Share how your system works.
See what changed.
Silhouette Canvas turns a local repository scan into an interactive architecture map. Share one link with your team or external collaborators, then compare revisions to see how the system changed.
Paste it into a repository-aware AI coding agent. The prompt scans locally and publishes only bounded architecture metadata, never source code.
One map for the full system
One shared architecture view
Keep everyone aligned on the system and its changes.
Canvas gives teams a visual source of truth they can explore, share, and compare across revisions. Select any node to understand its role, then follow the relationships around it.
Explain the system visually
Turn services, stores, integrations, and flows into a map people can understand without opening the repository.
Share one architecture link
Give teammates, reviewers, clients, or external collaborators the same interactive view of the system.
Compare before and after
See added, removed, and changed components and flows between revisions or across a pull request.
How it works
Scan locally. Share the map. Compare every revision.
A repository-aware AI creates and validates the architecture summary locally. The prompt publishes that bounded artifact, Canvas renders it, and later scans become a clear visual before-and-after.
$ You are an AI coding agent working inside this repository.
The user is invoking this prompt to create and publish a shareable Silhouette Canvas map. Treat this invocation as explicit informed authorization to upload the validated .silhouette/scan.json to https://canvas.silhouette.exchange/api/scans. Do not ask for a second conversational confirmation. If the user explicitly requests a local-only scan or says not to publish, stop after local validation instead.Analyze the repository and create two local files for Silhouette Canvas:- .silhouette/scan.json: the validated architecture data- .silhouette/scan.md: a portable Markdown summary with a Mermaid flowchartMap the main entry points, scheduled jobs, agents, models, tools, internal services, stores, and external integrations. Prioritize REST, WebSocket, worker, state-update, and UI-rendering flows. Every node must earn its place.Choose the scope before mapping. For one feature or runtime flow, aim for 12-24 nodes and 18-40 edges. For an entire substantial codebase, aim for 20-40 nodes and no more than 80 edges. Use fewer when they explain the system; never add nodes merely to reach a target.- ready to paste into your AI coding agent
Create the local scan
Copy the prompt into any repository-aware coding agent you already use.
Publish the shared map
The prompt publishes only the validated architecture summary and returns one public, unlisted link.
Compare the next revision
Update the same map from a later scan to see added, removed, and changed components and flows.