VS Code Copilot Extension
Architecture-aware answers in VS Code — type @prism in Copilot Chat
Live Demo
What developers see in VS Code@prism should I refactor auth?
PRISM Intelligence
Based on your architecture data: auth.session has 3 drift entries and is involved in a cycle with middleware/csrf.ts. Refactoring is recommended but requires resolving the cycle first.
Risk: High · Estimated: 4.5 developer-days
Next step: Run `/amber/capability-deps` to see the full dependency chain before touching auth.
Why Microsoft loves this
Copilot Extensions are GitHub's newest ecosystem (launched 2024). PRISM is the first architecture intelligence extension — giving Copilot codebase-specific context that training data can never provide. While GitHub Copilot knows general patterns, PRISM knows your specific capabilities, drift, and cycles in real time.
Setup
Create a GitHub App for PRISM in your GitHub organization or account settings.
→ Create GitHub AppRequired permissions: Copilot Chat
In your GitHub App settings, under "Copilot", set the webhook URL to your PRISM instance:
Open VS Code → Copilot Chat → Extensions. Search for PRISMand authorize it. You'll need the GitHub App ID from step 1.
Open GitHub Copilot Chat in VS Code (⌘+I) and type @prism. PRISM will load your live architecture data and answer immediately.
Webhook URL
This is the endpoint GitHub sends Copilot Chat requests to. It returns SSE-streamed responses in the GitHub Copilot Extensions format.
What developers can ask @prism
@prism what's my architecture health?Score, grade, top issues, improvement actions
@prism is this file safe to refactor?Capability mapping, drift status, cycle involvement
@prism what capabilities does auth/login.ts belong to?AMBER capability membership, coverage depth
@prism what would break if I change this service?Downstream capabilities, dep chain risk
@prism how long would the auth refactor take?Developer-day estimate based on drift + complexity
@prism show me the riskiest capabilities right nowRanked list by drift × dependency depth
How PRISM answers
PRISM reads live architecture data from your AMBER scan and GREEN analysis — not training data.
File references from Copilot are mapped to AMBER capabilities automatically.
Responses cite specific drift counts, capability names, and risk scores from your codebase.
PRISM uses Claude with a cached architecture context for fast, consistent answers.