/review
对任意研究制品(idea、proposal、experiment plan、paper draft、method)进行跨模型审查。
使用 Review LLM 作为独立审稿人,输出结构化评分、可操作的改进建议,以及与 wiki 实体的映射
(哪些 ideas/methods 需要加强,哪些 gaps 被发现)。
支持三种难度级别(standard / hard / adversarial)和四种审查焦点。
可独立使用,也被 /ideate、/refine、/exp-design 调用。
Inputs
artifact:要审查的制品,以下之一:
- wiki 页面的 slug(如
sparse-lora-for-edge-devices,从 ideas/experiments/methods/ 中查找)
- 文件路径(如
wiki/outputs/paper-draft-v1.md)
- 自由文本(直接粘贴的 proposal 或 idea 描述)
--difficulty(可选,默认 standard):
standard:单轮审查,给出结构化反馈
hard:多轮对话(最多 3 轮),Claude 对每个 weakness 进行 rebuttal
adversarial:多轮对话(最多 3 轮),Review LLM 额外尝试找致命缺陷,模拟最严苛的审稿人
--focus(可选,默认全面审查):
method:聚焦方法设计的正确性、创新性、可行性
evidence:聚焦证据是否充分、实验是否严谨、idea/method 是否得到充分支撑
writing:聚焦表达清晰度、结构组织、论证逻辑
completeness:聚焦是否遗漏关键内容(相关工作、ablation、baseline)
Outputs
- Review Report(输出到终端):
- Overall Score(1-10)
- Strengths(优点列表)
- Weaknesses(缺点列表,按严重程度排序)
- Questions(审稿人的疑问)
- Actionable Suggestions(可操作的改进建议,按优先级排序)
- Wiki Entity Mapping(哪些 ideas/methods 需要加强,哪些 gaps 被发现)
- Verdict:
ready / needs-work / major-revision / rethink
- 若
--difficulty >= hard:额外包含多轮对话记录和最终修正后的评分
- 该 skill 不直接修改 wiki,但会输出建议的 wiki 更新列表
Wiki Interaction
Reads
wiki/papers/*.md — 查找制品引用的论文,验证引用正确性
wiki/concepts/*.md — 理解制品涉及的技术概念
wiki/methods/*.md — 检查制品依赖的 methods 当前状态
wiki/experiments/*.md — 查找相关实验结果
wiki/ideas/*.md — 如果审查的是 idea,检查其上下文
wiki/graph/context_brief.md — 获取全局上下文
wiki/graph/open_questions.md — 对照 gap map 检查完整性
.claude/skills/shared-references/cross-model-review.md — 审稿独立性原则
Writes
- 无。Review 是只读查询操作。
- 审查结果输出到终端,由用户或调用方(如 /refine)决定是否应用。
Graph edges created
Workflow
前置:确认工作目录为 wiki 项目根(包含 wiki/、raw/、tools/ 的目录)。
Step 1: 加载上下文
解析 artifact:
- 若为 slug:按顺序在
wiki/ideas/、wiki/experiments/、wiki/methods/、wiki/papers/、wiki/outputs/ 中查找 {slug}.md
- 若为文件路径:直接读取
- 若为自由文本:直接使用
确定 artifact 类型:idea / experiment / method / paper-draft / proposal / other
加载相关 wiki 上下文:
- 读取
wiki/graph/context_brief.md 获取全局视角
- 读取
wiki/graph/open_questions.md 获取知识缺口列表
- 根据 artifact 类型,加载相关 wiki 页面:
- idea → 其 origin_gaps(concepts/topics)、相关 papers
- experiment → 其 linked_idea、相关 experiments
- method → 其 source_papers 和 parent_methods
- paper-draft → 其引用的所有 wiki 页面
读取 cross-model-review.md:确认 Review LLM 独立性原则
构建 reviewer system prompt(根据 --focus):
Base prompt(所有 focus):
You are a senior ML researcher reviewing a research artifact.
Be thorough, specific, and constructive. For every weakness, suggest a concrete fix.
Score on a 1-10 scale where:
- 1-3: Fundamental flaws, not salvageable in current form
- 4-5: Significant issues but core idea may have merit
- 6-7: Solid work with clear areas for improvement
- 8-9: Strong work, minor issues only
- 10: Exceptional, publication-ready
Focus-specific additions:
method:额外要求评估 technical correctness, novelty of approach, feasibility, comparison to alternatives
evidence:额外要求评估 experimental rigor, statistical significance, idea-evidence alignment, missing controls
writing:额外要求评估 clarity, logical flow, notation consistency, figure quality, related work coverage
completeness:额外要求评估 missing baselines, missing ablations, missing datasets, missing related work, reproducibility
Adversarial addition(仅 adversarial 模式):
Additionally: actively search for fatal flaws. A fatal flaw is anything that,
if true, would make the entire contribution invalid (incorrect proof, data leakage,
unfair comparison, published prior work). If you find one, flag it clearly.
Step 2: Review LLM 首轮审查
遵循 cross-model-review.md:不向 Review LLM 发送任何 Claude 的预判。
mcp__llm-review__chat:
system: {reviewer system prompt from Step 1}
message: |
## Artifact to Review
{artifact full text}
## Context from Knowledge Base
{relevant wiki context: related ideas/methods with status, related experiments, gap map entries}
## Review Instructions
Please provide:
1. **Strengths** (3-5 bullet points)
2. **Weaknesses** (ranked by severity, each with a concrete suggestion to fix)
3. **Questions** (things that are unclear or need clarification)
4. **Score** (1-10 with one-sentence justification)
5. **Verdict**: ready / needs-work / major-revision / rethink
6. **Idea-/method-level feedback**: For each idea or method referenced in the artifact, assess whether the evidence/justification is sufficient. List any ideas or methods that need stronger support.
7. **Knowledge gaps identified**: Any open questions or missing knowledge that would strengthen this work.
记录 Review LLM 返回的 threadId(用于 Step 3 多轮对话)。
Step 3: 多轮对话(hard / adversarial 模式)
若 --difficulty 为 standard,跳过此步。
对 Review LLM 的每个 weakness 进行回应(最多 3 轮):
Round N(N = 1, 2, 3):
Claude 分析 Review LLM 的 weaknesses,对每个 weakness 分类:
- 可反驳(rebuttal):Claude 有充分理由或 wiki 证据反驳 → 写出 rebuttal
- 承认(acknowledge):weakness 确实存在 → 承认并提出修复方案
- 需要更多信息(clarify):weakness 基于误解 → 提供澄清
将 Claude 的回应发送给 Review LLM:
mcp__llm-review__chat-reply:
threadId: {from Step 2}
message: |
Thank you for the review. Here are my responses:
{for each weakness: rebuttal / acknowledgment / clarification}
Please re-evaluate considering these responses. Update your score if warranted.
If --difficulty == adversarial: Also, please try harder to find any remaining
fatal flaws I may have missed.
Review LLM 回应新的评估和修正后的评分
若 Review LLM 的评分变化 < 0.5 且无新 weakness → 停止对话(收敛)
若已达 3 轮 → 停止对话
Step 4: 结构化输出
综合 Step 2 + Step 3 结果,生成结构化 Review Report:
# Review Report: {artifact title}
## Meta
- **Artifact type**: {idea / experiment / method / paper-draft / proposal}
- **Difficulty**: {standard / hard / adversarial}
- **Focus**: {method / evidence / writing / completeness / 全面}
- **Reviewer**: Review LLM(在 `.env` 中配置)
- **Rounds**: {1 for standard, N for hard/adversarial}
## Score: {final score}/10 — {verdict}
| Verdict | 含义 |
|---------|------|
| ready | 可直接使用或提交 |
| needs-work | 有明确的改进点,修复后可用 |
| major-revision | 核心部分需要重大修改 |
| rethink | 基本方向可能有问题,需重新考虑 |
## Strengths
1. {strength 1}
2. {strength 2}
...
## Weaknesses (by severity)
### Critical
- {weakness}: {具体描述} → **Fix**: {具体修复建议}
### Major
- {weakness}: {具体描述} → **Fix**: {具体修复建议}
### Minor
- {weakness}: {具体描述} → **Fix**: {具体修复建议}
## Questions
1. {question}
...
