by TsCarpe
AI-native SDLC skills for Claude Code: requirement intake & audit, adversarial review gate, AI test orchestration, decision re-check, release config audit
# Add to your Claude Code skills
git clone https://github.com/TsCarpe/claude-sdlc-skillsGuides for using ai agents skills like claude-sdlc-skills.
See how claude-sdlc-skills compares with popular alternatives.
claude-sdlc-skills is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by TsCarpe. AI-native SDLC skills for Claude Code: requirement intake & audit, adversarial review gate, AI test orchestration, decision re-check, release config audit. It has 50 GitHub stars.
claude-sdlc-skills's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/TsCarpe/claude-sdlc-skills" and add it to your Claude Code skills directory (see the Installation section above).
claude-sdlc-skills is primarily written in Python. It is open-source under TsCarpe on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh claude-sdlc-skills against similar tools.
No comments yet. Be the first to share your thoughts!
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM is an informational directory and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
AI-native SDLC skills for Claude Code (and other coding agents via skills.sh) — AI takes over the repetitive labor of requirement intake, test design, adversarial review, browser testing and release auditing; humans rule only at a few hard gates. Battle-tested end-to-end on a real production Java project. Docs in Chinese; skills trigger on Chinese and English phrases.
把 SDLC 里最耗人的重复劳动——需求梳理、测试用例、评审、浏览器测试、发版审计——交给 AI;红线由脚本机械拦截;人只在关口裁决。不是 prompt 合集,是带关卡、留档、降级路径的完整工作流。
npx skills add TsCarpe/claude-sdlc-skills # 一行安装,先跑 sdlc-intent 零依赖
AI 写代码已经够快,瓶颈移到了需求理解、测试与评审:这些环节要么仍靠人工硬扛,要么让 AI 自由发挥但不可信。本仓库的答案是三条纪律:
来自源项目(某 Java DDD 生产系统)的实测数字:57 条评审 issue 单轮关口产出(12 分歧 + 45 角色问题)、74 条用例走完浏览器执行、回归轮 0 agent token(用例资产化为 Playwright spec 后由 runner 跑)、6 个 skill 对齐官方 best-practices(5 个经多轮复核,sdlc-guardrails 按同标准新增)。
flowchart LR
A[PRD 输入] --> B[sdlc-intent<br/>梳理 + 体检]
B --> C1[技术设计]
B -.互不阅读.-> C2[测试用例<br/>sdlc-test cases]
C1 --> D[sdlc-gate<br/>4 子代理对抗评审]
C2 --> D
D --> E{{人 · 逐条裁决放行}}
E --> F[切片开发<br/>大需求走三段式]
F --> G[sdlc-test<br/>静态比对 → 浏览器执行 → 报告]
G --> H{{人 · 复验缺陷}}
DB[sdlc-doubt<br/>决策对抗复查] -. 任意阶段旁路 .-> F
CR[sdlc-config-review<br/>发版配置扫描] -. 发版前旁路 .-> G
六边形 = 人工关口,只有两处;旁路机制随时可插。带真实产物片段的 6 步走查见 example-walkthrough。
| skill | 阶段 | 触发示例 | 产物 | 依赖 |
|---|---|---|---|---|
| sdlc-intent | 需求进来时 | 「帮我梳理这个需求文档,完成后继续体检」 | intake/ 三件套:digest(结构化共识)/ audit(六层缺陷体检)/ pm-checklist |
无必需 |
| sdlc-test | 用例与测试期 | /sdlc-test cases|static|exec|spec|report <需求名> |
cases.md(用例+缺陷权威文件)+ 各轮 reports;通过用例可资产化为 Playwright spec | chrome-devtools / mysql / codegraph MCP(均带降级) |
| sdlc-gate | 设计+用例定稿后 | /sdlc-gate <需求名> |
issues 宽表(原文摘引+裁决列),放行后解锁开发与测试执行 | 无硬依赖 |
| sdlc-doubt | 旁路 · 任意阶段 | 「这个判断我不放心,帮我质疑一下」 | 会话内五步闭环(CLAIM→EXTRACT→DOUBT→RECONCILE→STOP) | 无 |
| sdlc-config-review | 旁路 · 发版前 | 「梳理上线配置清单」 | 配置 Key 清单 + xxl-job 任务/MQ 订阅平台操作清单 + 知会项 | git 即可 |
| sdlc-guardrails | 横切 · 写入瞬间 | 「给这个项目接红线拦截」 | hook 拦截(guardrails.yaml 四类规则)+ pre-commit 兜底 + --check 基线报告 |
python3 |
| sdlc-sync | 横切 · 产物出口/入口 | 「把三件套推上平台 / 拉一下 PM 的答复 / 用例上平台」 | 平台需求卡 + 澄清答复闭环 + 标准需求文档 + 用例登记表(平台本体:独立仓 sdlc-platform) | 平台服务运行中 + curl |
关键机制:关卡互认(sdlc-gate 放行视同 sdlc-test 关卡1 通过)、用例独立性红线(cases 禁止读设计——交叉审查的价值前提)、降级不炸(任一 MCP 缺失都有声明过的降级路径,见 faq)。
方式一:skills CLI(推荐)
npx skills add TsCarpe/claude-sdlc-skills # 装到当前项目 .claude/skills/
npx skills add TsCarpe/claude-sdlc-skills -g # 装到全局 ~/.claude/skills/
npx skills update <skill名> -g # 更新单个;跨文件迁移版本可能报 Failed to update
update 报 Failed to update? 常见于版本间有文件移动/删除时(如 v0.5.0 → v0.5.1 的 runner 模板迁域)——remove 后重新 add 即可恢复,步骤见 faq。
方式二:Claude Code 插件市场
/plugin marketplace add TsCarpe/claude-sdlc-skills
/plugin install claude-sdlc-skills@claude-sdlc-skills
装完先试这个(零外部依赖):对一个真实需求文档(飞书链接或本地 markdown)说——
帮我梳理这个需求文档 <链接或路径>,完成后继续体检
装完只是开始:sdlc-guardrails 红线拦截、回归 runner、测试环境文件在项目侧怎么搭,见 docs/project-setup.md。
不必一步到位,按依赖重量逐级上(论证见 workflow.md §7.3):
功能点 ≥5 / 涉及表结构变更 / 跨业务域的大需求不整块开发:第一段业务全貌(角色×状态×功能点主线串联)→ 第二段技术骨架(接口契约分级 / D 表八类别 / 公共资产六类)→ 第三段骨架 child 先行、逐功能点切片——三次小确认替代一次看不完的大确认。
全部文档带导读地图与三条阅读路径:
sdlc-platform(Spring Boot + Vue + SQLite),skill 源在本仓 skills/sdlc-sync/站在这些来源的肩膀上:addyosmani/agent-skills(纪律强制)、Claude Agent Skills 官方最佳实践(skill 工程规范)、AI-Native SDLC Playbook(工件链思想)。
MIT · 版本兼容:已在 Claude Code + skills CLI(2026-09 版)验证;skill 遵循 Agent Skills 开放标准,其他兼容 agent(Cursor 等)经 skills.sh 亦可安装。更新日志:CHANGELOG.md(最新 v0.4.0)
感谢 LINUX DO 社区的讨论与反馈:https://linux.do