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/marcellocurto/skills --skill user-journey-verifier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。
Review a specific code change for correctness, requirements, maintainability, and repository standards.
Implement and verify clearly scoped software work from an existing spec or tickets.
Reduce an overbuilt code proposal to the simplest production-quality design that meets the real requirements.
正在显示 SKILL.md
| name | user-journey-verifier |
| description | Verify completed software through the exact user workflow and resulting output, without fixing it. |
Verify the outcome through the same path the user relies on.
Verification does not authorize fixes. Do not edit code, install dependencies, change configuration, deploy, or mutate production or external systems unless the user separately requests that action.
Produce direct evidence that the requested user-visible behavior works from its real entry point to its final state or artifact.
Establish before running checks:
Use the user's stated path when one is provided. A substitute path cannot prove the requested path: directly invoking an export function does not verify a browser download button, and inspecting source data does not verify the generated workbook.
If the authoritative environment, reference artifact, or required access is missing, report the gap rather than silently substituting another source.
Exercise the narrowest complete journey. Use available browser, document, PDF, spreadsheet, image, log, or shell capabilities only as needed for the actual outcome.
Check the relevant layers:
Use representative data and scale when the defect or requirement depends on them. Do not claim broad coverage from a happy-path fixture that omits the reported condition.
Local or test verification may perform normal reversible actions required by the journey, such as uploading a fixture or creating disposable test state. Keep artifacts outside the working tree when practical, clean up through the application's normal mechanisms when safe, and do not treat test authority as permission for production or external mutation.
Classify each acceptance check as:
pass: directly observed through the required pathfail: observed behavior contradicts acceptanceunverified: the required path or evidence could not be exercisedAutomated tests, typechecks, builds, logs, and code inspection can support a verdict, but they cannot replace missing user-visible evidence when the outcome is visual, interactive, or artifact-based.
Lead with the journey-level verdict: pass, fail, or unverified.
Report:
| Acceptance check | Evidence | Result |
|---|
For a failure, include the shortest reliable reproduction and the observed-versus-expected difference. Name a likely implementation area only when evidence supports it; do not turn verification into speculative diagnosis.
For an unverified result, state the exact missing access, environment, data, or tool and the next check that would close the gap.
Stop when every material acceptance check has direct evidence or an explicit verification gap. Never report success solely because a different path, unit test, build, or generated artifact succeeded.