菜单
Opinionated language-agnostic test-admission and sans-I/O unit-testing rules. Use when proposing, writing, reviewing, or deleting tests; deciding whether behavior is project-owned; evaluating wrapper, schema, ORM, framework, mock, or third-party-library tests; or separating pure business policy from wiring and I/O.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Opinionated conventions for authoring, restructuring, and reviewing agent skills. Use when deciding whether guidance belongs in `SKILL.md`, a conditional reference, or a smaller skill; writing trigger descriptions and reference routers; naming procedural or documentary skills; creating examples; keeping skill context focused; or bundling static assets and their licenses.
Opinionated conventions for designing focused agent-skill plugins as encapsulated, installable capabilities. Use when defining a plugin boundary, deciding which skills are public entry points, hiding shared implementation skills, splitting broad plugins, designing plugin structure, minimizing client manifests, or removing auxiliary package documentation.
Opinionated language-agnostic design rules for implementation, refactoring, and code review. Use when shaping conditional control flow, finite states, invariants, error boundaries, resource lifetimes, abstractions, public APIs, validation boundaries, pagination, missing values, or the separation of sans-I/O decisions from imperative drivers.
Opinionated feature-first architecture conventions for cohesive modules, owned subtrees, entry points, shared-code promotion, and package boundaries. Use when designing or reviewing directory structure, placing UI or workflow code, splitting nested capabilities, extracting shared modules, enforcing feature import boundaries, deciding whether a feature deserves a package, or refactoring horizontal technical layers.
Opinionated `better-all` conventions for dependency-declared async task graphs, inferred task results, cancellation propagation, and failure semantics. Use when replacing serial `await` chains or manual `Promise.all` stages, parallelizing work with result dependencies, or writing and reviewing `all` and `allSettled` task definitions.
Opinionated logging and error-handling conventions for OpenTelemetry-compatible systems covering log levels, attribute naming, and error classification. Use when writing log statements, choosing log levels, or handling errors with traces.
| name | testing-guidelines |
| description | Opinionated language-agnostic test-admission and sans-I/O unit-testing rules. Use when proposing, writing, reviewing, or deleting tests; deciding whether behavior is project-owned; evaluating wrapper, schema, ORM, framework, mock, or third-party-library tests; or separating pure business policy from wiring and I/O. |
| license | MPL-2.0 |
| metadata | {"author":"Basti Ortiz <ortiz@bastidood.dev>","source":"https://github.com/BastiDood/skills"} |
This skill treats tests as protection for project-owned behavior, not as a coverage ritual. It favors cheap sans-I/O tests and rejects tests that merely restate third-party libraries or transparent wiring.
Test behavior that the project owns. Assume third-party libraries are correct and already tested.
Add a test only when it can fail because project-owned behavior is wrong.
Project-owned behavior includes:
Do not add a test when the implementation only forwards values to a dependency and returns its result unchanged.
Delete a test when it only:
Do not preserve a useless test for coverage. Coverage is evidence of execution, not evidence of meaningful verification.
Read as many linked references as are relevant to the current task before writing, approving, or retaining tests.
