live-canvas

name: live-canvas description: > Live visual canvas for agents. Use when running multi-step tasks (deploys, installs, batch jobs), gathering config from the user (forms), showing plan approval (flowchart + buttons), or displaying live dashboards (metrics, charts, tables). The canvas opens a browser panel that updates in real time as the agent pushes components over WebSocket. Users can click buttons or submit forms and the agent receives the action. Server auto-starts on first use. metadata: keywords: [canvas, visual, ui, dashboard, progress, form, plan, chart, board, task, deploy, config, interactive] port: 18793 templates: [progress_board, data_dashboard, config_form, plan_view]

Live Canvas Skill 🍀

Give any agent a live visual workspace. Instead of walls of markdown, push rich interactive panels—progress boards, dashboards, forms, and plan diagrams—that update in real time as work progresses. Users can click buttons and submit forms; the agent receives those actions as structured responses.

When to Use

Quick Start

import sys
sys.path.insert(0, '/opt/skills/live-canvas/claude/implementation')
from canvas import Canvas

c = Canvas()     # auto-generates session ID; auto-starts server
c.open()         # opens http://localhost:18793?session=<id> in browser

API Reference

c.render(components: list)                # full re-render
c.render_template(name, data)             # render a built-in template
c.update(node_id, changes)               # partial update (live, no flicker)
c.clear()                                # wipe canvas
c.open()                                 # open browser tab
action = c.wait_for_action(timeout=60)   # block until user clicks

wait_for_action returns a dict:

{"type": "action", "action_id": "approve", "data": {...}, "session_id": "...", "timestamp": "..."}
# or on timeout:
{"type": "timeout"}

Template 1: progress_board

Kanban board showing Todo / Running / Done columns with an overall progress bar.

Use for: deployments, installs, batch jobs, any multi-step workflow.

c.render_template("progress_board", {
    "title": "Deploy v2.1",
    "elapsed": "3m 12s",         # optional
    "steps": [
        {"name": "Pull repo",      "status": "done"},
        {"name": "npm build",      "status": "running"},
        {"name": "Restart svc",    "status": "pending"},
        {"name": "Health check",   "status": "pending"},
    ]
})

# Live update a step without full re-render:
c.update("step-1", {"status": "done"})    # index-based id
c.update("overall-progress", {"pct": 75})

Status values: done, running, pending, error, skip

Template 2: data_dashboard

Metrics row + optional line chart + optional table.

Use for: server status, Home Assistant devices, GitHub stats, monitoring.

c.render_template("data_dashboard", {
    "title": "Server Status",
    "metrics": [
        {"label": "CPU",  "value": "42%",  "trend": "up"},
        {"label": "RAM",  "value": "6.1 GB"},
        {"label": "Disk", "value": "78%",  "trend": "down"},
        {"label": "Svcs", "value": "4/4"},
    ],
    "chart": {                             # optional
        "label": "CPU last 1h",
        "labels": ["10m", "20m", "30m", "40m", "50m", "60m"],
        "datasets": [{"label": "CPU %", "data": [30, 45, 42, 55, 48, 42]}],
    },
    "table": {                             # optional
        "headers": ["Process", "CPU", "MEM"],
        "rows": [["node", "12%", "400 MB"], ["python", "5%", "120 MB"]],
    },
})

Template 3: config_form

Dynamic form rendered from a field schema. Returns submitted values.

Use for: collecting settings, rules, credentials, preferences from the user.

c.render_template("config_form", {
    "title": "Email Triage Rules",
    "description": "Configure how incoming emails are processed.",
    "fields": [
        {"name": "sender_filter", "label": "Sender filter", "type": "text",     "placeholder": "e.g. @spam.com"},
        {"name": "action",        "label": "Action",         "type": "select",   "options": ["Move", "Delete", "Flag"]},
        {"name": "notify",        "label": "Notify me",      "type": "checkbox", "default": True},
    ],
    "submit_label": "Apply Rules",
    "cancel_label": "Cancel",
})

action = c.wait_for_action(timeout=120)
if action["action_id"] == "submit":
    values = action["data"]   # {"sender_filter": "...", "action": "...", "notify": True}

Template 4: plan_view

Mermaid.js flowchart + Approve / Cancel buttons.

Use for: showing a plan before executing, getting explicit user approval.

c.render_template("plan_view", {
    "title": "Rotate SSL Certificates",
    "description": "This will renew and deploy 3 certificates.",
    "mermaid": """flowchart TD
  A[Check expiry] --> B{Expired?}
  B -->|Yes| C[Generate new cert]
  B -->|No| D[Skip]
  C --> E[Deploy to nginx]
  E --> F[Restart service]""",
    "approve_label": "Execute Plan",
    "cancel_label": "Cancel",
})

action = c.wait_for_action(timeout=300)
if action.get("action_id") == "approve":
    # proceed with execution
    pass

Building Custom Component Trees

For full control, pass a component tree to render():

c.render([
    {"type": "heading", "level": 2, "text": "My Dashboard"},
    {"type": "row", "children": [
        {"type": "metric", "id": "cpu-metric", "label": "CPU", "value": "42%"},
        {"type": "metric", "id": "ram-metric", "label": "RAM", "value": "8 GB"},
    ]},
    {"type": "progress", "id": "job-progress", "label": "Indexing…", "pct": 35},
    {"type": "button", "label": "Cancel Job", "action_id": "cancel", "variant": "danger"},
])

Component Quick Reference

Mobile Channel Behavior

When Foster is on Telegram or WebEx (mobile channels), canvas URLs must be optimized for phone viewports (375–430px wide).

Rules for mobile channels

• Append ?mobile=1 to every canvas URL sent via Telegram or WebEx
• The canvas HTML includes a CSS media query (@media (max-width: 480px)) AND a body.mobile-mode class applied when ?mobile=1 is set — both paths produce the same mobile-friendly layout
• Mobile layout: stacked cards (flex-column), 14px base font, 100% widths, min 44px tap targets
• All content fits within viewport — no horizontal scroll

Mobile CSS behaviour (auto-applied)

Channel detection in agent code

import os
channel = os.environ.get('WEE_CHANNEL', 'webui').lower()
mobile = channel in ('telegram', 'webex')

url = canvas.viewer_url()
if mobile:
    url += ('&' if '?' in url else '?') + 'mobile=1'

# Send url to user

Server Details

• Port: 18793 (override: CANVAS_PORT env var)
• Host: localhost by default (override: CANVAS_HOST env var — set to Tailscale IP for remote access)
• Bind: Server always binds to 0.0.0.0, so it accepts connections on all interfaces including Tailscale
• Auto-start: Canvas() starts canvas_server.py if not already running
• Auto-stop: Server stops after 30 minutes with no active WebSocket connections
• Session isolation: Each session_id has its own state; multiple agents can use the same server
• State persistence: Last rendered state restores on page refresh
• Install dependency: pip install websockets

Tailscale Access

To open the canvas on a remote device (phone, MacBook) via Tailscale:

export CANVAS_HOST=100.124.186.75   # lepbuntu Tailscale IP
python3 canvas.py
# Opens: http://100.124.186.75:18793/?session=...

Or set permanently in the agent's environment/.env.

The WebSocket in index.html also reads the host from the URL, so remote browsers connect to the correct Tailscale IP automatically.