Token导航 LogoToken导航TokenDH.com
开发只读github未标认证来源可访问许可证需确认审计异常

re-extract-requirements重新提取需求

Agent Skill

re-extract-requirements 用于处理 GitHub 仓库、Issue、Pull Request 和代码协作信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要围绕仓库状态、代码变更或协作事项进行整理时使用。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

428

周安装

18

GitHub Stars

公开资料未说明

下载量

150
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:re-extract-requirements(重新提取需求)
来源仓库:https://github.com/caldiaworks/caldiaworks-marketplace
仓库路径:skills/re-extract-requirements
安装命令:
npx skills add https://github.com/caldiaworks/caldiaworks-marketplace --skill re-extract-requirements
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/caldiaworks/caldiaworks-marketplace --skill re-extract-requirements

简介

re-extract-requirements 用于处理 GitHub 仓库、Issue、Pull Request 和代码协作信息。

  • 适合在 Codex、Claude、Cursor、Gemini CLI 中围绕仓库状态、代码变更或协作事项进行整理。
  • 通过 npx skills add 命令从指定仓库安装,需结合原始 README 确认具体用法。
  • 安装前建议确认权限范围、维护状态及是否触发联网、命令执行或文件读写操作。
  • 该技能适用于需要自动化提取和重组开发需求与协作信息的场景。

SKILL.md

Requirements Extraction — Reverse Engineering Phase 3

Extract system requirements from source code and logic diagrams using EARS notation. This skill reads targets from manifest.json (populated by Phase 2) and implements internal Doer/Critic verification within a single execution.

Three Principles

1. Code is Truth

  • Requirements reflect ACTUAL code behavior, not intended behavior
  • Bugs are documented as requirements with note: [Suspected Bug]
  • Separate As-Is requirements from Question List

2. Traceability to Line

  • Every requirement MUST cite file:line reference
  • Requirements without evidence are rejected as hallucination

3. Behavior over Intent

  • Requirements describe observable behavior: input → output, side effects
  • Use concrete values from code (e.g., "retry 3 times", not "retry several times")

EARS Notation Types

TypePatternWhen to Use
UbiquitousSystem SHALL [action]Unconditional, always-active behavior
Event-drivenWHEN [trigger], system SHALL [action]Triggered by external input or event
UnwantedIF [condition], system SHALL [action]Error handling, validation failures
State-drivenWHILE [state], system SHALL [action]Background services, stateful behavior
OptionalWHERE [config/feature], system MAY [action]Configurable or optional behavior

Execution

Step 1: Load Manifest and Determine Targets

Read docs/reverse/{analysis}/manifest.json.

Verify:

  • phase2.status is "verified" — if not, report error and stop
  • phase2.targets_for_phase3 contains entries

Determine targets:

  • If component argument is provided, process only that component
  • If omitted, process all entries in phase3.remaining (or all targets_for_phase3 if phase3 hasn't started)

Set phase3.status to "in_progress".

Load language reference if available.

Step 2: Doer — Extract Requirements

For each target component:

  1. Read the source code file
  2. Read the logic diagram from Phase 2 (if available)
  3. Systematically scan for each EARS type:

Ubiquitous — Scan for unconditional actions:

  • No surrounding if/switch/while
  • Direct method calls in main flow
  • Always-executed logging, persistence

Event-driven — Scan for event triggers:

  • Request/command handling methods
  • Event subscribers, message handlers
  • Callback invocations

Unwanted — Scan for error conditions:

  • Null checks, validation, type guards
  • Exception handling (try/catch)
  • Error response returns

State-driven — Scan for stateful behavior:

  • Background service loops
  • While conditions with state checks
  • Timer-based processing

Optional — Scan for configurable behavior:

  • Configuration/environment checks
  • Feature flags
  • Optional logging or processing

For each requirement, record:

  • EARS type and requirement statement
  • Source reference (file:line)
  • Evidence (code snippet)
  • Concrete values (thresholds, timeouts, error messages)

Step 3: Critic — Verify Each Requirement

For EVERY extracted requirement:

  1. Re-read the source file at the cited line
  2. Verify:

- Line number exists in source file - Code at that line matches the requirement description - Condition boundaries are correct (>= vs >, == vs!=) - Concrete values match exactly - No assumed behavior that doesn't exist in code

  1. Classify:

- ✅ APPROVED: Verified against source - ⚠️ CORRECTED: Line number or detail corrected - ❌ REJECTED: Hallucination — not present in code

Remove rejected requirements. Correct inaccurate ones. Log all decisions.

