Token导航 LogoToken导航

Agent Skill · 研究检索

develop(develop 演示文稿)

develop 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。

develop 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,第三方扫描有提醒项,按安装量在研究检索类的 73,675 个 Skill 里排第 61,604。

总安装 349周安装 14GitHub Stars 142下载 113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:develop(develop 演示文稿)
来源仓库:https://github.com/thebushidocollective/han
仓库路径:skills/develop
安装命令:
npx skills add https://github.com/thebushidocollective/han --skill develop
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/thebushidocollective/han --skill develop

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。

develop 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。通过 npx skills add 命令从指定 GitHub 路径安装并使用。安装前需确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

建议结合来源仓库和原始 README 核验具体用法和功能边界。

来源仓库内路径:skills/develop

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/thebushidocollective/han --skill develop;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

develop 怎么安装?

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

develop 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。

develop 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。

develop 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。

develop 用的人多吗?

公开数据是累计安装 349 次,近一周安装 14 次,下载 113 次,GitHub 142 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。

develop 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
develop

develop 演示文稿 · 本页

349142需要联网未明确有提醒

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

Feature Development Workflow

Name

han-core:develop - Comprehensive 8-phase workflow for developing new features with quality enforcement

Synopsis

/develop [arguments]

Description

Comprehensive 8-phase workflow for developing new features with quality enforcement

Implementation

A comprehensive, structured 8-phase workflow for developing new features with quality enforcement and quality principles.

Overview

This command guides you through a systematic feature development process:

  1. Discover - Understand requirements and context
  2. Explore - Analyze existing codebase patterns
  3. Clarify - Resolve ambiguities with user input
  4. Design - Create architecture with specialized agents
  5. Implement - Build with TDD and quality practices
  6. Review - Multi-agent quality review with confidence scoring
  7. Validate - Run all verification hooks and summarize
  8. Document - Write blog post announcing the feature

Phase 1: Discover

Understand requirements and gather context

Objective: Establish clear understanding of what needs to be built and why.

  1. Review the feature request:

- What is the user-facing goal? - What problem does this solve? - What are the acceptance criteria?

  1. Identify impacted areas:

- Which parts of the codebase will change? - What existing features might be affected? - Are there related issues or PRs?

  1. Check for similar features: # Search for similar implementations grep -r "similar_feature_name".
  2. Review project documentation:

- Check CLAUDE.md, CONTRIBUTING.md for standards - Review architecture docs if available - Identify any constraints or requirements

Output: Clear problem statement and high-level approach.


Phase 2: Explore (Parallel Agent Execution)

Analyze codebase with specialized agents

Objective: Understand existing patterns and identify integration points.

Launch multiple Explore agents in PARALLEL (single message with multiple Task calls):

  1. Code Explorer: Map existing features

- Find entry points and call chains - Identify data flow and transformations - Document current architecture

  1. Pattern Analyzer: Identify conventions

- How are similar features implemented? - What testing patterns are used? - What naming conventions exist?

  1. Dependency Mapper: Understand relationships

- What modules will be affected? - What are the integration points? - Are there circular dependencies to avoid?

Consolidation: Synthesize findings from all agents into a cohesive understanding.

Output: Comprehensive map of existing codebase patterns and integration points.


Phase 3: Clarify (Human Decision Point)

Resolve ambiguities before implementation

Objective: Get user input on unclear requirements and design choices.

Use AskUserQuestion tool to resolve:

  1. Architecture decisions:

- Which approach should we take? (if multiple valid options) - What are the trade-offs? (performance vs. simplicity)

  1. Scope clarifications:

- Should this include X feature? - What's the priority if time is limited?

  1. Integration choices:

- Should we extend existing module or create new one? - How should this integrate with system Y?

IMPORTANT: Do not proceed with assumptions. Get explicit user answers.

Output: Clear, unambiguous requirements with user-approved approach.


Phase 4: Design (Parallel Agent Execution)

Create architecture with specialized agents

Objective: Design the implementation before coding.

Select appropriate specialized agent(s) based on feature type:

  • Frontend feature? → frontend:presentation-engineer
  • Backend API? → backend:api-designer
  • Database changes? → databases:database-designer
  • Complex system? → architecture:solution-architect

Launch agents in PARALLEL for multi-disciplinary features:

  • Frontend + Backend agents simultaneously
  • Include security:security-engineer for sensitive features
  • Include performance:performance-engineer for high-traffic features

Agent responsibilities:

  • Define module structure and file organization
  • Specify interfaces and contracts
  • Identify testing strategy
  • Document key decisions and trade-offs

Consolidation: Review all design proposals, resolve conflicts, select final approach.

Output: Detailed implementation plan with module structure and interfaces.


Phase 5: Implement (TDD with Quality Enforcement)

Build the feature using test-driven development

Objective: Implement the designed solution with quality practices.

Apply TDD cycle (use tdd:test-driven-development skill):

For each component:
1. Write failing test (Red)
2. Implement minimum code to pass (Green)
3. Refactor for quality (Refactor)
4. Repeat

Implementation guidelines:

  • ✅ Start with tests, not implementation
  • ✅ Follow existing codebase patterns (from Phase 2)
  • ✅ Apply SOLID principles (han-core:solid-principles skill)
  • ✅ Keep it simple (KISS, YAGNI)
  • ✅ Apply Boy Scout Rule - leave code better than found
  • ❌ Don't over-engineer
  • ❌ Don't skip tests
  • ❌ Don't ignore linter/type errors

Integration:

  • Integrate incrementally (don't build everything then integrate)
  • Test integration points early
  • Validate against acceptance criteria continuously

Output: Working implementation with comprehensive tests.


