ln-005-agent-reviewer

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Paths: File paths (
shared/
,
references/
,
../ln-*
) are relative to skills repo root. If not found at CWD, locate this SKILL.md directory and go up one level for repo root.

路径： 文件路径（
shared/
、
references/
、
../ln-*
）是相对于技能仓库根目录的。如果在当前工作目录（CWD）中找不到，请找到此SKILL.md文件所在目录，然后向上一级即可到达仓库根目录。

Agent Reviewer (Universal)

Agent审查工具（通用版）

Runs parallel external agent reviews on arbitrary context, critically verifies suggestions, returns filtered improvements.

对任意上下文并行运行外部Agent审查，严格验证建议，返回经过筛选的改进方案。

Purpose & Scope

用途与适用范围

Standalone utility in 0XX category (like ln-003, ln-004)
Delegate any context to codex-review + gemini-review as background tasks in parallel
Context always passed via file references (never inline in prompt)
Process results as they arrive (first-finished agent processed immediately)
Critically verify each suggestion; debate with agent if Claude disagrees
Return filtered, deduplicated, verified suggestions

属于0XX类别的独立实用工具（类似ln-003、ln-004）
将任意上下文并行委托给codex-review和gemini-review作为后台任务运行
上下文始终通过文件引用传递（绝不放在提示词内联）
结果到达即处理（先完成的Agent会被立即处理）
严格验证每条建议；如果Claude不认同，会与Agent进行辩论
返回经过筛选、去重、验证的建议

When to Use

适用场景

Manual invocation by user for independent review of any artifact
Called by any skill needing external second opinion on plans, decisions, documents
NOT tied to Linear, NOT tied to any pipeline
Works with any context that can be saved to a file

用户手动调用，对任意产物进行独立审查
任何需要对计划、决策、文档获取外部第二意见的技能都可调用
不绑定Linear，不绑定任何流水线
适用于所有可保存为文件的上下文

Parameters

参数

Parameter	Value
`review_type`	`contextreview`
`skill_group`	`005`
`prompt_template`	`shared/agents/prompt_templates/context_review.md`
`verdict_acceptable`	`CONTEXT_ACCEPTABLE`

参数	值
`review_type`	`contextreview`
`skill_group`	`005`
`prompt_template`	`shared/agents/prompt_templates/context_review.md`
`verdict_acceptable`	`CONTEXT_ACCEPTABLE`

Inputs

输入项

Input	Required	Description
`context_files`	Yes	List of file paths containing context to review (relative to CWD)
`identifier`	No	Short label for file naming (default: `review_YYYYMMDD_HHMMSS` )
`focus`	No	List of areas to focus on (default: all 6)
`review_title`	No	Human-readable title (default: `"Context Review"` )

Context delivery rule: Context is ALWAYS passed via files.

If context already exists as files (plans, docs, code) -> pass file paths directly
If context is a statement/decision from chat -> caller creates a temporary file in
```
.agent-review/context/
```
with the content, then passes the file path

输入项	是否必填	说明
`context_files`	是	包含待审查上下文的文件路径列表（相对于当前工作目录）
`identifier`	否	用于文件命名的短标签（默认值： `review_YYYYMMDD_HHMMSS` ）
`focus`	否	重点关注的领域列表（默认：全部6个领域）
`review_title`	否	人类可读的标题（默认值： `"Context Review"` ）

上下文传递规则： 上下文始终通过文件传递。

如果上下文已经以文件形式存在（计划、文档、代码）-> 直接传递文件路径
如果上下文是来自聊天的陈述/决策 -> 调用方需要在
```
.agent-review/context/
```
目录下创建一个包含该内容的临时文件，再传递该文件路径

Workflow

工作流

MANDATORY READ: Load

shared/references/agent_review_workflow.md

for Health Check, Ensure .agent-review/, Run Agents, Critical Verification + Debate, Aggregate + Return, Fallback Rules, Critical Rules, and Definition of Done. Load

shared/references/agent_delegation_pattern.md

for Reference Passing Pattern, Review Persistence Pattern, Agent Timeout Policy, and Debate Protocol.

强制阅读： 加载

shared/references/agent_review_workflow.md

了解健康检查、.agent-review目录初始化、运行Agent、严格验证+辩论、聚合+返回、降级规则、关键规则和完成定义。加载

shared/references/agent_delegation_pattern.md

了解引用传递模式、审查持久化模式、Agent超时策略和辩论协议。

Unique Steps (before shared workflow)

独有步骤（共享工作流之前）

Health check: per shared workflow, filter by
```
skill_group
```
=
```
005
```
.
Resolve identifier: If
```
identifier
```
not provided, generate
```
review_YYYYMMDD_HHMMSS
```
. Sanitize: lowercase, replace spaces with hyphens, ASCII only.
Ensure .agent-review/: per shared workflow. Additionally create
```
.agent-review/context/
```
subdir if it doesn't exist (for materialized context files).
Materialize context (if needed): If context is from chat/conversation (not an existing file):
- Write content to
```
.agent-review/context/{identifier}_context.md
```
- Add this path to
```
context_files
```
  list
Build prompt: Read template
```
shared/agents/prompt_templates/context_review.md
```
.
- Replace
```
{review_title}
```
  with title or
```
"Context Review"
```
- Replace
```
{context_refs}
```
  with bullet list:
```
- {path}
```
  per context file
- Replace
```
{focus_areas}
```
  with filtered subset or
```
"All default areas"
```
  if no focus specified
- Save to
```
.agent-review/{identifier}_contextreview_prompt.md
```
  (single shared file -- both agents read the same prompt)

健康检查： 按照共享工作流要求，按
```
skill_group
```
=
```
005
```
过滤。
解析标识符： 如果未提供
```
identifier
```
，则生成
```
review_YYYYMMDD_HHMMSS
```
。格式清理：转为小写，空格替换为连字符，仅保留ASCII字符。
初始化.agent-review目录： 按照共享工作流要求。如果
```
.agent-review/context/
```
子目录不存在则额外创建（用于存放物化的上下文文件）。
物化上下文（如有需要）： 如果上下文来自聊天/对话（不是现有文件）：
- 将内容写入
```
.agent-review/context/{identifier}_context.md
```
- 将该路径添加到
```
context_files
```
  列表
构建提示词： 读取模板
```
shared/agents/prompt_templates/context_review.md
```
。
- 将
```
{review_title}
```
  替换为指定标题或
```
"Context Review"
```
- 将
```
{context_refs}
```
  替换为项目符号列表：每个上下文文件对应一行
```
- {path}
```
- 将
```
{focus_areas}
```
  替换为筛选后的子集，如果未指定关注领域则替换为
```
"All default areas"
```
- 保存到
```
.agent-review/{identifier}_contextreview_prompt.md
```
  （单个共享文件——两个Agent都读取同一个提示词）

Shared Workflow Steps

共享工作流步骤

6-8) Run agents, Critical Verification + Debate, Aggregate + Return: per shared workflow.

```
{review_type}
```
= review_title or "Context Review" (for challenge template
```
{review_type}
```
placeholder)
```
{story_ref}
```
placeholder in challenge template = identifier

6-8) 运行Agent、严格验证+辩论、聚合+返回： 按照共享工作流执行。

```
{review_type}
```
= 审查标题或「Context Review」（用于挑战模板的
```
{review_type}
```
占位符）
挑战模板中的
```
{story_ref}
```
占位符 = 标识符

Output Format

输出格式

yaml

verdict: CONTEXT_ACCEPTABLE | SUGGESTIONS | SKIPPED
suggestions:
  - area: "logic | feasibility | completeness | consistency | best_practices | risk"
    issue: "What is wrong or could be improved"
    suggestion: "Specific actionable change"
    confidence: 95
    impact_percent: 15
    source: "codex-review"
    resolution: "accepted | accepted_after_debate | accepted_after_followup | rejected"

Agent stats and debate log per shared workflow output schema.

yaml

verdict: CONTEXT_ACCEPTABLE | SUGGESTIONS | SKIPPED
suggestions:
  - area: "logic | feasibility | completeness | consistency | best_practices | risk"
    issue: "What is wrong or could be improved"
    suggestion: "Specific actionable change"
    confidence: 95
    impact_percent: 15
    source: "codex-review"
    resolution: "accepted | accepted_after_debate | accepted_after_followup | rejected"

按照共享工作流输出规范输出Agent统计数据和辩论日志。

Verdict Escalation

裁决升级规则

No escalation. Suggestions are advisory only.
Caller decides how to apply accepted suggestions.

无需升级。 建议仅作参考。
调用方决定如何应用已接受的建议。

Reference Files

参考文件

Shared workflow:

shared/references/agent_review_workflow.md

Agent delegation pattern:

shared/references/agent_delegation_pattern.md

Prompt template (review):

shared/agents/prompt_templates/context_review.md

Review schema:

shared/agents/schemas/context_review_schema.json

Version: 1.0.0 Last Updated: 2026-02-25

共享工作流：

shared/references/agent_review_workflow.md

Agent委托模式：

shared/references/agent_delegation_pattern.md

提示词模板（审查）：

shared/agents/prompt_templates/context_review.md

审查Schema：

shared/agents/schemas/context_review_schema.json

版本： 1.0.0 最后更新： 2026-02-25