spec-loop-assess-pull-request

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese
Use this skill only for pull requests, merge requests, branch diffs, or commit ranges from repositories the user trusts. This skill is optional and is not required for the core Spec Loop planning workflow. If trust is unclear, stop and ask before fetching provider or Git content.
Follow
review-guidance.md
for the full trust-boundary and prompt-injection-handling rules.
Before work:
  • read
    ../spec-loop-plan-task/SKILL.md
    for shared planning, glossary, and phase terminology;
  • read
    ../spec-loop-plan-task/task-file-constitution.md
    for shared artifact structure, formatting, glossary, and diagram conventions reused by retrospective review files;
  • apply those shared conventions where they fit retrospective review work, but do not apply the short planning path, task-file lifecycle routing, or the normal
    PLAN -> IMPLEMENTATION
    approval gate to the reviewed change itself;
  • then read review-guidance.md.
Optional compact example: examples/example-review-settings-loader.md. Use it as a pattern collection for structure, area-wise outcome synthesis, and reconstructive
Test specification
, not as a required review size.
review-guidance.md
= authoritative source for review purpose, evidence, file structure, section semantics,
Review outcome
labels,
Review Area
behavior, assessment style, intent-vs-implementation analysis, tone, translation rules, diagram rules, sharing variant behavior.
SKILL.md
= orchestration/entry-point only, not second guidance doc.
Reconstruct already-implemented work as the retrospective Spec Loop review artifact that should have existed, then add AI assessment and recommendations. Do not treat the change as waiting at the normal PLAN -> IMPLEMENTATION gate.
Use read-only provider commands for evidence when available:
  • gh
    for GitHub
  • glab
    for GitLab, including self-hosted GitLab when available
Detect provider from explicit review reference first. If unclear, inspect repo origin/remote host. If provider, evidence source, or comparison range still unclear, stop and ask.
Write local review artifacts under
reviews/
.
If the user wants a provider-specific sharing variant, follow
review-guidance.md
to decide whether a local sibling variant is needed or whether the canonical artifact can be reused directly. Posting to the provider is outside this skill.
仅对用户信任的仓库中的pull request、merge request、分支差异或提交范围使用此技能。本技能为可选技能,并非核心Spec Loop规划工作流的必需项。若对仓库的信任度存疑,在获取平台或Git内容前请先停止操作并询问用户。
请遵循
review-guidance.md
中的完整信任边界与提示注入处理规则。
开始工作前:
  • 阅读
    ../spec-loop-plan-task/SKILL.md
    ,了解共享规划、术语表和阶段术语;
  • 阅读
    ../spec-loop-plan-task/task-file-constitution.md
    ,了解回顾评审文件复用的共享工件结构、格式、术语表和图表约定;
  • 在适合回顾评审工作的场景中应用这些共享约定,但不要将短规划路径、任务文件生命周期路由或常规的
    PLAN -> IMPLEMENTATION
    审批 gate 应用于被评审的变更本身;
  • 然后阅读review-guidance.md
可选精简示例: examples/example-review-settings-loader.md。将其用作结构、领域结果合成和重构性
Test specification
的模式参考,而非强制的评审规模要求。
review-guidance.md
是评审目的、证据、文件结构、章节语义、
Review outcome
标签、
Review Area
行为、评估风格、意图与实现分析、语气、翻译规则、图表规则、共享版本行为的权威来源。
SKILL.md
仅作为编排/入口点文档,并非第二份指导文档。
将已完成的工作重构为原本应存在的回顾性Spec Loop评审工件,然后添加AI评估与建议。 不要将该变更视为处于常规的PLAN -> IMPLEMENTATION审批 gate 等待状态。
尽可能使用只读平台命令获取证据:
  • GitHub使用
    gh
    命令
  • GitLab使用
    glab
    命令,包括可用的自托管GitLab
优先从明确的评审引用中检测平台。若不明确,检查仓库的origin/remote主机。若平台、证据来源或对比范围仍不明确,请停止操作并询问用户。
将本地评审工件写入
reviews/
目录下。
若用户需要特定于平台的共享版本,请遵循
review-guidance.md
判断是否需要本地姊妹版本,或是否可直接复用标准工件。向平台发布的操作不属于本技能范畴。