agent-canvas（Agent 画布）

Agent Canvas

A CLI tool to interact with an Excalidraw canvas for creating diagrams and visualizations.

Installation

Before using, check if CLI is installed:

which agent-canvas && agent-canvas --version

• If not installed: Ask the user which package manager they prefer (bun or npm), then install: bun add -g @agent-canvas/cli@0.13.0 # or npm install -g @agent-canvas/cli@0.13.0
• If installed but version differs from 0.13.0: Upgrade using the same package manager:

- Path contains .bun → bun add -g @agent-canvas/cli@0.13.0 - Otherwise → npm install -g @agent-canvas/cli@0.13.0

• After install/upgrade: Verify with agent-canvas --version to confirm version is 0.13.0

Quick Start

1. Start the canvas (opens in browser): agent-canvas start &
2. Use CLI commands to draw on the canvas.

Port Configuration

Default ports: 39820 (WebSocket), 39821 (HTTP). If ports conflict, configure via:

agent-canvas config set port 39820         # Set WebSocket port
agent-canvas config set http-port 39821    # Set HTTP port
agent-canvas config list                   # Show all config (value + source)
agent-canvas config get port               # Show current port
agent-canvas config reset port             # Reset to default

Config is saved to ~/.agent-canvas/config.json. One-time override via start:

agent-canvas start --port 8000 --http-port 8001 &

Priority: --port flag > env var AGENT_CANVAS_WS_PORT > config file > default.

Commands Reference

Start Canvas

agent-canvas start &                  # Start server in the background (will close automatically when no action for a while)

Load File

agent-canvas load file.excalidraw     # Load .excalidraw file into current canvas

When loading from file: Remember the file path and save back to it with agent-canvas save <original-file>.

Canvas Management

The canvas supports multiple canvases. Each canvas is stored independently and can be switched between.

agent-canvas list                     # List all canvases ([U]=User active, [A]=Agent active, [F]=Folder)
agent-canvas new -n "Name" [--use]    # Create new canvas, optionally switch to it
agent-canvas use "Name"               # Switch to canvas by name
agent-canvas rename "New Name"        # Rename current canvas

Folder Management — Organize canvases into folders:

agent-canvas create-folder -n "Name"              # Create a new folder
agent-canvas delete-folder "Name"                  # Delete folder (canvases become ungrouped)
agent-canvas move-to-folder "Canvas" "Folder"      # Move canvas into a folder
agent-canvas move-to-folder "Canvas" --ungrouped   # Remove canvas from its folder

Notes:

• Canvas names are case-insensitive and must be unique
• Delete canvases via UI (hover over canvas in sidebar, click "..." menu)
• Each canvas has its own scene data; switching automatically saves current canvas
• Deleting a folder does NOT delete the canvases inside it — they become ungrouped

Add Text

agent-canvas add-text -t "<text>" --ax <x> --ay <y> [options]

• Options: --font-size <size>, --text-align <left|center|right>, -a/--anchor <anchor>, --stroke-color <hex>, -n/--note <text>
• Font sizes: S=16, M=20 (default), L=28, XL=36
• Anchor (-a): Since text size is unknown until rendered, anchor gives you precise positioning control by specifying which point of the text bounding box aligns to (--ax, --ay). Default: bottomLeft. Anchor Common Text Types topLeft Badge, Tag, Icon label topCenter Subtitle, Description below shape topRight Timestamp, Version, Status leftCenter Side annotation (right of shape) center Centered title, Main label rightCenter Side annotation (left of shape) bottomLeft Footnote, Note bottomCenter Title, Header above shape bottomRight Footnote, Signature
• Returns: Text created (id: <id>, x: <x>, y: <y>, <width>x<height>) — actual top-left position and dimensions for precise layout

Add Drawing Elements

All drawing commands share common style options:

• Stroke: --stroke-color <hex> (default: #1e1e1e), --stroke-width <1-4> (default: 2), --stroke-style <solid|dashed|dotted> (default: solid)
• Fill (shapes only): --background-color <hex> (default: transparent), --fill-style <solid|hachure|cross-hatch> (default: solid)
• Meta: -n/--note <text> - semantic description for the element. Use liberally - notes help understand diagram intent when reading back later.
• Animated: --animated - auto-scroll viewport to show the newly added element. Small elements zoom in, large elements zoom out, normal elements scroll minimally.

