Token导航 LogoToken导航

Agent Skill · 研究检索

spec(规格说明)

spec 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 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,权限为操作浏览器;许可证信息还不明确,第三方扫描有提醒项,按安装量在研究检索类的 73,675 个 Skill 里排第 52,216。

总安装 563周安装 23GitHub Stars 8下载 180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(规格说明)
来源仓库:https://github.com/phrazzld/claude-config
仓库路径:skills/spec
安装命令:
npx skills add https://github.com/phrazzld/claude-config --skill spec
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/phrazzld/claude-config --skill spec

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 用于查找和筛选相关信息。适合在 Codex、Claude、Cursor、Gemini CLI 中根据线索定位结果。通过 GitHub 安装,建议结合来源仓库和 README 核验用法。需确认是否会触发联网或命令执行后再使用。spec 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/spec

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/phrazzld/claude-config --skill spec;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

    它需要操作浏览器,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 怎么安装?

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

spec 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 需要什么权限?

权限级别是操作浏览器。该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

spec 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 用的人多吗?

公开数据是累计安装 563 次,近一周安装 23 次,下载 180 次,GitHub 8 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 安全吗?

第三方扫描有提醒项,说明存在需要人工确认的行为。安装前请打开源码看清它读了什么、联网到哪里。

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。

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。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 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

规格说明 · 本页

5638操作浏览器未明确有提醒

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

"People don't want a quarter-inch drill. They want a quarter-inch hole." — Levitt

Role

Product lead clarifying WHAT to build and WHY. Technical HOW comes in /architect.

Dual Mode

Exploration mode — User invokes directly. Full interactive discovery. Quick mode — Autopilot invokes on well-defined issue. Streamlined interview, then spec.

Detection: Well-defined issue with clear problem statement + context = quick mode. Raw idea, vague issue, no issue, or user explicitly exploring = exploration mode.


Exploration Mode

Phase 1: Understand

Accept input: raw idea (string), issue ID, or observation.

  1. If no issue exists: create skeleton immediately (gh issue create)
  2. If issue exists: gh issue view $1 --comments
  3. Load project.md if present (falls back to vision.md)
  4. Read relevant codebase context (adjacent features, existing patterns)

Present: "Here's what I understand about this idea. Let me investigate the problem space."

Phase 2: Investigate

Launch agents in parallel:

AgentFocus
Problem explorerWhat's the real problem? Who has it? How painful?
Web research agentHow do others solve this? Current best practices, docs (Gemini)
Codebase explorerExisting patterns, affected files, reusable utilities
User impact analystWho's affected, how much, cost of inaction

Synthesize findings. Present: "Here's what we know about this problem space."

Use AskUserQuestion to validate understanding:

  • "Is this the right problem framing?"
  • "Who's the primary user here?"
  • "What does success look like?"

Phase 3: Brainstorm Product Approaches

For raw or vague ideas, open with /brainstorming before entering the spec loop. It surfaces design options via dialogue with user approval gates before writing anything.

Generate 3-5 approaches from a product perspective. For each:

  • What the user experiences — Concrete interaction flow
  • What value it delivers — Why someone would care
  • Scope — Small / Medium / Large
  • What it enables downstream — Future possibilities

Recommend one. Present all with clear reasoning.

Phase 4: Discussion Loop

Iterate with the user. This loop continues until approach is locked:

  1. Present approaches with recommendation
  2. User questions, challenges, steers ("What about X?", "I prefer Y because Z")
  3. Agents dig deeper on areas of interest
  4. Refine approaches based on discussion
  5. User locks direction — or explores more

Use AskUserQuestion for structured decisions. Plain conversation for open-ended exploration.

No limit on rounds. The spec isn't ready until the user says it is.

Phase 5: Codify

Write product spec on the issue:

## Product Spec

### Problem
[Validated problem — 2-3 sentences]

### Users
**Primary**: [Role] — [context, pain, goal]

### Recommended Approach
[Chosen direction and why]

### Acceptance Criteria
- [ ] Given [precondition], when [action], then [outcome]
- [ ] Given [precondition], when [action], then [outcome]

### Affected Files
- `path/file.ts` — [what changes and why]

### User Stories
- As [persona], I want [action] so that [value]

### Success Metrics
| Metric | Target | How Measured |

### Non-Goals
- [What we're NOT building]

### Boundaries
- Do NOT modify [X]
- Do NOT change [Y]

### Verification

[commands to verify the implementation works]


### Open Questions for Architect

[Technical unknowns surfaced during exploration]

## Flow

flowchart LR A([User opens X]) --> B[Does Y] B --> C{Decision?} C -- Yes --> D[Path A] C -- No --> E[Path B] D & E --> F([Goal achieved])

Diagram selection:

  • Feature with user interaction → flowchart LR (user journey)
  • Feature with branching logic → flowchart TD (decision tree)
  • Integration/API feature → sequenceDiagram

Load ~/.claude/skills/visualize/references/github-mermaid-patterns.md for annotated examples and GitHub gotchas.

If scope is large: yield multiple atomic issues linked via epic.

Post spec as comment: gh issue comment $1 --body "..."

Update labels:

gh issue edit $1 --remove-label "status/needs-spec" --add-label "status/needs-design"

Stress-test with /critique $1 to find gaps.


Quick Mode (Autopilot)

Triggered when autopilot calls /spec on a well-defined issue.

  1. Read issue + comments: gh issue view $1 --comments
  2. Brief interview (1-2 AskUserQuestion max) — only if ambiguities exist
  3. Draft spec with Codex: codex exec "DRAFT product spec for [issue]. Problem: [X]. Users: [Y]. Include user stories, success metrics, non-goals." \ --output-last-message /tmp/codex-spec.md 2>/dev/null
  4. Validate with Thinktank: thinktank /tmp/spec-review.md./README.md --synthesis
  5. Post spec, update labels to status/needs-design

Completion

Exploration mode: "Product spec locked — includes flow diagram. Ready for /architect $1 or /shaping $1." Quick mode: "Product spec complete — includes flow diagram. Next: /architect $1"

Note: For interactive exploration with formal Shape Up methodology (Rs/Ss notation, fit checks), prefer /shaping. /spec is the product-only primitive used by autopilot's quick mode.

Visual Deliverable

After completing the core workflow, generate a visual HTML summary:

  1. Read ~/.claude/skills/visualize/prompts/spec-overview.md
  2. Read the template(s) referenced in the prompt
  3. Read ~/.claude/skills/visualize/references/css-patterns.md
  4. Generate self-contained HTML capturing this session's output
  5. Write to ~/.agent/diagrams/spec-{feature}-{date}.html
  6. Open in browser: open ~/.agent/diagrams/spec-{feature}-{date}.html
  7. Tell the user the file path

Skip visual output if:

  • The session was trivial (single finding, quick fix)
  • The user explicitly opts out (--no-visual)
  • No browser available (SSH session)

平台分布

Codex

37.4%
按下载量换算67

Claude

26.03%
按下载量换算47

Cursor

17.69%
按下载量换算32

Gemini CLI

9.18%
按下载量换算17

安全审计

Gen Agent Trust Hub

可疑

Socket

可疑

Snyk

可疑

权限和风险

操作浏览器

该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

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。来源安全扫描存在提醒或未通过项,安装前请自行复核。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