Token导航 LogoToken导航

Agent Skill · 研究检索

spec-workflow(规格工作流程)

spec-workflow 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。

spec-workflow 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为需要联网;许可证信息还不明确,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 50,882。

总安装 605周安装 26GitHub Stars 1下载 212核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

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:spec-workflow(规格工作流程)
来源仓库:https://github.com/stack-shifter/skills
仓库路径:skills/spec-workflow
安装命令:
npx skills add https://github.com/stack-shifter/skills --skill spec-workflow
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/stack-shifter/skills --skill spec-workflow

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

spec-workflow 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词或任务场景快速定位候选结果时使用。适用于研究检索类任务,支持基于来源线索和仓库路径进行信息聚合与过滤。

通过 npx skills add 命令从指定 GitHub 仓库安装,需结合原始 README 确认具体用法。安装前建议核实权限范围、维护状态,并注意是否涉及联网、命令执行或文件读写操作。适用宿主包括 Codex、Claude、Cursor、Gemini CLI,接入前应确认版本、权限和运行环境要求。

来源仓库内路径:skills/spec-workflow

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/stack-shifter/skills --skill spec-workflow;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要需要联网,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

spec-workflow 怎么安装?

在终端执行 npx skills add https://github.com/stack-shifter/skills --skill spec-workflow。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

spec-workflow 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

spec-workflow 需要什么权限?

权限级别是需要联网。该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

spec-workflow 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

spec-workflow 用的人多吗?

公开数据是累计安装 605 次,近一周安装 26 次,下载 212 次,GitHub 1 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

spec-workflow 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 第三方扫描已通过,常见的高风险行为没有命中。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
spec-workflow

规格工作流程 · 本页

6051需要联网未明确通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

Spec Workflow Skill

Purpose

Keep one stable spec per feature as the source of truth. Get user approval before starting implementation. Check in after each logical chunk of work before continuing.

  • Spec = what — locked requirements, stable across the entire feature
  • Plan mode = how — in-session design derived from the spec before each implementation run

When to Use

  • Spec-first workflow for a new feature
  • Stable requirements document driving phased implementation
  • "Do only the next phase" loop
  • Locked requirements with explicit user approval gates

Typical trigger phrases:

  • Spec: build...
  • "create a spec for this feature"
  • "implement the next phase"
  • "use the spec as locked requirements"
  • "set this up in docs/specs"

Do not use for trivial one-off edits unless the user explicitly asks for the workflow.

Canonical Layout

docs/
  specs/
    010_task-management.spec.md
    020_user-invitations.spec.md

If CLAUDE.md exists in the repository, treat it as an additional instruction source alongside AGENTS.md.

Naming

File names: <id>_<slug>.spec.md

Examples: 010_task-management, 011_user-invitations, 012_billing-webhooks

To choose the next prefix, list existing spec files and increment by 1. Do not use vague slugs like stuff, misc, or new-feature.

Spec File

docs/specs/<id>_<slug>.spec.md must contain:

# <Feature Title>

## Overview

## Scope / Non-Scope

### In Scope

### Out of Scope

## Requirements

## Acceptance Criteria

## What This Touches

> A loose list of areas, files, and surfaces likely affected. For human reference — not an ordered plan.

- ...

When drafting, mark every assumption or gap inline rather than guessing:

[NEEDS CLARIFICATION: <specific question>]

Examples:

- Users receive an email on assignment [NEEDS CLARIFICATION: should this also fire for reassignment?]
- Auth method not specified [NEEDS CLARIFICATION: email/password, SSO, or OAuth?]

Do not resolve ambiguities by assuming. Surface them in the draft and ask the user to fill them in before the spec is treated as locked.

After writing a new spec, always output a summary card:

Spec:    <id>_<slug>.spec.md
Branch:  dev-<slug>
Summary: <one-sentence description of the feature>
Open Questions: <count of [NEEDS CLARIFICATION] markers>

Holistic Validation Pass

Run this before drafting or executing against a spec:

  1. Surface audit — find every symbol, config key, route, UI string, contract, and behavior the spec touches. Identify all impacted layers — app, UI, API, data, infrastructure, tests, docs — not just files already named.
  2. Scope consistency — check for contradictions between Overview, Scope, Requirements, and Acceptance Criteria. If an acceptance criterion requires changing an out-of-scope layer, narrow the criterion or bring the layer into scope.
  3. Phase executability — verify each planned phase can be validated using only work from that phase and earlier. Reorder phases when later-phase work is a prerequisite.
  4. Explicit ownership — name every caller, consumer, adapter, handler, test, doc, and config that must change. Do not leave high-risk coupling as implied cleanup.
  5. Contract boundaries — distinguish product, application, storage, integration, and infrastructure behavior. State clearly when a change affects one boundary but intentionally leaves another unchanged.

Present findings in order: contradictions → missing surface → phase/ordering risks → test gaps → doc fixes.

Workflow

Stage 1: Create or refine the spec

  1. Gather context from the repo and the user request.
  2. Run the holistic validation pass.
  3. Draft the spec in docs/specs/ (or the repo's established location). Mark every ambiguity or missing detail with [NEEDS CLARIFICATION: <specific question>] rather than assuming.
  4. Output the summary card.
  5. After outputting the summary card, print all [NEEDS CLARIFICATION] markers as a numbered list so the user can answer them without opening the file. The spec is not locked while markers remain.

When a spec already exists, treat it as the source of truth and run the holistic validation pass before accepting it as execution-ready.

Stage 2: Implementation

Before starting implementation, check the current git branch. If on main or master, ask the user to confirm before creating a dev-<slug> branch (e.g., dev-010-task-management). Do not create it automatically. If already on a dev-* branch, proceed.

Use the locked spec as requirements. Get user approval before executing each phase of work. Stop after each phase and wait for explicit instruction before continuing.

Spec Lock Rule

Treat the spec as locked once the user approves it or asks for implementation without requesting requirement changes.

Do not start implementation while any [NEEDS CLARIFICATION] markers remain. Resolve them first, then treat the cleaned spec as the lockable source of truth.

During implementation, the user may change direction. Use judgment to decide whether the change is small or large:

Small change — absorb it. Update the spec inline and continue. A change is small if it:

  • doesn't invalidate work already completed
  • doesn't change the core approach or architecture
  • can be absorbed into the current or next phase without restructuring

Large change — stop and escalate before touching anything. A change is large if it:

  • invalidates completed work
  • significantly expands or restructures scope
  • changes the core approach or introduces new surfaces

Do not silently absorb a large change by treating it as a small one.

Spec Change Escalation

When a change is too large to absorb inline, say so explicitly before proceeding. Present:

  1. What the user is asking for
  2. Why it is a large change (what it invalidates or restructures)
  3. Proposed spec revision
  4. Revised implementation approach

Wait for user approval before editing the spec or continuing implementation.

Choosing the Active Spec

When multiple specs exist, determine the active file in this order:

  1. Explicit user-provided file or feature ID
  2. Files referenced in the current task context
  3. Files matching the current branch or active feature name
  4. The only matching spec in the repository

If multiple candidates remain, ask the user which to use.

Guardrails

  • Do not advance past the current phase without explicit user instruction.
  • Do not silently change locked requirements by editing the spec during implementation.
  • Do not guess which spec is active when multiple candidates exist.
  • Do not store execution state in the spec.
  • Use the spec as the single source of truth.

平台分布

Codex

35.15%
按下载量换算75

Claude

27.02%
按下载量换算57

Cursor

18.67%
按下载量换算40

Gemini CLI

9.85%
按下载量换算21

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

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。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills