Agent Canvas

What it does

Agent Canvas provides a CLI for creating and managing TLDraw whiteboards: geometric shapes, arrows, markdown notes, screenshots, code diffs and other high-signal visual artifacts. It lets agents produce diagrams, incident timelines, PR review boards, and interactive previews programmatically or during an agent-run workflow. The CLI can read and write shapes, capture screenshots, and embed interactive HTML or test results so boards become a replayable evidence trail. Board operations that modify state require the board to be open in a browser (WebSocket relay).

When to use it

Use Agent Canvas whenever an agent needs to explain architecture, present a step-by-step incident postmortem, show focused code diffs during a review, or capture deployment/test outputs as visual evidence. Typical triggers include: documenting a debugging timeline, creating architecture or flow diagrams, preparing a demo board for stakeholders, or assembling PR review artifacts alongside test results and logs.

What's included

• Scripts: CLI commands for board and shape management (create, read, update, delete), screenshot capture, and comments threading. (has_scripts: false in repo manifest, but CLI usage is documented in the skill body)
• References: none bundled in the skill directory (has_references: false).
• Instructions: detailed guidance on when to use specific shape types (geo, markdown, snippet, code-diff, ai-terminal, stack-trace, test-results, web-preview), tempId usage for cross-references, and practical examples for PR review and incident workflows.

Compatible agents

Compatible with agents that can run CLI tools and manage files or shell commands (Copilot/Codex-like agents and local CLI-capable agents). The skill assumes an environment where the agent can launch or communicate with a browser relay to operate TLDraw boards.