staged-review-validator
Staged Review Validator
用于对已有审查报告做二次验证,不负责重新发起完整扫描。
何时使用
- 用户已经有一份
staged-changes-review结果,想确认每条问题是否真实成立 - 用户问“这个问题到底要不要改”
- 用户希望获得“逐条裁决 + 证据 + 修改建议”
输入契约
必需输入
- 用户提供一段
staged-changes-review文本(至少包含问题描述)
推荐输入
- 文件路径
- 行号
- 风险级别
- 置信度
- 原始建议
缺失输入处理
如果用户没有粘贴审查结果:
- 明确提示“请先粘贴 staged-changes-review 结果”
- 提供一个最小输入模板(见
references/output-template.md) - 不自动改为扫描
git diff --cached
核验流程
1. 解析问题清单
从用户提供的报告提取每条问题:
- 标题
- 文件与行号
- 原风险级别
- 原置信度
- 原建议
2. 逐条证据核验
对每条问题进行核验:
- 文件/行号是否存在
- 代码上下文是否匹配问题描述
- 该问题是否已经在当前变更中修复
- 是否存在上下文导致“看似问题但实际合理”
可使用但不限于以下命令:
git diff --cached
rg -n "关键符号" <path>
sed -n '<start>,<end>p' <path>
git show HEAD:<path>
3. 给出裁决
每条问题只能输出以下三种之一:
成立:证据直接支持问题描述不成立:证据与描述冲突,或问题已修复待确认:证据不足(路径/行号缺失、上下文不完整等)
4. 推导“是否需要修改”
成立且风险为中/高:需要修改成立且风险低:建议修改不成立:不需要修改待确认:暂不修改,先补充信息
输出要求(固定)
先给总览,再给逐条裁决。
总览
- 问题总数
- 成立数
- 不成立数
- 待确认数
- 建议修改数
逐条裁决块(每条必含)
- 问题标题
- 文件位置
- 风险级别
- 置信度
- 裁决(成立/不成立/待确认)
- 是否需要修改(是/否/待补充)
- 证据(命令结果或代码定位)
- 最小改动建议
结尾给出编号式“可执行下一步”。
与 staged-changes-review 的关系
staged-changes-review:发现候选问题staged-review-validator:验证候选问题是否真实成立
串联时,后者不重复完整扫描逻辑,重点做证据核验与裁决。
参考文档
references/verification-rubric.md:裁决标准、证据优先级、置信度映射references/output-template.md:固定输出模板与示例
More from adonis0123/adonis-skills
staged-changes-review
Checklist-driven review of staged Git changes with deterministic rule scanning and semantic analysis. Use when the user wants to review staged changes, check for errors before commit, or validate code quality before committing.
25ruler-rules-init
Migrate and bootstrap Ruler-based AI rules across repositories with English-first templates, preset-based project detection, and safe, idempotent setup. Use when creating or standardizing `.ruler/*` files, generating `AGENTS.md`/`CLAUDE.md`, wiring `ruler:apply` into `package.json`, adding generated-file ignores in `.gitignore`, and designing `applyTo` scoping patterns such as `**` and `web/**`.
18discuss-before-plan
Enforces 'decide then plan' discipline - the pre-planning decision gate. Use when the user asks for a plan or starts a change while key decisions are unresolved: architecture tradeoffs, data flow, public interfaces, unclear requirements, multi-module scope, or roughly 5+ files affected. Also triggers when the user explicitly wants to discuss, compare options, or review architecture before committing. Core job: reduce incorrect-execution cost by confirming decisions before producing executable plans.
17claude-skills-sync-init
Bootstrap Claude skill-sync automation in any repository. Use when users ask to set up `.agents/skills` to `.claude/skills` syncing, generate `skills:sync:*` scripts, inject package.json scripts/postinstall hooks, or initialize this sync workflow in a new or existing project.
16tailwindcss-next-init
Initialize Tailwind CSS v4 and Iconify for Next.js App Router projects with script-driven templates. Use when you need a fast setup or migration with consistent style-entry conventions.
12weekly-report
Generate structured weekly reports from Git commit history across one or multiple repositories. Use when you need concise, project-grouped progress summaries for status reporting.
11