Agent Canvas
Agent Canvas: A Whiteboard for Coding Agents When Claude Code investigates bugs, reviews PRs, or plans features, it dump walls of text.
Core trick
Agents draw shapes and diagrams onto a shared whiteboard via a CLI, with threaded inline comments anchored to markdown shapes.
Tags
Description
Agent Canvas: A Whiteboard for Coding Agents
When Claude Code investigates bugs, reviews PRs, or plans features, it dump walls of text. This turns that into visual whiteboards where it can draw diagrams, organize evidence spatially, and let humans iterate through inline comments.
1. PR Review Boards: Agents split diffs into logical chunks (not file dumps), group related changes, capture before/after screenshots with Playwright
2. Iterative Planning: Markdown shapes with Google Docs-style threaded comments let you iterate on plans inline. Agents draw freeform editable diagrams showing architecture and sequence flows.
3. Deployment Debugging: Visualize using MCP servers (Vercel, Sentry) with stack traces, terminal logs, test results as interactive components and see how the agent worked over time
4. Codebase Onboarding: Agents draw diagrams with code snippets at different abstractions, letting engineers explore interactively from high-level overview to implementation details.
README excerpt
# Agent Canvas A whiteboard CLI for coding agents like Claude Code, Codex, OpenCode, Amp, Gemini CLI, and more. Use the `agent-canvas` CLI to orchestrate swarms of agents on the same board, compare multiple implementations of the same feature side-by-side, and keep each agent isolated with git worktrees while you pick the best approach. ## Main Use Cases ### 1) Architecture planning and feature iteration Ask a coding agent to draw architecture and sequence diagrams so you can: - plan architecture before implementation - understand how an existing feature works end-to-end - iterate on design and flow changes quickly  ### 2) PR review with logical diff chunks and tested screenshots Ask a coding agent to lay out PR changes as small, logical chunks of related diffs, then attach screenshots of the tested feature (for example with Playwright or agent-browser).  ## Prerequisites - [Bun](https://bun.sh/) (v1.0+) ## Quick Start (Linked CLI) Build everything and link the `agent-canvas` command globally: ```bash bun install bun run build:link ``` This runs `build` (shared, server, web, cli), bundles the web app into the CLI, and links it via `bun link`. ### Add the Skill to Claude Code The `skills/` directory contains a skill that teaches Claude Code how to use agent-canvas. Skills are directories with a `SKI