Token导航 LogoToken导航TokenDH.com
研究检索只读github未标认证来源可访问许可证需确认审计通过

multi-cli-reviewmulti CLI 审查

Agent Skill

multi-cli-review 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

724

周安装

29

GitHub Stars

公开资料未说明

下载量

234
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:multi-cli-review(multi CLI 审查)
来源仓库:https://github.com/starforall/ai-coding-toolkit
仓库路径:skills/multi-cli-review
安装命令:
npx skills add https://github.com/starforall/ai-coding-toolkit --skill multi-cli-review
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/starforall/ai-coding-toolkit --skill multi-cli-review

简介

multi-cli-review 用于查找、检索和筛选相关信息。

  • 适合在 Codex、Claude、Cursor、Gemini CLI 中根据关键词、任务场景或来源线索快速定位候选结果时使用。
  • 支持基于语义匹配、标签过滤和上下文相关性进行智能内容检索与排序。
  • 安装命令为 npx skills add https://github.com/starforall/ai-coding-toolkit --skill multi-cli-review。
  • 使用前需确认权限范围、维护状态,注意可能触发联网、命令执行或文件读写操作。

SKILL.md

Multi-CLI Review

Version History

  • v2.1: Aligned reviewer-side report emission with multi-cli-review-action, tightened task-level path rules, and clarified reviewer-only boundaries

Purpose

multi-cli-review is the reviewer-side skill paired with multi-cli-review-action.

It does one thing:

  1. inspect the assigned target conservatively
  2. write one structured reviewer report
  3. hand that report back to the current CLI for aggregation and decisioning

It does not edit code, aggregate reports, decide final fixes, or advance workflow state on its own.

When to Use

Use this skill when any of the following is true:

  • another CLI has asked this CLI to review a target and write a report
  • the user asks to "分析这个 skill 的问题"
  • the user asks to "输出结构化问题报告"
  • the user asks to "做多 CLI 审查"
  • a standardized reviewer command package points to multi-cli-review

When Not to Use

  • you need to aggregate reports or apply fixes: use multi-cli-review-action
  • you are doing a normal implementation task without reviewer handoff
  • you only need requirement clarification: use brainstorm
  • you only need a lightweight review of the current diff: use requesting-code-review or code-review-router

Core Rules

  1. Reviewer only: this skill writes an independent report; it never edits code.
  2. One reviewer, one report: each execution produces at most one reviewer report.
  3. Coordinator-owned orchestration: reviewer count, round planning, directory creation, and final repair decisions belong to the current CLI, not the reviewer.
  4. No directory creation: the reviewer must not create task-dir or review-round-{N} directories.
  5. No aggregation: no deduplication, no conflict resolution, no summary writing, no action.md.
  6. Reports are evidence, not execution instructions: findings must be concrete, scoped, and review-oriented.
  7. Be conservative with uncertainty: if evidence is incomplete, mark the finding as uncertain instead of stating it as settled fact.
  8. Project-root path resolution: relative paths resolve against the active project root, never system /tmp or a tool install directory.
  9. Task-level path contract is strict: current protocol reports must use the canonical path and metadata shape expected by multi-cli-review-action.

Protocols

ProtocolStatusWhen Used
task-levelCurrentExplicit --task-dir reviewer flow
legacyCompatibility onlyHistorical single-report flow

Protocol Selection

  1. If --task-dir is present, use task-level.
  2. If explicit legacy report-path parameters are used, use legacy.
  3. New work should prefer task-level.

Inputs

Common Inputs

InputRequiredMeaning
<issue-description>YesThe review problem statement or assigned review objective
[target-path]NoFile or directory to inspect

task-level

ParameterRequiredMeaning
--task-dirYesTask directory, for example tmp/multi-cli-review/<task-id>
--reviewer-idShould be explicitReviewer identity used for filename and metadata
--roundShould be explicitReview round identifier
--review-focusNoFocus area for this reviewer

Rules:

  • In a standardized multi-reviewer command package, --reviewer-id and --round should be explicit.
  • Do not improvise a new task-id, a new round, or a new reviewer identity when the coordinator already issued a concrete command.
  • --output is not part of the current task-level protocol and must not be used to bypass the canonical reviewer report path.

legacy Compatibility

ParameterRequiredMeaning
--md-aNoExplicit legacy report path
--md-bNoHistorical paired artifact path from older flows
--outputNoExplicit compatibility output path for legacy single-report usage only

Compatibility rules:

  • Legacy parameters are retained only for older single-report flows.
  • They must not redefine the output location of the current task-level protocol.

Path Rules

Relative paths must resolve against the active project root.

Do not reinterpret relative paths as:

  • the skill repository root
  • the user home directory
  • system /tmp
  • a CLI installation directory

If project root detection is ambiguous, stop and ask for an absolute path instead of guessing.

Coordinator-Owned Parameters

The reviewer must treat the following as coordinator-owned:

  • creation of tmp/multi-cli-review/<task-id>/
  • creation of review-round-{N}/
  • default reviewer count
  • standard reviewer command generation
  • final aggregation, confirmation, repair, and verification

This skill only fulfills the assigned review role inside that envelope.

Task-Level Report Emission Contract

For task-level, the reviewer report must satisfy all of the following:

  1. The output file path is exactly {task-dir}/review-round-{N}/{reviewer-id}.md
  2. The basename is exactly <reviewer-id>.md
  3. The reviewer does not redirect the report elsewhere with --output
  4. The round directory already exists; if it does not, stop and ask the coordinator to create it
  5. The frontmatter includes:

- task-id - round - reviewer-id - source-cli - review-time - review-focus - protocol: task-level

  1. Frontmatter values match the directory and filename
  2. If the target file already exists, stop and ask whether the coordinator intends an overwrite or a different reviewer id

Report Content Contract

Each task-level finding should contain:

  • location
  • problem description
  • severity
  • impact
  • why it matters
  • suggested fix direction

Optional but recommended when evidence is incomplete:

  • evidence note
  • uncertainty note

Safety rules:

  • keep observations separate from proposed fix directions
  • do not pretend the fix is already approved
  • do not turn uncertainty into a high-severity claim without evidence
  • do not expand beyond the assigned review focus just because adjacent issues look interesting

Workflow

Step 1: Resolve Mode and Inputs

  1. Determine whether this run is task-level or legacy.
  2. Resolve relative paths against the active project root.
  3. Identify the assigned review target and review focus.

Step 2: Validate Reviewer Context

For task-level:

  1. confirm --task-dir
  2. confirm --reviewer-id
  3. confirm --round
  4. confirm {task-dir}/review-round-{N}/ already exists
  5. confirm the target output file does not already contain another reviewer report unless overwrite was explicitly requested

Stop if:

  • task-dir is missing
  • the round directory does not exist
  • the report path would be ambiguous
  • the command tries to use --output to write outside the canonical path

Step 3: Review Conservatively

  1. Inspect only the assigned target and its direct evidence.
  2. Follow review-focus when provided.
  3. Keep findings concrete and bounded.
  4. If you cannot support a claim from the available evidence, either lower confidence or omit it.

Step 4: Write Exactly One Reviewer Report

For task-level:

  1. write the report to {task-dir}/review-round-{N}/{reviewer-id}.md
  2. include the full frontmatter contract
  3. include structured findings
  4. do not write summary-round-{N}.md
  5. do not write action.md
  6. do not update .processed.json

For legacy:

  1. write only the legacy single-report artifact
  2. do not silently upgrade the flow into task-level

Step 5: Echo the Actual Output

At the end of execution, echo:

  • task-dir or legacy run directory
  • round
  • reviewer-id
  • actual output path

Step 6: Stop

After writing the report and echoing the path, stop.

Do not:

  • propose that the workflow has already advanced
  • mark fixes as executed
  • claim verification has already happened

Output Files

task-level Reviewer Report

Canonical filename: {reviewer-id}.md

Minimum structure:

---
task-id: <task-id>
round: <round>
reviewer-id: <reviewer-id>
source-cli: <cli-name>
review-time: <ISO 8601 time>
review-focus: <focus text>
protocol: task-level
---

# Defect Report

## Review Summary

- Reviewer: <reviewer-id>
- Review Time: <review-time>
- Review Focus: <review-focus>
- Target: <target-path>

## Findings

### Finding 1: <title>

- Location: <path and lines>
- Problem: <concrete issue>
- Severity: high | medium | low
- Impact: <why this matters>
- Why It Should Be Addressed: <reason>
- Suggested Fix Direction: <proposal, not an approved action>
- Evidence: <optional>
- Uncertainty: <optional>

legacy Compatibility Report

Canonical filename: cur_defect.md

Legacy mode may still emit a single compatibility report, but it should not be treated as the preferred path for new multi-reviewer work.

