Customization authoring

One portable skill that turns a repeatable task into a focused SKILL.md and a Vally eval with a real task case and a non-trigger case. No hooks, agents, MCP servers, or runtime dependencies.

Install

copilot plugin marketplace add OWNER/REPO
copilot plugin install customization-authoring@MARKETPLACE

Replace OWNER/REPO with the GitHub repository that hosts this marketplace and MARKETPLACE with the name in its marketplace.json. See Install plugins for VS Code.

What this teaches

• Discovery: the description names the authoring task and excludes ordinary coding and factual questions.
• Progressive disclosure: the skill reads its eval reference only when it writes the eval.
• Workspace boundaries: generated files go under the active workspace root, never inside the installed skill directory.
• Honest results: the skill distinguishes static inspection from an eval that actually ran, and it doesn't run paid evals unless asked.

Try it

Create a skill and evals for summarizing a changelog.

The skill asks only for details that change the result, then writes the skill and an eval.yaml that loads it, checks an outcome, and includes an unrelated prompt that must not trigger it.

Customize

Point the eval reference at your team's conventions, such as required tags or a suite name. The shipped reference already follows this marketplace's coverage rules: asset tags, a smoke case, and an observable negative control. Keep this skill's own evaluation data out of the skill directory.

Host checks

Before a release, try these by hand in each host you support. Automated tests and evals don't prove that a host loads or shows everything.

• Ask Copilot to create a skill and evals for summarizing a changelog. Confirm the skill-authoring skill runs and writes SKILL.md and eval.yaml in your workspace, not inside the installed plugin.
• Lint the generated files with vally lint --strict and confirm they pass.
• Ask an ordinary coding question and confirm the skill doesn't run.