Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create or register a canvas extension in the awesome-copilot repository. Use when asked to scaffold a new canvas extension, create its plugin.json, add a reusable extension to one or more plugins, or migrate extension metadata. Extensions are reusable source under extensions/; shippable plugin manifests belong under plugins/.
.claude/skills/github-create-canvas-extension/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 180% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 305% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 129% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 272% | 0% |
Use this skill only for the github/awesome-copilot repository. Canvas extensions are reusable source components. They do not have a plugin.json under extensions/.
Before creating files, ask for each missing value:
1.0.0)MIT when appropriate)extension.mjs. If not, create a minimal entrypoint only when the user provides enough implementation details; otherwise create the directory and leave an explicit TODO.assets/preview.png path or ask the user to add it. Do not invent a binary image or silently use a misleading placeholder.plugins/<extension-id>../extensions/<extension-id> to that plugin's extensions.com.github.awesome-copilot.extensions.For a new extension plugin, create this structure:
textextensions/<extension-id>/ ├── extension.mjs └── assets/ └── preview.png plugins/<extension-id>/ ├── plugin.json └── README.md
The extension source may contain additional files such as package.json, canvas assets, or supporting modules. Keep all reusable implementation files under extensions/<extension-id>/.
Create plugins/<extension-id>/plugin.json with this shape:
json{ "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json", "name": "<extension-id>", "description": "<description>", "version": "1.0.0", "author": { "name": "<author>" }, "repository": "https://github.com/github/awesome-copilot", "license": "MIT", "keywords": [ "<keyword>" ], "extensions": { "com.github.copilot": { "logo": "assets/preview.png" }, "com.github.awesome-copilot": { "extensions": [ "./extensions/<extension-id>" ] } } }
Keep Agent Plugins fields at the manifest top level. Repository composition belongs only under extensions.com.github.awesome-copilot; do not put agents, commands, hooks, or skills at the top level or directly under extensions. MCP servers are declared in mcp.json at the plugin root, never in plugin.json. Do not add x-awesome-copilot, standalone, or other repository-specific top-level fields.
For an existing parent plugin, create or update:
textplugins/<parent-plugin>/plugin.json (`extensions.com.github.awesome-copilot.extensions`)
Its extensions property must contain sorted repository-relative paths:
json{ "extensions": [ "./extensions/<extension-id>" ] }
Do not copy the extension source into the parent plugin. Materialization resolves the IDs from the root extensions/ directory, so the same source can be bundled by multiple plugins.
extensions/<extension-id> and plugins/<extension-id> do not already conflictextensions.com.github.awesome-copilot.extensions for every parent plugin that should ship the extension. Keep paths alphabetically sorted and unique.extensions/<extension-id>/.github/plugin/plugin.json.bash npm run plugin:validate npm run build npm run website:data
When migrating an existing extension:
plugins/<extension-id>/plugin.json.extensions.com.github.copilot.logo shape.extensions/<extension-id>.extensions.com.github.awesome-copilot.extensions.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 14,388 | 16,338 | +14% | 1 | 1 | 0% | 383 | 1,642 | +329% | 0 | 0 | — |
case-02 | fail→fail | 14,325 | 15,437 | +8% | 1 | 1 | 0% | 298 | 1,627 | +446% | 0 | 0 | — |
case-03 | fail→fail | 14,606 | 16,240 | +11% | 1 | 1 | 0% | 338 | 1,591 | +371% | 0 | 0 | — |
case-04 | fail→pass | 12,652 | 18,719 | +48% | 1 | 1 | 0% | 1,477 | 4,130 | +180% | 0 | 0 | — |
case-05 | fail→pass | 16,271 | 5,642 | -65% | 1 | 1 | 0% | 2,196 | 2,402 | +9% | 0 | 0 | — |
case-06 | fail→pass | 9,407 | 15,163 | +61% | 1 | 1 | 0% | 843 | 3,416 | +305% | 0 | 0 | — |
case-07 | fail→fail | 13,749 | 11,127 | -19% | 1 | 1 | 0% | 1,839 | 1,833 | -0% | 0 | 0 | — |
case-08 | fail→fail | 13,559 | 19,405 | +43% | 1 | 1 | 0% | 1,782 | 2,155 | +21% | 0 | 0 | — |
case-09 | fail→pass | 7,723 | 16,622 | +115% | 1 | 1 | 0% | 1,600 | 3,662 | +129% | 0 | 0 | — |
case-10 | fail→pass | 7,689 | 7,689 | 0% | 1 | 1 | 0% | 484 | 1,802 | +272% | 0 | 0 | — |
case-11 | fail→pass | 23,836 | 8,244 | -65% | 1 | 1 | 0% | 1,121 | 1,653 | +47% | 0 | 0 | — |
case-12 | fail→pass | 16,653 | 6,447 | -61% | 1 | 1 | 0% | 2,232 | 2,429 | +9% | 0 | 0 | — |
case-13 | fail→pass | 10,322 | 15,604 | +51% | 1 | 1 | 0% | 2,159 | 3,604 | +67% | 0 | 0 | — |
case-14 | fail→pass | 11,736 | 4,583 | -61% | 1 | 1 | 0% | 2,002 | 2,200 | +10% | 0 | 0 | — |
case-15 | fail→fail | 5,500 | 11,793 | +114% | 1 | 1 | 0% | 1,061 | 1,569 | +48% | 0 | 0 | — |
case-16 | fail→pass | 27,413 | 8,765 | -68% | 1 | 1 | 0% | 5,041 | 1,986 | -61% | 0 | 0 | — |
case-17 | fail→pass | 7,164 | 9,548 | +33% | 1 | 1 | 0% | 1,313 | 2,192 | +67% | 0 | 0 | — |
case-18 | fail→fail | 12,239 | 8,548 | -30% | 1 | 1 | 0% | 1,254 | 1,799 | +43% | 0 | 0 | — |
case-19 | fail→pass | 6,156 | 16,604 | +170% | 1 | 1 | 0% | 983 | 2,704 | +175% | 0 | 0 | — |
case-20 | pass→fail | 21,293 | 22,997 | +8% | 1 | 1 | 0% | 3,391 | 2,198 | -35% | 0 | 0 | — |
case-21 | pass→pass | 11,514 | 11,155 | -3% | 1 | 1 | 0% | 1,211 | 2,400 | +98% | 0 | 0 | — |
case-22 | pass→pass | 18,099 | 17,648 | -2% | 1 | 1 | 0% | 3,664 | 4,961 | +35% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 14 counted toward the lift figure. The other 8 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +50 percentage points is the difference between those two pass rates over the 14 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/28/2026 | +25% |
Other measured skills in the registry, with their headline benchmark lift.