| name | ai-tooling-evaluation |
| description | Evaluate or design AI tooling for VS Code Copilot: custom agents, skills, instructions, hooks, MCP servers, plugins, prompts, model routing, and workflow automation. Use when setting up or improving an AI-assisted development environment. |
| argument-hint | Describe the workflow or AI tooling problem |
AI Tooling Evaluation
Use this skill when deciding how to improve the AI development environment for this workspace.
Procedure
- Define the workflow pain point in one sentence.
- Check existing
.github\agents, .github\skills, .github\instructions, .github\hooks, and .github\prompts before adding anything.
- Pick the smallest primitive that solves the problem:
- Instruction: guidance that should be loaded automatically.
- Skill: repeatable multi-step workflow with optional references or scripts.
- Custom agent: focused role, context isolation, or restricted tools.
- Hook: deterministic lifecycle enforcement or context injection.
- MCP server/plugin: only when the workflow needs live external data, APIs, tools, or state that files cannot provide.
- Keep descriptions keyword-rich and include trigger phrases such as "Use when: ...".
- Validate new files:
- JSON hooks parse cleanly.
- Markdown customization files have valid frontmatter when required.
- Skill folder name matches the
name field.
MCP and plugin threshold
Do not add an MCP server or plugin just because it is possible. Recommend one only when at least one of these is true:
- The agent needs authenticated access to a system outside the repo.
- The workflow needs live state that changes independently of files.
- Existing CLI/file workflows are too slow, brittle, or unsafe.
- The integration can be configured without committing secrets.
Output
Return a short recommendation with:
- chosen primitive
- files to create or update
- why MCP/plugin is or is not justified
- validation command or check