Skip to main content

testing

Use when adding, changing, deleting, or reviewing tests in Anarlog (TypeScript, Rust, or node --test scripts), deciding whether a change needs a test, fixing a failing test, or pruning test suites. Covers what is worth testing, what to leave untested, test budgets, and the test-weight report.

소스 정보

저장소
fastrepl/anarlog
최근 소스 활동
2026년 9월 29일 10:01
감지된 SKILL.md 언어
영어
스타
9,517
포크
796

설치 방법

기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.

소스 파일 검토

설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.

파일 탐색기
2 개 파일

SKILL.md 표시 중

SKILL.md
소스 지침 · 읽기 전용 미리보기
name
testing
description
Use when adding, changing, deleting, or reviewing tests in Anarlog (TypeScript, Rust, or node --test scripts), deciding whether a change needs a test, fixing a failing test, or pruning test suites. Covers what is worth testing, what to leave untested, test budgets, and the test-weight report.
# Testing Use this skill when adding, changing, deleting, or reviewing tests in Anarlog, deciding whether a change needs a test, or pruning a test suite. Tests are code we pay for on every CI run and every refactor; keep the ones that catch real regressions and nothing else. ## Write a test when - A bug was fixed: add one regression test that fails without the fix. One, not a matrix of neighbours. - The behavior is observable by a user or another component: public APIs, tauri-specta bindings, OpenAPI contracts, MCP and CLI output, serializers and parsers, TipTap/ProseMirror JSON conversion. - Losing it loses data or trust: SQLite migrations and desktop/mobile schema parity, CloudSync and E2EE, auth, billing and Stripe webhooks, permissions and RLS, recording and STT lifecycles. - It is a real concurrency or recovery invariant: capture start/stop races, crash recovery, retry, cancellation, ordering between actors or tasks. - It encodes release or store policy that a workflow edit could silently break (for example `scripts/desktop-release-provenance.test.mjs`). ## Do not write a test for - CSS classes, DOM structure, exact copy, or log text, unless the text is a contract (CLI output parsed by another tool, i18n keys checked by CI). - Mock call counts or call order when the observable result already proves the behavior. - Private state, internal helpers already covered through their public caller, or trivial delegation. - Framework or library behavior (React, TanStack, Tauri, Zustand, Tokio, serde). - Constants, enum tables, type re-exports, or anything that mirrors source line by line. - Package or file layout. Structural and import-boundary rules belong in oxlint/ESLint rules, not tests that read files from disk. - `test.skip` / `#[ignore]` without a linked issue. ## Budget - Prefer strengthening an existing test over adding a new one. Prefer one table-driven case over five copies only when the rows exercise different branches. - A new test must name the regression it prevents in its title. - Deleting a test is fine when the behavior is covered elsewhere, the code is gone, or it matches the "do not write" list. Say which in the commit or PR. - Keep fixtures and helpers next to the tests that use them; delete them with the last caller. - If a module's test file outgrows its source several times over, stop adding and consolidate first. ## Test-weight report Run `node .agents/skills/testing/scripts/test-weight.mjs` from the repo root to list packages and files whose test code most outweighs their source. Use `--help` for thresholds and JSON output. Audit the top entries against the rules above before adding more tests to them. ## Conventions Which tests run for which component is listed under "Pre-commit verification" in the root `AGENTS.md`; the workflows in `.github/workflows/` are the source of truth. Axum transport choice for API tests follows `../axum-test-transport/SKILL.md`.
GitHub에서 보기