## Wiki Entity Mapping
### Ideas / methods needing stronger support
| Entity | Signal | Issue | Suggested action |
|--------|--------|-------|------------------|
| [[idea-slug]] | novelty_score 2/5 | Novelty argument is thin | Run /novelty rerun |
| [[method-slug]] | source_papers sparse | Missing source paper backing | Ingest the missing paper, then rerun /check |
### Knowledge gaps identified
| Gap | Related to | Suggested action |
|-----|-----------|------------------|
| {描述} | [[slug]] | /ingest, /exp-run, or /query |
### Suggested wiki updates
- `wiki/ideas/{slug}.md`: add risk factor from review
- `wiki/methods/{slug}.md`: tighten Tradeoff profile / Limitations
- `wiki/graph/open_questions.md`: will be updated on next rebuild
## Dialogue History (hard/adversarial only)
### Round 1
**Review LLM**: {summary of initial review}
**Claude**: {summary of rebuttals/acknowledgments}
### Round 2
**Review LLM**: {updated assessment}
...
## Actionable Items (ranked)
1. [CRITICAL] {action item}
2. [MAJOR] {action item}
3. [MINOR] {action item}
Constraints
- 审稿独立性:严格遵循
shared-references/cross-model-review.md,不向 Review LLM 泄露 Claude 的预判
- 不修改 wiki:review 只输出建议,不直接修改任何 wiki 页面。wiki 修改由调用方(如 /refine)执行
- score 必须有 justification:不接受没有理由的分数
- weakness 必须有 fix:每个 weakness 必须附带具体的、可操作的修复建议,不接受空洞批评
- entity-level mapping 必须:输出必须包含 Wiki Entity Mapping 部分,将 review 发现映射到具体 wiki 实体(ideas、methods 等)
- adversarial 模式必须搜索致命缺陷:如已发表的完全相同工作、证明错误、数据泄露等
- 多轮对话最多 3 轮:防止无限循环,若 3 轮后仍未收敛则以当前状态输出
- 引用 wiki 页面时使用 [[slug]]:所有对 wiki 页面的引用使用 wikilink 语法
Error Handling
- artifact 找不到:提示用户检查 slug 或路径,列出可能的候选页面
- Review LLM 不可用:降级为 Claude 自我审查模式,报告标注「single-model review, cross-model verification unavailable」,建议用户稍后用 Review LLM 重新审查
- wiki 为空:正常执行审查,但 Wiki Entity Mapping 部分标注「wiki empty, no entity mapping available」
- artifact 太长:若超过 Review LLM 上下文窗口,按 section 分段审查,最后合并
- Review LLM 返回无效响应:重试一次,若仍无效则使用 Claude 自审降级方案
- 多轮对话中 Review LLM 不收敛:3 轮后强制结束,输出最后一轮的评分和总结
Dependencies
Tools(via Bash)
MCP Servers
mcp__llm-review__chat — Review LLM 首轮审查(Step 2)
mcp__llm-review__chat-reply — Review LLM 多轮对话(Step 3)
Claude Code Native
Read — 读取 artifact 和 wiki 页面
Glob — 查找 artifact 对应的 wiki 页面
Shared References
.claude/skills/shared-references/cross-model-review.md — 审稿独立性原则(必读)
Called by
/ideate Phase 4(审查 top ideas)
/refine 每轮迭代(审查当前版本)
/exp-design --review(审查实验计划)
1---2name: review3description: 通用跨模型审查:Review LLM 对任意研究制品进行独立评审,输出结构化评分、wiki 实体映射与改进建议4---56# /review78> 对任意研究制品(idea、proposal、experiment plan、paper draft、method)进行跨模型审查。9> 使用 Review LLM 作为独立审稿人,输出结构化评分、可操作的改进建议,以及与 wiki 实体的映射10> (哪些 ideas/methods 需要加强,哪些 gaps 被发现)。11> 支持三种难度级别(standard / hard / adversarial)和四种审查焦点。12> 可独立使用,也被 /ideate、/refine、/exp-design 调用。