Top half — prose for humans: purpose, business rules, edge cases, design decisions
• Bottom half — YAML for AI tools: props, state, dependencies, render conditions

src/
  PaymentForm/
    PaymentForm.tsx
    PaymentForm.test.tsx
    PaymentForm.module.css
    PaymentForm.context.md    ← generated

AI tools like Claude Code, Cursor, and Copilot pick up .context.md files automatically through colocation. No plugins. No configuration. Just convention.

→ Format reference · How it works

Install

npm install contextify-ai --save-dev

Quick Start

# Initialize in your project (interactive wizard)
npx contextify init

# Or if installed globally
npm install -g contextify-ai
contextify init

The setup wizard will:

1. Ask for your LLM provider (Ollama, GitHub Models, Gemini, Claude, or OpenAI)
2. Let you pick a model
3. Configure include/exclude file patterns
4. Select hook mode (pre-commit or post-commit)
5. Set up AI tool integrations (Claude Code, Cursor, Copilot, Windsurf)
6. Install git hooks
7. Optionally generate .context.md for existing files

Supported Providers

| Provider | Config value | Cost | Env Variable | Setup | | --- | --- | --- | --- | --- | | Ollama | ollama | Free | None needed | ollama.com — ollama pull llama3 | | GitHub Models | github | Free | GITHUB_TOKEN | github.com/settings/tokens — no special scopes | | Google Gemini | gemini | Free tier | GEMINI_API_KEY | aistudio.google.com/apikey | | Claude | claude | Paid | ANTHROPIC_API_KEY | console.anthropic.com | | OpenAI | openai | Paid | OPENAI_API_KEY | platform.openai.com |

Configuration

Running contextify init creates a .contextifyrc file in your project root:

{
  "provider": "ollama",
  "model": "llama3",
  "mode": "pre-commit",
  "include": [
    "**/index.ts",
    "**/index.tsx",
    "**/index.js",
    "**/index.jsx"
  ],
  "exclude": [
    "**/*.test.*",
    "**/*.spec.*",
    "**/*.stories.*",
    "**/*.story.*",
    "**/*.context.md",
    "**/*.d.ts",
    "**/node_modules/**",
    "**/dist/**",
    "**/build/**"
  ],
  "output": "colocated",
  "concurrency": 5,
  "smartDiff": true,
  "commitTags": true,
  "tools": {
    "claudeCode": false,
    "cursor": false,
    "copilot": false,
    "windsurf": false
  },
  "systemPrompt": ""
}

You can also use .contextifyrc.json, .contextifyrc.yaml, .contextifyrc.js, or contextify.config.js.

Custom System Prompt

By default, contextify-ai uses a built-in prompt that produces the standard .context.md format. To use your own:

{
  "systemPrompt": "./my-prompt.md"
}

The path is resolved relative to your project root. If the file is not found, contextify falls back to the built-in prompt with a warning. See docs/format.md for what the default prompt produces and what to preserve if you customize it.

How it Works

• Smart diff — parses your code with Babel, hashes structural elements (props, exports, hooks), and skips the LLM call when nothing meaningful changed. 50–70% fewer LLM calls on active codebases.
• Developer intent — prompts you for what changed and why at commit time, or accepts --message. Your explanation feeds the LLM as the primary source for edge cases, business rules, and the decision log.
• Hook ordering — runs before lint-staged so generated files are staged and lintable. Works with plain .git/hooks and Husky.
• Commit tagging — tags every commit with [context: generated|updated|no-change|skipped|error].

→ Full details in docs/how-it-works.md

CLI Commands

contextify init

Interactive setup wizard. Creates config, installs git hooks, sets up AI tool integrations.

contextify generate

Manually generate .context.md files outside of the commit flow.

contextify generate --dry-run               # preview without calling the LLM
contextify generate src/components/Button.tsx
contextify generate                         # all files matching your configured scope
contextify generate --force                 # ignore smart diff
contextify generate --concurrency 3

contextify audit

Check your project's context coverage.

contextify audit           # find files missing .context.md
contextify audit --stale   # also flag stale context files

contextify hook

Called automatically by git hooks. Supports flags:

contextify hook -m "refactored auth to use JWT, session edge cases no longer apply"
CONTEXTIFY_SKIP=true git commit -m "ci: update deps"   # skip context for this commit

contextify reset

Delete .context.md files. Asks for confirmation unless --force.

contextify reset src/components/PaymentForm.tsx   # single file
contextify reset                                   # entire project
contextify reset --force                           # skip confirmation

contextify regen

Force-regenerate .context.md files, passing the existing context back to the LLM so it can preserve accurate edge cases while updating what changed.

contextify regen src/components/PaymentForm.tsx
contextify regen -m "switched from session tokens to JWT across the board"
contextify regen --concurrency 3

Full refresh workflow:

contextify reset --force   # delete all .context.md
contextify regen -m "..."  # regenerate from scratch, guided by your message

AI Tool Integration

During contextify init, you can select which AI tools you use. contextify-ai adds references so each tool knows to look for .context.md files:

| Tool | What gets created | | --- | --- | | Claude Code | Adds context reference to CLAUDE.md | | Cursor | Creates .cursor/rules/contextify.mdc with auto-attach rules | | Copilot | Updates .github/copilot-instructions.md | | Windsurf | Updates .windsurfrules |

A project-wide index is auto-generated at .contexts/index.md — a table of contents listing every context file, categorized by type.

Why Colocation

The .context.md file sits next to the component it describes — same convention as .test.tsx, .module.css, and .stories.tsx.

AI tools with file system access discover these files without any integration work. Claude Code reads PaymentForm.tsx, checks the same directory for PaymentForm.context.md, and understands the props, state, dependencies, and business rules before writing a single line.

Convention over configuration.

Comparison

| Feature | contextify-ai | Repomix | code-contextify | JSDoc | AI commit generators | | --- | --- | --- | --- | --- | --- | | Per-component | ✅ | ❌ | ❌ | ✅ | ❌ | | Commit-hooked | ✅ | ❌ | ❌ | ❌ | ✅ | | LLM-powered | ✅ | ❌ | ✅ | ❌ | ✅ | | Dual-audience | ✅ | ❌ | ❌ | ❌ | ❌ | | Smart diff | ✅ | ❌ | ❌ | ❌ | ❌ | | Developer intent | ✅ | ❌ | ❌ | ❌ | ❌ | | Provider-agnostic | ✅ | N/A | Partial | N/A | Partial | | Business context | ✅ | ❌ | Partial | ❌ | ❌ |

Roadmap

• VS Code extension for inline context previews
• MCP server integration
• CI/CD pipeline validation
• Python and Go support
• Context file diffing in PR reviews

Research

This tool is backed by a research paper:

"Automated Context Generation for AI Code Assistants: An LLM-Powered Framework for Developer Intent Capture and Documentation Automation" Althaf Khan Pattan, Independent Researcher, 2026

Contributing

Contributions are welcome. Open an issue first to discuss what you'd like to change.

git clone https://github.com/AlthafPattan/contextify-ai.git
cd contextify-ai
npm install
npm test

See docs/how-it-works.md for testing against a real project.

License

MIT

If this tool helps you, consider giving it a ⭐ on GitHub. Every star, fork, and contribution helps keep this project alive.