Phase 6: Review (Parallel Multi-Agent Review)

Quality review with confidence-based filtering

Objective: Identify high-confidence issues before final validation.

Launch review agents in PARALLEL (single message with multiple Task calls):

  1. Code Reviewer (han-core:code-reviewer skill):

- General quality assessment - Confidence scoring ≥80% - False positive filtering

  1. Security Engineer (security:security-engineer):

- Security vulnerability scan - Auth/authz pattern verification - Input validation review

  1. Discipline-Specific Agent:

- Frontend: frontend:presentation-engineer (accessibility, UX) - Backend: backend:backend-architect (API design, scalability) - etc.

Review consolidation:

  • Merge findings from all agents
  • De-duplicate issues
  • Filter for confidence ≥80%
  • Organize by: Critical (≥90%) → Important (≥80%)

Present findings to user with options:

Found 3 critical and 5 important issues.

Options:
1. Fix all issues now (recommended)
2. Fix critical only, defer important
3. Review findings and decide per-issue

Output: Consolidated review with high-confidence issues only.


Phase 7: Validate & Summarize

Final verification and change summary

Objective: Ensure all quality gates pass and document the change.

Run all validation hooks:

# All validation plugins automatically run on Stop
# Verify: tests, linting, type checking, etc.

Validation checklist:

  • All tests pass
  • Linting passes
  • Type checking passes
  • No security vulnerabilities introduced
  • Documentation updated
  • No breaking changes (or properly coordinated)

Generate change summary:

  1. What changed: Files modified and why
  2. How to test: Steps to verify functionality
  3. Breaking changes: None, or list with migration guide
  4. Follow-up tasks: Any deferred work or tech debt

Create TODO list (using TaskCreate tool):

  • Document any follow-up tasks
  • Track deferred improvements
  • Note any tech debt introduced

Output: Ready-to-commit feature with comprehensive documentation.


Phase 8: Document (Blog Post)

Write a blog post announcing the feature

Objective: Share the new feature with the community and explain its value.

Blog post creation:

  1. Research context (optional):

- Use reddit to find related community discussions - Identify pain points the feature addresses - Understand how users talk about this problem

  1. Write the blog post: Location: website/content/blog/{feature-slug}.md --- title: "{Feature Name}: {Compelling subtitle}" description: "{One-line description of what problem this solves}" date: "{YYYY-MM-DD}" author: "The Bushido Collective" tags: ["{relevant}", "{tags}"] category: "Feature" --- {Opening hook - what problem does this solve?} ## The Problem {Describe the pain point this feature addresses} ## The Solution {Explain how the feature works} ### Key Capabilities {List main features with examples} ## Getting Started {How to use the feature} ## What's Next {Future improvements or related features}
  2. Writing guidelines:

- Technical but accessible - 500-1000 words for feature announcements - Include working code examples - Be honest about limitations - Make it actionable

IMPORTANT: Every significant feature should have a blog post. This is not optional.

Output: Published blog post in website/content/blog/.


Usage

Basic usage

/feature-dev

Then describe the feature you want to build.

With feature description

/feature-dev Add user authentication with JWT tokens

Best Practices

DO

  • ✅ Follow all 8 phases in order
  • ✅ Launch agents in parallel when independent
  • ✅ Use AskUserQuestion to resolve ambiguities
  • ✅ Apply confidence scoring to all reviews
  • ✅ Run TDD cycle for all new code
  • ✅ Pause for user input at decision points
  • ✅ Write a blog post for every significant feature

DON'T

  • ❌ Skip phases (especially Explore, Review, and Document)
  • ❌ Start coding before design (Phases 1-4)
  • ❌ Implement without tests
  • ❌ Report low-confidence review findings
  • ❌ Make architectural decisions without user input
  • ❌ Commit without running validation hooks
  • ❌ Ship features without documentation

Example Workflow

User: /feature-dev Add pagination to user list API

Phase 1: Discover
- Feature: Add pagination to GET /api/users
- Acceptance: Support page/limit query params, return total count
- Impact: Backend API, database queries

Phase 2: Explore (parallel agents)
- Found existing pagination in products API
- Pattern: Uses offset/limit with total count in response
- Testing: Integration tests verify pagination logic

Phase 3: Clarify
Q: Should we use cursor-based or offset-based pagination?
A: [User selects offset-based for consistency]

Phase 4: Design
- Agent: backend:api-designer
- Design: Extend existing UserService with pagination
- Interface: getUsersPaginated(page, limit) -> { users, total }

Phase 5: Implement
- Write test for pagination
- Implement pagination logic
- Test passes ✅

Phase 6: Review (parallel agents)
- Code reviewer: No issues (confidence N/A)
- Security engineer: No issues (confidence N/A)
- Backend architect: No issues (confidence N/A)

Phase 7: Validate
- Tests: ✅ Pass
- Linting: ✅ Pass
- Types: ✅ Pass
- Ready to commit

Phase 8: Document
- Blog post: website/content/blog/user-list-pagination.md
- Title: "Pagination: Handling Large Data Sets Gracefully"
- Tags: [api, pagination, performance]

Summary: Added pagination to user list API
Files: services/user.service.ts, tests/user.service.test.ts, website/content/blog/user-list-pagination.md
Testing: Run GET /api/users?page=1&limit=10

See Also

  • /review - Run multi-agent review only
  • /commit - Smart commit workflow
  • /create-blog-post - Research and write blog posts
  • tdd:test-driven-development - TDD skill
  • han-core:code-reviewer - Review skill

平台分布

Codex

34.24%
按下载量换算39

Claude

32.16%
按下载量换算36

Cursor

19.58%
按下载量换算22

Gemini CLI

10.34%
按下载量换算12

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

可疑

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。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