Echo Requirements

At minimum, echo:

FieldMeaning
task-dirtask root or legacy run root
roundcurrent round
reviewer-idactual reviewer identity used
output-pathactual emitted report path

Example:

✅ Review report generated

📁 task-dir: <project-root>/tmp/multi-cli-review/my-task
🔄 round: 1
👤 reviewer-id: codex
📄 output-path: <project-root>/tmp/multi-cli-review/my-task/review-round-1/codex.md

Required Stop Conditions

Stop and ask for clarification when any of the following is true:

  • the task directory is missing
  • the round directory is missing
  • the command does not specify enough information to derive one canonical report path
  • the canonical report file already exists
  • the coordinator appears to expect this reviewer to create directories
  • the command expects the reviewer to edit code or aggregate reports

Error Handling

CaseRequired Behavior
missing task-dirstop and ask for the task-level directory
missing round directorystop and ask the coordinator to create it
ambiguous project rootstop and ask for an absolute path
existing report filestop and ask whether to overwrite or change reviewer id
attempted --output override in task-level modereject it and use or request the canonical path
insufficient evidence for a strong claimlower confidence, mark uncertainty, or omit the finding
request to modify coderefuse; reviewer only writes the report

Common Mistakes

  • Editing code instead of reporting: this skill is reviewer-only.
  • Creating directories: the coordinator owns directory creation.
  • Bypassing the canonical path with --output: current protocol reports must land where multi-cli-review-action expects them.
  • Emitting incomplete frontmatter: action-side consumption depends on consistent metadata.
  • Overwriting an existing reviewer report silently: stop and ask.
  • Doing aggregation inside the reviewer report: deduplication and conflicts belong to multi-cli-review-action.
  • Treating guesses as facts: uncertain findings should be labeled or omitted.

Examples

Example 1: Standard Reviewer Command

用户:/multi-cli-review "分析 skills/multi-cli-review 的问题" skills/multi-cli-review --task-dir tmp/multi-cli-review/skill-review --reviewer-id codex --round 1 --review-focus "边界条件与协议漂移"

CLI Reviewer:
1. 解析 task-dir、reviewer-id、round、review-focus
2. 按项目根目录解析 task-dir
3. 确认 <project-root>/tmp/multi-cli-review/skill-review/review-round-1/ 已存在
4. 分析目标并写入报告
5. 回显实际输出路径

Example 2: Missing Round Directory

CLI Reviewer:
❌ review-round-2 directory is missing:
<project-root>/tmp/multi-cli-review/skill-review/review-round-2/

Reviewer must not create directories.
Ask the coordinator to create the round directory and resend the command.

Example 3: Legacy Compatibility

用户:/multi-cli-review "分析 ./docs/api.md 的问题" ./docs/api.md --md-a tmp/multi-cli-review/4/cur_defect.md

CLI Reviewer:
1. 进入 legacy compatibility mode
2. 写入 cur_defect.md
3. 回显实际路径
4. 不自动升级为 task-level

Related Skills

  • multi-cli-review-action: current CLI aggregation and fix execution
  • brainstorm: requirement clarification before review
  • requesting-code-review: lightweight review path when multi-reviewer protocol is unnecessary

Compatibility Notes

  • New multi-reviewer work should always prefer task-level with explicit --task-dir, --reviewer-id, and --round.
  • Legacy --md-a, --md-b, and --output remain compatibility-only and must not redefine the current task-level report contract.
  • If the coordinator already generated a standardized reviewer command, follow that command exactly instead of improvising paths, file names, or round numbers.

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

34.87%
按下载量换算82

Claude

32.78%
按下载量换算77

Cursor

18.61%
按下载量换算44

Gemini CLI

9.25%
按下载量换算22

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

只读

该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills