reviewing by desplega-ai/ai-toolbox
npx skills add https://github.com/desplega-ai/ai-toolbox --skill reviewing您正在对文档(研究、计划或头脑风暴)进行结构化评审,以识别其中的缺口、弱点和质量问题。
这些指令确立了您与用户之间的工作协议。关键原则如下:
AskUserQuestion 是您的主要沟通工具 - 每当您需要询问用户任何事项(澄清、偏好、决策)时,请使用 AskUserQuestion 工具。不要以纯文本形式输出问题 - 始终使用结构化工具,以便用户能够高效响应。
预先确定偏好 - 在工作流程开始时询问用户偏好,而不是在用户可能想要继续推进时才在结束时询问。
自主模式指导交互级别 - 用户选择的自主级别决定了您检查的频率,但 AskUserQuestion 仍然是所有问题的询问机制。
在开始审阅之前(除非自主模式为 Autopilot),请确定以下偏好:
输出模式 - 使用 AskUserQuestion 并附带:
| 问题 | 选项 |
|---|---|
| "我应该如何呈现审阅结果?" | 1. 将勘误部分附加到文档末尾(推荐),2. 将审阅写入单独的审阅文件到 thoughts/*/reviews/ |
文件审阅偏好 - 检查 file-review 插件是否可用(在可用命令中查找 file-review:file-review)。
广告位招租
在这里展示您的产品或服务
触达数万 AI 开发者,精准高效
如果 file-review 插件已安装,请使用 AskUserQuestion 并附带:
| 问题 | 选项 |
|---|---|
| "在自动审阅之后,您是否希望使用 file-review 进行内联反馈?" | 1. 是的,审阅后打开 file-review(推荐),2. 不,自动审阅已足够 |
存储这些偏好并在审阅过程中据此执行。
此技能在以下情况下激活:
/review 命令在审阅开始时,根据自主模式调整您的交互级别:
| 模式 | 行为 |
|---|---|
| Autopilot | 运行完整审阅,自动修复次要问题,最后呈现摘要 |
| Critical (默认) | 询问关键/重要发现,自动修复次要发现 |
| Verbose | 逐步检查每个发现,在进行任何更改前确认 |
自主模式由调用命令传递。如果未指定,则默认为 Critical。
完整阅读输入文档。根据其路径和内容结构确定文档类型:
| 路径包含 | 类型 |
|---|---|
/research/ | 研究文档 |
/plans/ | 计划 |
/brainstorms/ | 头脑风暴 |
如果类型不明确,请根据内容结构推断或使用 AskUserQuestion 进行澄清。
根据文档类型验证所需部分是否存在:
研究文档:
计划:
头脑风暴:
应用特定类型的质量标准:
研究文档:
计划:
头脑风暴:
查找缺失或假设的内容:
将所有发现按严重性级别分为三类:
| 严重性 | 含义 | 操作 |
|---|---|---|
| Critical | 阻碍正确性或完整性 — 必须解决 | 与用户讨论 |
| Important | 重大缺口或弱点 — 应该解决 | 与用户讨论(或在 Autopilot 模式下自动修复) |
| Minor | 格式、拼写错误、小的不一致 | 除非是 Verbose 模式,否则自动修复 |
以文本输出形式呈现摘要,并按严重性对发现进行分组。
根据输出模式偏好:
如果选择"附加勘误":
直接在文档中自动修复次要问题
对于关键/重要项目,在文档末尾附加一个 ## 审阅勘误 部分,内容如下:
## 审阅勘误
_审阅日期:YYYY-MM-DD 由 [审阅者]_
### 关键问题
- [ ] [发现描述和建议操作]
### 重要问题
- [ ] [发现描述和建议操作]
### 已解决
- [x] [次要问题] — 已自动修复
如果选择"单独文件":
thoughts/*/reviews/YYYY-MM-DD-review-of-<original-slug>.md关键:审阅者识别问题 — 审阅者不重写文档。呈现发现,让原作者去处理它们。唯一的例外是次要问题的自动修复(拼写错误、格式)。
如果 file-review 插件可用,并且用户在用户偏好设置期间选择了"是":
/file-review:file-review <路径> 进行内联人工评论file-review:process-review 技能处理反馈每周安装次数
1
仓库
GitHub 星标数
13
首次出现
1 天前
安全审计
安装于
zencoder1
amp1
cline1
openclaw1
opencode1
cursor1
You are performing a structured critique of a document (research, plan, or brainstorm) to identify gaps, weaknesses, and quality issues.
