Skip to content
Back to skills

Ha Screenshot

ASecurity

Capture and visually verify how a Home Assistant dashboard, view, or Lovelace card looks. Use when the user wants to see/check a dashboard's appearance, or after editing Lovelace to confirm the change rendered correctly.

  • 3 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 20, 2026
ai-agentsgobash

Works with

  • cli
  • mcp

Security analysis

A100/100

Scanned October 20, 2026

npx -y skills add LayerTM/ClaudeInHA --skill ha-screenshot --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Ha Screenshot?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Ha Screenshot
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/layertm-ha-screenshot/badge)](https://www.skillsdirectory.com/skills/layertm-ha-screenshot)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: ha-screenshot
description: Capture and visually verify how a Home Assistant dashboard, view, or Lovelace card looks. Use when the user wants to see/check a dashboard's appearance, or after editing Lovelace to confirm the change rendered correctly.
---

Visually verify Home Assistant UI by capturing a screenshot, then **reading the PNG** to evaluate. Never claim a dashboard "looks right" without looking at a fresh screenshot.

## Primary: `ha-shot`

Screenshots an authenticated Lovelace dashboard to PNG. Requires the add-on's **HA Token** (Long-Lived Access Token) option to be set — if unset, `ha-shot` prints a hint; tell the user to set it in the add-on config.

```bash
ha-shot <dashboard-path> [output.png] [WIDTHxHEIGHT]
```

```bash
# Default dashboard, first view
ha-shot /lovelace/0 /tmp/dash.png 1280x800

# A named view within the default dashboard
ha-shot /lovelace/lights /tmp/lights.png 1280x800

# A specific dashboard by its url_path
ha-shot /lovelace-energy/0 /tmp/energy.png 1280x800

# Mobile-width check
ha-shot /lovelace/0 /tmp/mobile.png 414x896
```

Then evaluate it — **Read the PNG** (`/tmp/dash.png`) so you actually see the result. Judge layout, card contents, empty/error states, and whether the edit took effect.

### Path reference
- `/lovelace/0` — default dashboard, view index `0` (also `/1`, `/2`, …).
- `/lovelace/<view-path>` — a view by its `path:` slug.
- `/<dashboard-url_path>/<view>` — a non-default dashboard (its `url_path` from config).

### Sizing
- Desktop: `1280x800` (default-ish). Wide dashboards: `1920x1080`.
- Mobile: `414x896`. Use it to catch responsive/column breakage.

## Iterating on a card
After changing a card in the Lovelace config (edit the `!include`d dashboard file, 2-space YAML, never tabs):

1. If the change is in `configuration.yaml`-loaded (YAML-mode) Lovelace, run `ha core check` then `ha core reload` (prefer reload over restart). Storage-mode dashboards update live — no reload needed.
2. Re-shoot the same path to the same output file.
3. Read the new PNG and compare against the goal. Repeat until correct.

## Interactive / stateful cases: Playwright MCP
Use when you must click, hover, toggle, open a dialog, wait for data, or capture a state `ha-shot` can't reach.

```
browser_navigate  → $HA_URL<path>   e.g. $HA_URL/lovelace/0   (echo $HA_URL first)
browser_snapshot        # accessibility tree — inspect structure / find elements
browser_take_screenshot # save PNG, then Read it
```

The browser is preinstalled. Navigate, interact (click/hover/wait), screenshot, then Read the PNG to evaluate. Prefer `ha-shot` for plain "show me this dashboard" — reach for Playwright only when interaction is required.

## Rules
- Always **Read the captured PNG** — a screenshot you don't look at proves nothing.
- Take a screenshot *after* any Lovelace edit to confirm it rendered.
- `ha-shot` failing with a token hint → HA Token add-on option is unset; do not fall back to guessing.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.