work-canvaslisted
Install: claude install-skill JingbiaoMei/work-canvas-skill
# work-canvas
Produce **one self-contained HTML file** that renders an agent's work for a human to understand and
act on. The output is a capture of work (a report/board), not an application.
## Workflow
1. **Pick the artifact type** and read its reference:
| The user wants to… | Type | Read |
|---|---|---|
| See where a long/messy task stands, what broke, what's next | **Progress / status report** | `references/progress-report.md` |
| Close out a `/goal` command session with a status artifact | **Progress / status report** | `references/progress-report.md` |
| Review analysis or a plan, comment, and decide | **Review / decision memo** | `references/review-decision.md` |
| Weigh options head-to-head and pick a winner | **Comparison / leaderboard** | `references/comparison.md` |
If it's a blend (e.g. a progress report that ends in a decision), lead with the primary type and
borrow blocks from the other.
2. **Write the body.** Copy `assets/starter.html` and fill the header, the **What needs your input**
box, and the body sections from the reference's component recipes. Leave the `[[PASTE base.css]]`
and `[[PASTE interactions.js]]` markers in place; reference local images with `[[IMG path]]`.
3. **Inline into one self-contained file.**
- **Default (cheapest):** `python assets/assemble.py body.html canvas.html` — inlines + minifies
`base.css` and `interactions.js` and embeds `[[IMG]]` images as `data:` URIs. The model never
reads