Recommended Colors (from Excalidraw palette):

Shapes

agent-canvas add-shape -t <type> -x <x> -y <y> [-w <width>] [-h <height>] [-l <label>]

• Types: rectangle, ellipse, diamond
• Use -l/--label to add text inside the shape (fontSize: 16 by default), --label-font-size <n> to adjust
• Label color inherits from --stroke-color by default; use --label-stroke-color <hex> to override
• Returns: Shape created (id: <id> x=<x> y=<y> w=<width> h=<height>) — actual dimensions after auto-sizing for labels

⚠️ Label Sizing - CRITICAL: Calculate BEFORE drawing

If shape size is too small, Excalidraw auto-expands, breaking arrow coordinates. You MUST:

1. Calculate minimum dimensions using the formulas below
2. Use the calculated values directly — NEVER estimate or use smaller values

Step 1: Calculate text dimensions (fontSize=16 by default)
  textWidth = charCount × fontSize × 0.6  (English/numbers)
  textWidth = charCount × fontSize        (CJK characters)
  textHeight = lineCount × fontSize × 1.35

Step 2: Calculate minimum shape size (use these values, not smaller!)
  rectangle: width = textWidth + 50,  height = textHeight + 50
  ellipse:   width = textWidth × 1.42 + 55,  height = textHeight × 1.42 + 55
  diamond:   width = textWidth × 2 + 60,  height = textHeight × 2 + 60

Example: Label "Message Queue" (13 chars) in ellipse

textWidth = 13 × 16 × 0.6 = 124.8
textHeight = 1 × 16 × 1.35 = 21.6
ellipse width = 124.8 × 1.42 + 55 = 232
ellipse height = 21.6 × 1.42 + 55 = 86
→ Use: -w 232 -h 86 (or round up to -w 240 -h 90)

Tip: For long labels, insert \n manually, then recalculate with updated lineCount.

Lines & Arrows

agent-canvas add-line -x <x1> -y <y1> --end-x <x2> --end-y <y2>
agent-canvas add-arrow -x <x1> -y <y1> --end-x <x2> --end-y <y2>

• Arrow-specific: --start-arrowhead, --end-arrowhead (arrow, bar, dot, triangle, diamond, none)

Arrow Types (--arrow-type):

Intermediate Points (--via): Use --via to specify intermediate points as absolute coordinates in format "x1,y1;x2,y2;...":

# Round arrow: one control point determines curve direction
# Vertical arrow curving left (control point at x=50, left of the line)
agent-canvas add-arrow -x 100 -y 100 --end-x 100 --end-y 300 --arrow-type round --via "50,200"

# Elbow arrow: multiple points for 90° turns
# Loop back pattern: down → left → up (for flowchart iterations)
agent-canvas add-arrow -x 175 -y 520 --end-x 175 --end-y 280 --arrow-type elbow --via "120,520;120,280"

Tips:

• For round: curve bends toward the control point (offset from straight path)
• For elbow: points define the corners of the 90° path

Polygon

agent-canvas add-polygon -p '[{"x":0,"y":0},{"x":100,"y":0},{"x":50,"y":100}]'

Image

agent-canvas add-image -f <path> -x <x> -y <y> [-w <width>] [-h <height>] [-n <note>]

• Supported formats: PNG, JPEG, GIF, SVG, WebP
• Width/height default to original image dimensions; specify one to scale proportionally
• Image data is embedded as base64 in the canvas (stored in browser IndexedDB)
• Returns: Image added (id: <id>, x: <x>, y: <y>, <width>x<height>)

Manipulate Elements

agent-canvas delete-elements -i <id1>,<id2>,...
agent-canvas rotate-elements -i <id1>,<id2>,... -a <degrees>
agent-canvas move-elements -i <id1>,<id2>,... --delta-x <dx> --delta-y <dy>
agent-canvas resize-elements -i <id1>,<id2>,... [--top <n>] [--bottom <n>] [--left <n>] [--right <n>]
agent-canvas group-elements -i <id1>,<id2>,...
agent-canvas ungroup-element -i <id>

Resize Elements (resize-elements): Expand or contract element edges (rectangle, ellipse, diamond, image). Values are in element's local coordinate system (respects rotation).

