Skip to content

instructions.md

Latest commit

 

History

History
305 lines (204 loc) · 4.51 KB

instructions.md

File metadata and controls

305 lines (204 loc) · 4.51 KB
Raw

GLAT Change Card + Supabase Integration Prompt

You are helping build a VS Code extension called GLAT (Global Local Agent Transport).

GLAT allows developers using AI coding agents (like GitHub Copilot) to share structured context about uncommitted changes before commits.

Your task is to implement the change card generation pipeline and Supabase storage layer.

Keep the implementation minimal and production‑like.

SYSTEM OVERVIEW

GLAT works by converting local code changes into Change Cards.

A change card represents:

  • what changed
  • who changed it
  • which files may be impacted

These cards are stored in Supabase so other developers can retrieve relevant context.

WORKFLOW

Broadcasting Changes

When a developer runs:

GLAT: Broadcast Local Changes

The extension should:

  1. Detect changed files using git
  2. Generate a change card
  3. Store the card in Supabase

Retrieving Context

When another developer runs:

GLAT: Prepare Context for Copilot

The extension should:

  1. Detect the active file
  2. Query Supabase for change cards impacting that file
  3. Inject relevant summaries into a Copilot prompt

CHANGE CARD STRUCTURE

Use the following schema:

{
  id: string
  author: string
  timestamp: string
  changed_files: string[]
  impacted_files: string[]
  summary: string
}

Example:

{
  "id": "uuid",
  "author": "kiko",
  "timestamp": "2026-03-14T19:00:00Z",
  "changed_files": ["backend/user.ts"],
  "impacted_files": ["frontend/UserCard.tsx"],
  "summary": "Backend user response field fullName was renamed to displayName"
}

SUPABASE SETUP

Use Supabase as the backend database.

Create a table:

change_cards

Schema:

id uuid primary key
author text
timestamp timestamptz
changed_files jsonb
impacted_files jsonb
summary text

This allows fast querying of impacted files.

SUPABASE CLIENT SETUP

Install Supabase client:

npm install @supabase/supabase-js

Initialize client:

import { createClient } from "@supabase/supabase-js"

const supabase = createClient(
  process.env.SUPABASE_URL!,
  process.env.SUPABASE_ANON_KEY!
)

BROADCAST COMMAND IMPLEMENTATION

Command ID:

glat.broadcastChanges

Steps:

  1. Get workspace root

Use:

vscode.workspace.workspaceFolders
  1. Run git command
git diff --name-only

Use Node:

child_process.exec

  1. Parse changed files into array.

  2. Generate impacted_files

For MVP:

impacted_files = changed_files
  1. Create change card object.

Example:

const card = {
  id: crypto.randomUUID(),
  author: "developer",
  timestamp: new Date().toISOString(),
  changed_files: changedFiles,
  impacted_files: changedFiles,
  summary: "Local code changes detected"
}
  1. Insert into Supabase.
await supabase
  .from("change_cards")
  .insert(card)
  1. Notify the user.
vscode.window.showInformationMessage("GLAT change card broadcasted")

RETRIEVAL COMMAND IMPLEMENTATION

Command ID:

glat.prepareContext

Steps:

  1. Get active editor
const editor = vscode.window.activeTextEditor
  1. Extract active file path
const activeFile = editor.document.fileName
  1. Query Supabase for relevant change cards
const { data } = await supabase
  .from("change_cards")
  .select("*")
  .contains("impacted_files", [activeFile])
  1. Ask the user for a prompt
vscode.window.showInputBox()
  1. Build an augmented prompt.

Example:

Task:
${userPrompt}

Relevant teammate changes:
${cards.map(c => "- " + c.summary).join("\n")}

Current file:
${fileContents}
  1. Open prompt in a new editor so the user can paste it into Copilot.
vscode.workspace.openTextDocument({
  content: prompt,
  language: "markdown"
})

vscode.window.showTextDocument(document)

KEY DESIGN PRINCIPLES

Keep the system simple.

Do not implement:

  • embeddings
  • dependency graphs
  • automatic Copilot integration
  • complex indexing

Use only:

  • git diff
  • change cards
  • Supabase storage
  • file‑level retrieval

SUCCESS CRITERIA

The feature works if:

  1. A developer runs GLAT Broadcast Local Changes.
  2. A change card is inserted into Supabase.
  3. Another developer runs GLAT Prepare Context for Copilot.
  4. Relevant change cards are retrieved.
  5. The extension generates a prompt containing teammate change summaries.

OPTIONAL IMPROVEMENTS (IF TIME PERMITS)

  • AI‑generated change summaries
  • AI prediction of impacted files
  • symbol‑level indexing
  • real‑time updates using Supabase subscriptions