Quick Context

Generate clean, LLM-ready markdown prompts from selected files and folders in VS Code.

Quick Context creates one or more quick-context-N.md documents containing, for each file:

1. File name
   Relative path

   (followed by the original file contents wrapped in a language-aware fenced code block when generated in VS Code)

You paste those documents directly into ChatGPT / other LLMs as “catch-up” context for new conversations.

Features

File & folder support

• Works on files, folders, or a mixed selection in the Explorer.
• Folders are processed recursively.
• Common build / meta directories are ignored:
  • node_modules, .git, .next, dist, build, out.

Language-aware code fences

• Uses the file extension to choose the fenced code language:
  • .ts → ts
  • .tsx → tsx
  • .js / .jsx → javascript / jsx
  • .json → json
  • .md → markdown
  • .css, .scss, .sass → css / scss
  • .html, .htm → html
  • .py → python
  • .rb → ruby
  • .php → php
  • everything else → text

This makes the generated prompt much easier to read inside LLM UIs.

Automatic splitting into multiple prompts

• Keeps each generated markdown document under a configurable character limit
  (INPUT_LIMIT_CHARS in src/extension.ts, default ≈ 60 000 characters).
• When the combined content would exceed that limit, it:
  • Starts a new document.
  • Continues with the next file.
• Never splits a single file across documents.

Generated documents are named:

• quick-context-1.md
• quick-context-2.md
• quick-context-3.md
• …

Safety guard on huge files

• If a single file’s markdown block (name + path + fenced code) would exceed the INPUT_LIMIT_CHARS limit by itself, the extension:
  • Aborts the operation.
  • Shows an error message indicating which file is too large.

This prevents you from accidentally trying to paste something that one LLM input cannot realistically handle.

Usage

1. In the Explorer:
   • Select one or more files, folders, or both.
2. Right-click the selection.
3. Choose Quick Context: Generate Catch-up Prompt.
4. One or more tabs open:
   • quick-context-1.md
   • quick-context-2.md
   • …

Each document contains a numbered list of files with their relative paths and fenced contents, for example (conceptually):

1. OrderForm.tsx
src/components/OrderForm.tsx

    // full contents of OrderForm.tsx
    // (wrapped in a TSX code fence in the generated markdown)

You can now copy the contents of quick-context-1.md (and subsequent files) directly into your LLM chat.

If you have no selection in the Explorer but an active editor file, the command uses the currently open file.

Command

• Quick Context: Generate Catch-up Prompt
  • Command ID: quickContext.generateFromSelection
  • Available from:
    • Explorer context menu (files and folders)
    • Editor context menu (active file, if nothing is selected in Explorer)

Configuration (developer-side)

There is currently no user-facing settings UI.
Two key constants live in src/extension.ts:

// Soft limit per generated markdown document.
// Used to keep each "prompt" under a typical ChatGPT input size.
const INPUT_LIMIT_CHARS = 60000;

// Directories ignored when recursing through folders.
const IGNORE_DIRS = ["node_modules", ".git", ".next", "dist", "build", "out"];

If you fork or customize the extension, adjust these constants, rebuild, and republish.

Limitations

• If a single file’s markdown block exceeds INPUT_LIMIT_CHARS, the entire operation is aborted and an error is shown.
• There is no per-workspace configuration yet; limits and ignore lists are compile-time constants in src/extension.ts.

Release Notes

1.0.1

• Added README and repository metadata.
• Clarified behavior and documented limits.

1.0.0

• Initial release.
• File/folder selection.
• Recursive folder traversal with ignore list.
• Language-aware code fences.
• Automatic splitting into quick-context-N.md with per-document size limit.
• Guard against single-file overflow.