# Test

> Runs a persistent verification-and-fix loop, executing existing test suites and browser/desktop checks until the requested surface is verified or a real blocker remains.

- Skill: `comeonoliver/test` (Agent Skill)
- Install (CLI): `npx skillmds@latest add comeonoliver/test`
- Raw SKILL.md: https://api.skillmd.com/api/skills/comeonoliver/test/raw
- Safety review: CAUTION (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools, Testing & QA
- Tags: Chrome Devtools, Cypress, Jest, Playwright, Tauri, Testing, Verification, Vitest
- Author: ComeOnOliver (https://skillmd.com/u/comeonoliver)
- Updated: 2026-08-22
- Page: https://skillmd.com/skills/comeonoliver/test

---


# Test

Use this as a verification-and-fix mode after implementation or during bug hunts. This is not just "run one test command". It is a persistent loop that keeps narrowing failures and rerunning checks until the requested behavior is verified.

## Default Loop

1. Define the verification target.
- affected files, routes, surfaces, and user-visible claims
- exact stopping condition for signoff
2. Build the smallest credible test matrix from the repo and task.
- code-level checks first
- integration and e2e next
- live runtime QA last
3. Run the relevant existing automated checks before inventing custom ones.
- prefer repo-native scripts and targeted specs or files
- use the stack already present: Vitest, Jest, Playwright, Cypress, pytest, go test, etc.
- widen scope only after the narrow failing checks pass
4. If the surface is interactive, verify the real runtime too.
- web: run relevant e2e flows or drive the app directly
- before browser DevTools inspection, probe `http://127.0.0.1:9333/json/version`
- if it is already live, reuse it
- if it is not live and a `chrome` launcher exists, start it automatically, wait for the port, then use `chrome-devtools`
- browser diagnosis: use `chrome-devtools` for console, network, DOM, accessibility, or performance inspection
- before Pix desktop DevTools inspection, probe `http://127.0.0.1:9334/json/version`
- if it is already live, reuse it
- if it is not live and a `tauri-pix` launcher exists, start it automatically, wait for the port, then use `tauri-devtools`
- Tauri desktop: use `tauri-devtools` and the fixed Tauri debug path when diagnosis is needed
- direct browser control: use the repo's Playwright setup or a persistent `js_repl` Playwright session when targeted manual proof is still needed
5. On failure, fix and rerun.
- repeat inspect -> patch -> rerun narrow checks -> rerun broader confidence checks
- do not stop after the first or second failed attempt
6. Before signoff, rerun the checks that support the final claims.

## Priority Rules

- Existing repo tests beat ad hoc manual clicks.
- Narrow targeted checks beat whole-suite runs unless the task truly warrants broader coverage.
- Browser or desktop verification complements automated tests; it does not replace them.
- When the repo has both unit-level tests and Playwright e2e, usually run the relevant unit or component checks first, then the relevant Playwright flow, then live diagnosis only if something is still unclear.
- When the required DevTools launcher exists, do not wait for the user to open it manually; check first, auto-launch if absent, and continue.
- If the user implies "I am stepping away" or "keep going until it is clean", keep iterating without asking after each failure.

## Good Targets

- "/test"
- "나 점심 먹고 올 테니 테스트랑 검증 끝까지 돌려"
- "버그 없어질 때까지 계속 확인해"
- "회귀까지 보고 진짜 끝내줘"
- "웹, 모바일, Tauri까지 확인하고 막아줘"

## Avoid

- Do not stop at "probably fixed".
- Do not run the whole monorepo matrix by reflex.
- Do not skip existing automated tests just because manual QA looked okay.
- Do not claim success if any required surface was not actually checked.
- Do not keep brute-forcing when a real external blocker is the limit; name it clearly.

## Finish

Return:
- verification target
- automated checks run
- runtime surfaces checked
- pass or fail
- remaining blockers or residual risk