1314## Inputs1516- `artifact`:要审查的制品,以下之一:17 - wiki 页面的 slug(如 `sparse-lora-for-edge-devices`,从 ideas/experiments/methods/ 中查找)18 - 文件路径(如 `wiki/outputs/paper-draft-v1.md`)19 - 自由文本(直接粘贴的 proposal 或 idea 描述)20- `--difficulty`(可选,默认 `standard`):21 - `standard`:单轮审查,给出结构化反馈22 - `hard`:多轮对话(最多 3 轮),Claude 对每个 weakness 进行 rebuttal23 - `adversarial`:多轮对话(最多 3 轮),Review LLM 额外尝试找致命缺陷,模拟最严苛的审稿人24- `--focus`(可选,默认全面审查):25 - `method`:聚焦方法设计的正确性、创新性、可行性26 - `evidence`:聚焦证据是否充分、实验是否严谨、idea/method 是否得到充分支撑27 - `writing`:聚焦表达清晰度、结构组织、论证逻辑28 - `completeness`:聚焦是否遗漏关键内容(相关工作、ablation、baseline)2930## Outputs3132- **Review Report**(输出到终端):33 - Overall Score(1-10)34 - Strengths(优点列表)35 - Weaknesses(缺点列表,按严重程度排序)36 - Questions(审稿人的疑问)37 - Actionable Suggestions(可操作的改进建议,按优先级排序)38 - Wiki Entity Mapping(哪些 ideas/methods 需要加强,哪些 gaps 被发现)39 - Verdict:`ready` / `needs-work` / `major-revision` / `rethink`40- 若 `--difficulty >= hard`:额外包含多轮对话记录和最终修正后的评分41- 该 skill **不直接修改 wiki**,但会输出建议的 wiki 更新列表4243## Wiki Interaction4445### Reads46- `wiki/papers/*.md` — 查找制品引用的论文,验证引用正确性47- `wiki/concepts/*.md` — 理解制品涉及的技术概念48- `wiki/methods/*.md` — 检查制品依赖的 methods 当前状态49- `wiki/experiments/*.md` — 查找相关实验结果50- `wiki/ideas/*.md` — 如果审查的是 idea,检查其上下文51- `wiki/graph/context_brief.md` — 获取全局上下文52- `wiki/graph/open_questions.md` — 对照 gap map 检查完整性53- `.claude/skills/shared-references/cross-model-review.md` — 审稿独立性原则5455### Writes56- **无**。Review 是只读查询操作。57 - 审查结果输出到终端,由用户或调用方(如 /refine)决定是否应用。5859### Graph edges created60- **无**。6162## Workflow6364**前置**:确认工作目录为 wiki 项目根(包含 `wiki/`、`raw/`、`tools/` 的目录)。6566### Step 1: 加载上下文67681. **解析 artifact**:69 - 若为 slug:按顺序在 `wiki/ideas/`、`wiki/experiments/`、`wiki/methods/`、`wiki/papers/`、`wiki/outputs/` 中查找 `{slug}.md`70 - 若为文件路径:直接读取71 - 若为自由文本:直接使用722. **确定 artifact 类型**:idea / experiment / method / paper-draft / proposal / other733. **加载相关 wiki 上下文**:74 - 读取 `wiki/graph/context_brief.md` 获取全局视角75 - 读取 `wiki/graph/open_questions.md` 获取知识缺口列表76 - 根据 artifact 类型,加载相关 wiki 页面:77 - idea → 其 origin_gaps(concepts/topics)、相关 papers78 - experiment → 其 linked_idea、相关 experiments79 - method → 其 source_papers 和 parent_methods80 - paper-draft → 其引用的所有 wiki 页面814. **读取 cross-model-review.md**:确认 Review LLM 独立性原则825. **构建 reviewer system prompt**(根据 --focus):8384 **Base prompt(所有 focus):**85 ```86 You are a senior ML researcher reviewing a research artifact.87 Be thorough, specific, and constructive. For every weakness, suggest a concrete fix.88 Score on a 1-10 scale where:89 - 1-3: Fundamental flaws, not salvageable in current form90 - 4-5: Significant issues but core idea may have merit91 - 6-7: Solid work with clear areas for improvement92 - 8-9: Strong work, minor issues only93 - 10: Exceptional, publication-ready94 ```9596 **Focus-specific additions:**97 - `method`:额外要求评估 technical correctness, novelty of approach, feasibility, comparison to alternatives98 - `evidence`:额外要求评估 experimental rigor, statistical significance, idea-evidence alignment, missing controls99 - `writing`:额外要求评估 clarity, logical flow, notation consistency, figure quality, related work coverage100 - `completeness`:额外要求评估 missing baselines, missing ablations, missing datasets, missing related work, reproducibility101102 **Adversarial addition(仅 adversarial 模式):**103 ```104 Additionally: actively search for fatal flaws. A fatal flaw is anything that,105 if true, would make the entire contribution invalid (incorrect proof, data leakage,106 unfair comparison, published prior work). If you find one, flag it clearly.107 ```108109### Step 2: Review LLM 首轮审查110111**遵循 cross-model-review.md**:不向 Review LLM 发送任何 Claude 的预判。