菜单
Use when running routine Gradle tests, builds, lint, formatting, generated-output checks, or verbose output collection in stripe-android. Delegate independent low-judgment work to a low-cost subagent to preserve the main context.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/stripe/stripe-android --skill delegate-low-reasoning-work命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Use when writing, converting, or reviewing parameterized JVM, Robolectric, instrumentation, or Paparazzi tests in stripe-android. Covers TestParameterInjector runners, source-set dependencies, method versus class parameters, value providers, case matrices, and stable test names.
Use when writing or structuring unit tests in stripe-android — covers runScenario pattern, fakes, Turbine Flow testing, and Truth assertions
Reproduce, diagnose, and fix flaky stripe-android tests with focused, repeated validation and evidence-driven synchronization or determinism fixes. Use for intermittent unit, coroutine, Robolectric, Compose, screenshot, instrumentation, Gradle Managed Device, PaymentSheet E2E, Chrome, WebView, UiAutomator, or Espresso failures, while ensuring temporary ShampooRule edits never enter commits or pull requests.
基于 SOC 职业分类
正在显示 SKILL.md
| name | delegate-low-reasoning-work |
| description | Use when running routine Gradle tests, builds, lint, formatting, generated-output checks, or verbose output collection in stripe-android. Delegate independent low-judgment work to a low-cost subagent to preserve the main context. |
Delegate routine, verbose work when the client has a lower-cost subagent available. Keep responsibility for intent, implementation choices, diagnosis, and the final conclusion in the main conversation.
./gradlew test, build, lint, formatting, or generated-output command.Use the least capable available agent that can execute the task reliably. Prefer the client's low-cost profile, such as Haiku or Luna when available. If no appropriate subagent or delegation slot is available, run the command in the main session and say why.
Do not delegate code changes, architectural decisions, failure diagnosis, security-sensitive review, or user-facing conclusions.
State the exact command, working directory, relevant timeout or device constraint, and that the subagent must not edit files. Ask for a compact result containing the exit status, elapsed time, failing test names if any, and no more than 80 relevant log lines.
Run ./gradlew :paymentsheet:testDebugUnitTest from the repository root. Do not edit files.
Return the exit status, elapsed time, failing test names if any, and at most 80 relevant log lines.
Treat the returned output as evidence, not a conclusion. Triage failures and decide the next action in the main session. If the task grows beyond the original contract, delegate a newly bounded command or handle the reasoning directly.