These instructions establish a working agreement between you and the user. The key principles are:
AskUserQuestion is your primary communication tool - Whenever you need to ask the user anything (clarifications, preferences, decisions), use the AskUserQuestion tool. Don't output questions as plain text - always use the structured tool so the user can respond efficiently.
Establish preferences upfront - Ask about user preferences at the start of the workflow, not at the end when they may want to move on.
Autonomy mode guides interaction level - The user's chosen autonomy level determines how often you check in, but AskUserQuestion remains the mechanism for all questions.
Before starting review (unless autonomy is Autopilot), establish these preferences:
Output Mode - Use AskUserQuestion with:
| Question | Options |
|---|---|
| "How should I present the review findings?" | 1. Append errata section to the document (Recommended), 2. Write a separate review file to thoughts/*/reviews/ |
File Review Preference - Check if the file-review plugin is available (look for file-review:file-review in available commands).
If file-review plugin is installed, use AskUserQuestion with:
| Question | Options |
|---|---|
| "Would you like to use file-review for inline feedback after the automated review?" | 1. Yes, open file-review after review (Recommended), 2. No, the automated review is sufficient |
Store these preferences and act on them during the review process.
This skill activates when:
/review commandAt the start of review, adapt your interaction level based on the autonomy mode:
| Mode | Behavior |
|---|---|
| Autopilot | Run full review, auto-fix minor issues, present summary at end |
| Critical (Default) | Ask about Critical/Important findings, auto-fix Minor ones |
| Verbose | Walk through each finding, confirm before any changes |
The autonomy mode is passed by the invoking command. If not specified, default to Critical.
Read the input document fully. Determine the document type from its path and content structure:
| Path contains | Type |
|---|---|
/research/ | Research document |
/plans/ | Plan |
/brainstorms/ | Brainstorm |
If the type is ambiguous, infer from content structure or use AskUserQuestion to clarify.
Verify required sections exist based on document type:
Research documents:
Plans:
Brainstorms:
Apply type-specific quality criteria:
Research documents:
Plans:
Brainstorms:
Look for what's missing or assumed:
Categorize all findings into three severity levels:
| Severity | Meaning | Action |
|---|---|---|
| Critical | Blocks correctness or completeness — must be addressed | Discuss with user |
| Important | Significant gap or weakness — should be addressed | Discuss with user (or auto-fix in Autopilot) |
| Minor | Formatting, typos, small inconsistencies | Auto-fix unless Verbose mode |
Present a summary as text output with findings grouped by severity.
Based on output mode preference:
If "Append errata":
Auto-fix Minor issues directly in the document
For Critical/Important items, append a ## Review Errata section at the end of the document with:
## Review Errata
_Reviewed: YYYY-MM-DD by [reviewer]_
### Critical
- [ ] [Finding description and recommended action]
### Important
- [ ] [Finding description and recommended action]
### Resolved
- [x] [Minor issue] — auto-fixed
If "Separate file":
thoughts/*/reviews/YYYY-MM-DD-review-of-<original-slug>.mdCRITICAL : The reviewer identifies issues — the reviewer does NOT rewrite the document. Present findings and let the original author address them. The only exception is Minor auto-fixes (typos, formatting).
If the file-review plugin is available and the user selected "Yes" during User Preferences setup:
/file-review:file-review <path> for inline human commentsfile-review:process-review skillWeekly Installs
1
Repository
GitHub Stars
13
First Seen
1 day ago
Security Audits
Gen Agent Trust HubPassSocketPassSnykPass
Installed on
zencoder1
amp1
cline1
openclaw1
opencode1
cursor1
Skills CLI 使用指南:AI Agent 技能包管理器安装与管理教程
46,600 周安装
DuckDuckGo搜索技能 - 开源网络搜索工具,集成DuckDuckGo API,支持多种开发环境
106 周安装
Nuxt 4 数据管理:useFetch、useAsyncData、useState 组合式函数与 SSR 状态管理
70 周安装
Tauri v2 前端调用 Rust 后端函数教程:命令系统与参数传递详解
108 周安装
定价策略师指南:SaaS定价模型、价值指标与分层定价策略框架
106 周安装
Godot组合式架构技能包:组件化游戏开发最佳实践与脚本示例
108 周安装
Firebase 开发指南:Firestore 数据建模、安全规则与实时应用开发
108 周安装