112113```114mcp__llm-review__chat:115 system: {reviewer system prompt from Step 1}116 message: |117 ## Artifact to Review118 {artifact full text}119120 ## Context from Knowledge Base121 {relevant wiki context: related ideas/methods with status, related experiments, gap map entries}122123 ## Review Instructions124 Please provide:125 1. **Strengths** (3-5 bullet points)126 2. **Weaknesses** (ranked by severity, each with a concrete suggestion to fix)127 3. **Questions** (things that are unclear or need clarification)128 4. **Score** (1-10 with one-sentence justification)129 5. **Verdict**: ready / needs-work / major-revision / rethink130 6. **Idea-/method-level feedback**: For each idea or method referenced in the artifact, assess whether the evidence/justification is sufficient. List any ideas or methods that need stronger support.131 7. **Knowledge gaps identified**: Any open questions or missing knowledge that would strengthen this work.132```133134记录 Review LLM 返回的 `threadId`(用于 Step 3 多轮对话)。135136### Step 3: 多轮对话(hard / adversarial 模式)137138若 `--difficulty` 为 `standard`,跳过此步。139140**对 Review LLM 的每个 weakness 进行回应**(最多 3 轮):141142**Round N(N = 1, 2, 3):**1431441. Claude 分析 Review LLM 的 weaknesses,对每个 weakness 分类:145 - **可反驳(rebuttal)**:Claude 有充分理由或 wiki 证据反驳 → 写出 rebuttal146 - **承认(acknowledge)**:weakness 确实存在 → 承认并提出修复方案147 - **需要更多信息(clarify)**:weakness 基于误解 → 提供澄清1481492. 将 Claude 的回应发送给 Review LLM:150 ```151 mcp__llm-review__chat-reply:152 threadId: {from Step 2}153 message: |154 Thank you for the review. Here are my responses:155156 {for each weakness: rebuttal / acknowledgment / clarification}157158 Please re-evaluate considering these responses. Update your score if warranted.159 If --difficulty == adversarial: Also, please try harder to find any remaining160 fatal flaws I may have missed.161 ```1621633. Review LLM 回应新的评估和修正后的评分1641654. 若 Review LLM 的评分变化 < 0.5 且无新 weakness → 停止对话(收敛)1665. 若已达 3 轮 → 停止对话167168### Step 4: 结构化输出169170综合 Step 2 + Step 3 结果,生成结构化 Review Report:171172```markdown173# Review Report: {artifact title}174175## Meta176- **Artifact type**: {idea / experiment / method / paper-draft / proposal}177- **Difficulty**: {standard / hard / adversarial}178- **Focus**: {method / evidence / writing / completeness / 全面}179- **Reviewer**: Review LLM(在 `.env` 中配置)180- **Rounds**: {1 for standard, N for hard/adversarial}181182## Score: {final score}/10 — {verdict}183184| Verdict | 含义 |185|---------|------|186| ready | 可直接使用或提交 |187| needs-work | 有明确的改进点,修复后可用 |188| major-revision | 核心部分需要重大修改 |189| rethink | 基本方向可能有问题,需重新考虑 |190191## Strengths1921. {strength 1}1932. {strength 2}194...195196## Weaknesses (by severity)197198### Critical199- {weakness}: {具体描述} → **Fix**: {具体修复建议}200201### Major202- {weakness}: {具体描述} → **Fix**: {具体修复建议}203204### Minor205- {weakness}: {具体描述} → **Fix**: {具体修复建议}206207## Questions2081. {question}209...210211## Wiki Entity Mapping212213### Ideas / methods needing stronger support214| Entity | Signal | Issue | Suggested action |215|--------|--------|-------|------------------|216| [[idea-slug]] | novelty_score 2/5 | Novelty argument is thin | Run /novelty rerun |217| [[method-slug]] | source_papers sparse | Missing source paper backing | Ingest the missing paper, then rerun /check |218219### Knowledge gaps identified220| Gap | Related to | Suggested action |221|-----|-----------|------------------|222| {描述} | [[slug]] | /ingest, /exp-run, or /query |223224### Suggested wiki updates225- `wiki/ideas/{slug}.md`: add risk factor from review226- `wiki/methods/{slug}.md`: tighten Tradeoff profile / Limitations227- `wiki/graph/open_questions.md`: will be updated on next rebuild228229## Dialogue History (hard/adversarial only)230231### Round 1232**Review LLM**: {summary of initial review}233**Claude**: {summary of rebuttals/acknowledgments}234235### Round 2236**Review LLM**: {updated assessment}237...238239## Actionable Items (ranked)2401. [CRITICAL] {action item}2412. [MAJOR] {action item}2423. [MINOR] {action item}243```244245## Constraints246247- **审稿独立性**:严格遵循 `shared-references/cross-model-review.md`,不向 Review LLM 泄露 Claude 的预判248- **不修改 wiki**:review 只输出建议,不直接修改任何 wiki 页面。wiki 修改由调用方(如 /refine)执行249- **score 必须有 justification**:不接受没有理由的分数250- **weakness 必须有 fix**:每个 weakness 必须附带具体的、可操作的修复建议,不接受空洞批评251- **entity-level mapping 必须**:输出必须包含 Wiki Entity Mapping 部分,将 review 发现映射到具体 wiki 实体(ideas、methods 等)252- **adversarial 模式必须搜索致命缺陷**:如已发表的完全相同工作、证明错误、数据泄露等253- **多轮对话最多 3 轮**:防止无限循环,若 3 轮后仍未收敛则以当前状态输出254- **引用 wiki 页面时使用 [[slug]]**:所有对 wiki 页面的引用使用 wikilink 语法255256## Error Handling257258- **artifact 找不到**:提示用户检查 slug 或路径,列出可能的候选页面259- **Review LLM 不可用**:降级为 Claude 自我审查模式,报告标注「single-model review, cross-model verification unavailable」,建议用户稍后用 Review LLM 重新审查260- **wiki 为空**:正常执行审查,但 Wiki Entity Mapping 部分标注「wiki empty, no entity mapping available」261- **artifact 太长**:若超过 Review LLM 上下文窗口,按 section 分段审查,最后合并262- **Review LLM 返回无效响应**:重试一次,若仍无效则使用 Claude 自审降级方案263- **多轮对话中 Review LLM 不收敛**:3 轮后强制结束,输出最后一轮的评分和总结264265## Dependencies266267### Tools(via Bash)268- 无直接工具调用(review 不需要确定性工具)269270### MCP Servers271- `mcp__llm-review__chat` — Review LLM 首轮审查(Step 2)272- `mcp__llm-review__chat-reply` — Review LLM 多轮对话(Step 3)273274### Claude Code Native275- `Read` — 读取 artifact 和 wiki 页面276- `Glob` — 查找 artifact 对应的 wiki 页面277278### Shared References279- `.claude/skills/shared-references/cross-model-review.md` — 审稿独立性原则(必读)280281### Called by282- `/ideate` Phase 4(审查 top ideas)283- `/refine` 每轮迭代(审查当前版本)284- `/exp-design --review`(审查实验计划)