Examples:

# Expand bottom edge by 50px (increase height)
agent-canvas resize-elements -i abc123 --bottom 50

# Expand both right and bottom (like dragging bottom-right corner)
agent-canvas resize-elements -i abc123 --right 50 --bottom 30

# Contract left edge by 20px (decrease width)
agent-canvas resize-elements -i abc123 --left -20

# Expand all sides uniformly
agent-canvas resize-elements -i abc123 --top 25 --bottom 25 --left 25 --right 25

Read Scene

agent-canvas read                # TOON format (compact, ~7% of JSON size)
agent-canvas read --with-style   # Include stroke/bg colors
agent-canvas read --json         # Raw Excalidraw scene JSON

TOON output structure:

shapes[N]{id,type,x,y,w,h,angle,labelId,note}       # rectangle, ellipse, diamond, polygon
lines[N]{id,type,x,y,endX,endY,via,angle,note}      # line, arrow
labels[N]{id,containerId,content,x,y,w,h}           # text bound to shapes (via labelId)
texts[N]{id,content,x,y,w,h,angle,note}              # standalone text elements
groups[N]{id,elementIds}                             # element groupings

• labelId in shapes links to id in labels
• via shows intermediate points in same format as --via input ("x1,y1;x2,y2" or null if none)
• --with-style adds stroke, bg fields
• --json returns full Excalidraw format (use with jq to query specific elements)

Save, Export and Clear

agent-canvas save file.excalidraw
agent-canvas export -o out.png [--scale 2] [--dark] [--no-background]
agent-canvas clear                # Clear all elements from the canvas, use with caution!

Note: Before running clear, ask the user if they want to save or export the current canvas first. ⚠️ALWAYS prefer agent-canvas new over agent-canvas clear only use clear when user has confirmed!⚠️

Design Philosophy

You are a perfectionist. If it looks slightly off, it IS off. Fix it.

Core principle: Consistency reflects meaning.

• Same relationship → Same alignment & spacing

- Elements with the same relationship should share identical alignment - Gaps between same-level elements should be equal throughout - Snap to grid (e.g., 10px units) for precision

• Same type → Same color & size

- Same-type nodes share identical dimensions - Same color = same meaning; max 3-4 colors; less is more - Important elements = larger size

• Details matter

- Arrows connect precisely to shape edges - Review via export and fix any imperfections

IMPERATIVE GUIDE

1. Coordinates: Origin (0,0) is top-left. X→right, Y→down. Colors in hex (#FF5733) or transparent.
2. Workflow: Read canvas → Plan layout → Draw shapes → Add arrows/lines(if necessary) → Adjust.

- IMPORTANT: Canvas content is auto-saved to browser localStorage. Always run agent-canvas read first to check for existing content before drawing. - If old content exists, ask the user whether to: (a) continue editing, (b) clear and start fresh, or (c) save/export first then clear. - Shapes define the layout and provide exact coordinates - Arrow endpoints depend on shape positions — drawing arrows first leads to misalignment - Adjust: After initial draft, run read and export to review. Check against Design Philosophy: - Alignment issues? → move-elements to snap to grid - Inconsistent spacing? → move-elements to equalize gaps - Overlapping elements? → move-elements or delete-elements and redraw - Wrong sizes? → delete-elements and redraw - Misaligned arrows? → delete-elements and redraw with correct endpoints - Container size issue? → resize-elements to adjust to perfect size

1. Progressive Canvas Reading:

- read - Start here. Compact TOON format (~7% of JSON size) - read --with-style - Add color info when styling matters - export -o canvas.png + view image - For visual/spatial understanding - read --json | jq '.elements[] | select(.id=="<id>")' - Query specific element details

1. Batch Commands: Chain with && for efficiency. DO NOT WRITE BASH COMMENT IN DRAWING COMMANDS agent-canvas add-shape -t rectangle -x 100 -y 100 -l "A" && \ agent-canvas add-shape -t rectangle -x 300 -y 100 -l "B" && \ agent-canvas add-arrow -x 220 -y 130 --end-x 300 --end-y 130

Drawing Tutorials

Before drawing, identify the diagram type and check for tutorials:

1. Determine what type of diagram the user wants (flowchart, architecture, mindmap, UI mockup, etc.)
2. Read references/REFERENCE.md — this is the tutorial index listing all available diagram tutorials
3. If a matching tutorial exists: Read the specific tutorial FIRST before drawing — tutorials contain type-specific rules, layout patterns, and best practices
4. Apply the tutorial guidelines while drawing