# Test Generation

> Cover the behavior that can actually break: edges, errors, and the regression that bit you. Use when the user asks to write tests.

- Skill: `delzarsolutionsllc/test-generation` (Agent Skill)
- Install (CLI): `npx skillmds@latest add delzarsolutionsllc/test-generation`
- Raw SKILL.md: https://api.skillmd.com/api/skills/delzarsolutionsllc/test-generation/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: delzarsolutionsllc (https://skillmd.com/u/delzarsolutionsllc)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/delzarsolutionsllc/test-generation

---


# Test Generation

Generate tests that would have caught a real bug. Coverage percentage is not the goal.

## Method
1. Read the unit under test and list the behaviors: happy, empty, invalid, boundary, concurrent if it shares state.
2. Prefer existing test helpers and fixtures. Match the repo's style exactly.
3. Each test: arrange, act, assert one behavior. No giant "kitchen sink" cases.
4. Include the error path with the exact error type/message the code promises.
5. For I/O, freeze time, clock, and network. Do not hit live services.

## Priorities
1. Regression for a stated bug
2. Boundary and invalid input
3. Authz deny cases
4. Idempotency / retry
5. Snapshot UI only when layout is the contract

Delete any generated test that only restates the implementation line-for-line. If you cannot name the bug it would catch, drop it.