Step 4: Generate Output

Write to docs/reverse/{analysis}/03-requirements-{component}.md:

# EARS Requirements: {component}

**Analysis Date**: {YYYY-MM-DD}
**Source File**: [{file}]({file})
**Requirements Extracted**: {count}
**Verification**: All requirements verified against source

## Requirements Table

| REQ-ID | EARS Type | Requirement | Source | Concrete Values | Notes |
|:-------|:----------|:------------|:-------|:----------------|:------|
| REQ-U001 | Ubiquitous | {statement} | [{file}:{line}]({file}:{line}) | {values} | {notes} |
| REQ-E001 | Event-driven | WHEN {trigger}, system SHALL {action}. | [{file}:{line}]({file}:{line}) | {values} | {notes} |
| REQ-W001 | Unwanted | IF {condition}, system SHALL {action}. | [{file}:{line}]({file}:{line}) | {values} | {notes} |

## Detailed Requirements

### Ubiquitous Requirements

#### REQ-U001: {title}
**Statement**: {EARS statement}

**Source**: [{file}:{line}]({file}:{line})

**Evidence**:

// Line {N} {code snippet}


**Concrete Values**: {list of specific values}

**Verification**: ✅ Approved — {brief justification}

---

{repeat for all requirements}

## Verification Log

| REQ-ID | Verdict | Issue | Corrective Action | Evidence |
| --- | --- | --- | --- | --- |
| REQ-U001 | ✅ Approved | — | — | Line {N} verified |
| REQ-E003 | ❌ Rejected | Hallucination | Removed | No validation at L50 |

## Concrete Values Summary

### Thresholds and Constants

| Value Type | Value | Source | Context |
| --- | --- | --- | --- |

### Error Messages

| Error Type | Message | Source | Trigger |
| --- | --- | --- | --- |

### Configuration Keys

| Key | Type | Default | Source | Effect |
| --- | --- | --- | --- | --- |

## Question List

### Suspected Bugs

- **{description}** — [{file}:{line}](https://github.com/caldiaworks/caldiaworks-marketplace/blob/HEAD/skills/re-extract-requirements/%7Bfile%7D:%7Bline%7D)

### Unclear Logic

- **{description}** — [{file}:{line}](https://github.com/caldiaworks/caldiaworks-marketplace/blob/HEAD/skills/re-extract-requirements/%7Bfile%7D:%7Bline%7D)

Step 5: Update Manifest

After each component:

  • Add to phase3.completed: {"component": "{name}", "output": "03-requirements-{name}.md", "verification": null}
  • Remove from phase3.remaining

After all components:

  • Set phase3.status to "completed"
  • Update updated timestamp

Validation Before Completion

  • [ ] Every requirement has an EARS type classification
  • [ ] Every requirement has a file:line source reference
  • [ ] All requirements verified by Critic step (re-read source)
  • [ ] Verification log documents all approval/rejection decisions
  • [ ] Concrete values extracted from code (not guessed)
  • [ ] Suspected bugs noted in Question List, not removed from requirements
  • [ ] No ambiguous words (appropriate, handle, properly, etc.)

Prohibited Actions

  • Do NOT output requirements as chat response only — MUST create markdown file
  • Do NOT add requirements for behavior not present in code
  • Do NOT use vague terms ("appropriate", "reasonable", "several")
  • Do NOT skip the Critic verification step
  • Do NOT publish requirements without line numbers
  • Do NOT modify source code files

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

36.19%
按下载量换算54

Claude

29.42%
按下载量换算44

Cursor

19.92%
按下载量换算30

Gemini CLI

9.72%
按下载量换算15

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

未通过

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。来源安全扫描存在 warning/failed 结果,不能写成本站确认安全。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