## Use the packaged VC knowledge The managed Copilot workspace includes VC-focused GitHub Copilot instructions and skills. They help Copilot choose an appropriate procedure for specialized work without making each chat repeat the same background. The packaged set includes focused skills such as `vc-component-modeling`, `vc-py3-authoring`, and `vc-pm-configuration-authoring`. The workspace instructions direct Copilot to load the most specific skill before specialized VC work. Start from [Work with chats and the Copilot workspace](https://inspector.4111778.xyz/docs/copilot/chat-sessions-and-workspace/) to open the managed workspace in VS Code. ## Distill a repeatable lesson Use a local skill when a VC-specific fact changes how an agent should reason, not merely how it should phrase an answer. For example, a skill can require context that changes which model structure to inspect, which values to preserve, or how to verify the resulting change. After completing an exercise, ask Copilot to capture the information worth keeping with this prompt: ```text Write a new lessons learned markdown document to the workspace about this exercise of What did you struggle with, what information the user had to provide to move forward? Were any issues or limitations encountered with the MCP tools? What information or tools would have made the task easier and faster? ``` 1. Work through a representative task with Copilot and correct missing VC context as it appears. 2. Ask Copilot for a short lessons-learned report that separates durable rules, evidence to inspect, common failure modes, and verification steps. 3. Keep only the rules that apply repeatedly to one task family. Put the result in `.github/skills//SKILL.md` in the managed workspace. 4. Test the skill by asking Copilot to complete a comparable task. Check that it loads the skill, inspects the required structure, and reports the specified verification rather than guessing. Some VS Code installations may offer a `create_skill` skill to help create the file. When it is unavailable, create the folder and `SKILL.md` directly. ## Make the skill discoverable Give the skill a task-specific name and a description that says when Copilot should use it. The packaged `vc-component-modeling` skill uses this verified frontmatter style: For guidance on writing focused, effective skill instructions, see [Creating effective skills](https://awesome-copilot.github.com/learning-hub/creating-effective-skills/). ```yaml --- name: vc-component-modeling description: "Use for Visual Components component modeling: CAD cleanup, links, joints, frames, interfaces, parametrics, live MCP inspection, and component-scoped Python 3 work that should also load vc-py3-authoring." argument-hint: "Inspect or author a VC component-modeling task" user-invocable: false disable-model-invocation: false --- ``` For a local skill, replace the name, description, and hint with the narrow problem it owns. Include the conditions that activate it, the VC objects or properties to inspect, rules that must not be inferred, and a completion check. Avoid a broad collection of unrelated tips: it is harder for Copilot to select and harder for you to validate. ## Choose how updates apply Creating a new local skill leaves the bundled skill itself unchanged. This is the preferred choice for organization-specific lessons. Editing a bundled `SKILL.md` is deliberate customization. The package installer preserves a pre-existing ordinary packaged file instead of overwriting it, so both a locally changed bundled skill and an unchanged pre-existing bundled skill are retained when a later package update is adopted. Do not expect an automatic replacement or merge of bundled `SKILL.md` files; compare future packaged guidance as appropriate. Share a proposed skill or improvement with your organization or Inspector workspace maintainers when it would help other users. They can evaluate it for a future managed package while you keep a tested local version for current work. ## Result The local skill is useful when a representative request causes Copilot to apply the repeatable VC rule, inspect the stated evidence, and report the expected verification. Keep refining it from real lessons instead of trying to encode every possible VC task.