Canvas Cowork

Gives Claude a live cursor on a shared spatial canvas where you can generate images, video, text, and agent responses from the CLI while the user watches in their browser. The core insight is velocity: use submit-batch for independent prompts, chain with --follow for iterative work, and spawn parallel subagents for bulk operations. Mode switching is inline, tree structure is intentional (branches vs chains matter), and you're expected to bring results back with taste, not just say "check the canvas." Most useful when you need visual generation with structure, or when the user is mobile and you're their eyes. The --bot flag and recall system mean past work becomes shared memory across sessions.

Install to Claude Code

npx -y skills add flowith-ai/canvas-cowork --skill canvas-cowork --agent claude-code

Installs to .claude/skills