Use Cases & Success Criteria
Before any code: name the 2-3 use cases, pick a category, and write down what success looks like.
Start With Use Cases, Not Code
Before writing a single line of SKILL.md, identify two or three concrete use cases the skill should enable. A use case has a trigger (what the user says), a set of steps (what happens), and a result (what success looks like). If you cannot write those three elements down, the skill is not ready to be built.
Use Case: Project Sprint Planning
Trigger: User says "help me plan this sprint" or "create sprint tasks"
Steps:
1. Fetch current project status from Linear (via MCP)
2. Analyze team velocity and capacity
3. Suggest task prioritization
4. Create tasks in Linear with proper labels and estimates
Result: Fully planned sprint with tasks createdThree Use-Case Categories
Anthropic's internal taxonomy groups skills into three categories. Knowing which one you are building drives downstream decisions — what the description emphasizes, whether scripts are bundled, and whether the skill leans on an MCP server.
| Category | Used For | Real Example | Key Techniques |
|---|---|---|---|
| 1. Document & Asset Creation | Consistent, high-quality outputs: docs, decks, code, designs | frontend-design, docx/pptx/xlsx/pdf skills | Embedded style guides, templates, quality checklists |
| 2. Workflow Automation | Multi-step processes with consistent methodology | skill-creator | Step-by-step flow, validation gates, iterative refinement |
| 3. MCP Enhancement | Workflow guidance layered on an MCP server | sentry-code-review (Sentry MCP) | Sequenced MCP calls, domain expertise, error handling |
A skill that tries to serve all three categories usually fails all three. Pick one, execute it well, and ship a second skill if you need the other capability.
Define Success Criteria Up Front
Treat these as aspirational targets rather than precise thresholds — skills evaluation tooling is still maturing. The value is forcing yourself to describe, before building, what "this skill works" actually means.
Quantitative
- ▸Skill triggers on ~90% of relevant queries. Measure by running 10–20 test queries and counting auto-loads vs explicit invocation.
- ▸Workflow completes in X tool calls. Compare the same task with and without the skill; count tokens and calls.
- ▸Zero failed API calls per workflow. Monitor MCP logs during test runs; track retry rates and error codes.
Qualitative
- ▸Users don't need to prompt Claude about next steps. Count redirect/clarification turns during beta testing.
- ▸Workflows complete without user correction. Run the same request 3–5 times; compare outputs for structural consistency.
- ▸Consistent results across sessions. Can a new user accomplish the task on first try with minimal guidance?
Problem-First vs Tool-First
Before finalizing a pattern, decide how users will arrive at the skill. Problem-first skills activate because the user describes an outcome ("I need to set up a project workspace") — the skill orchestrates the right MCP calls behind the scenes. Tool-first skills activate because the user already has an integration ("I have Notion MCP connected") and the skill teaches optimal workflows on top. Most skills lean one direction; knowing which shapes the description, examples, and error handling.
Think of it like Home Depot. Problem-first: you walk in saying "I need to fix a kitchen cabinet" and an employee points you to the right tools. Tool-first: you pick out a new drill and ask how to use it. Skills work the same way — frame yours the way your users actually approach the task.