메뉴
按规格文档的 Ticket 逐个实现代码，每个 Ticket 独立上下文避免幻觉。内置 todo 进度追踪，支持 LSP 诊断、worktree 隔离、子智能体委派等按需升级。Use when user says "implement ticket", "实现这个任务", "sdd implement", "开始实现", or wants to execute tasks from a spec. Reads .plan/spec.md for task breakdown.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Extremely strict maintainability review focused on structural simplification, abstraction quality, and spaghetti-condition growth. Not a general correctness review -- this is a deep code quality audit that pushes for "code judo" moves that delete complexity. Use when asked to "code review", "maintainability review", "code quality audit", "/code-review", or "/review-quality".
Spawns a verifier subagent to audit your work: traces what was done, reviews the diff, runs builds/tests, and delivers a PASS/FAIL verdict. Auto-retries fixes up to 3 rounds. Use when asked to "check work", "verify changes", "self-verify", "/self-verify", "/check", or "/verify".
轻量需求头脑风暴 — 先扫代码库确认前置条件，再用 2-3 轮聚焦讨论锁定范围和方案。Use when user says "brainstorm this feature", "需求讨论", "sdd brainstorm", or wants to explore a feature idea before writing specs. Replaces heavyweight brainstorming workflows that burn tokens.
将头脑风暴结论转化为结构化规格文档，包含需求、技术设计、任务拆解和决策记录。Use when user says "write spec", "写规格", "sdd spec", or wants to formalize a brainstormed idea into an implementable specification. Reads .plan/brainstorm.md if available.
Comprehensive quality analysis tool for Claude Skills. Evaluates skills from GitHub URLs, marketplace links, ZIP files, or local directories across five dimensions (structure, security, UX, code quality, integration) using balanced scoring and three output modes (comprehensive report, interactive review, pass/fail certification). Triggers on "analyze skill quality", "review this skill", "evaluate skill", or "skill quality check".
Comprehensive security analysis for Claude Code skills with P0-P3 enhancements. Detects malicious code, obfuscation, YAML injection, typosquatting, time bombs, sandbox escapes, and 40+ attack patterns. Use for analyzing skills before installation or auditing existing skills.
| name | sdd-implement |
| description | 按规格文档的 Ticket 逐个实现代码，每个 Ticket 独立上下文避免幻觉。内置 todo 进度追踪，支持 LSP 诊断、worktree 隔离、子智能体委派等按需升级。Use when user says "implement ticket", "实现这个任务", "sdd implement", "开始实现", or wants to execute tasks from a spec. Reads .plan/spec.md for task breakdown. |
| argument-hint | Ticket number or 'all' to implement sequentially |
| disable-model-invocation | true |
按 .plan/spec.md 的任务拆解，逐个 Ticket 实现代码。
实现任何 Ticket 之前，先回答三个问题：
判断顺序：直接调用 > 组合适配 > 参考模式 > 新建。
常见反模式（禁止）：
formatDate 工具函数，Ticket 里再写一个实现要点：每个 Ticket 完成后报告中必须注明「复用情况」——用了什么现有代码、或为什么确认没有可复用的。
.plan/spec.md — 必须存在，否则提示先运行 /sdd-spec读取 .plan/spec.md，解析任务列表，用 todo_write 建立进度追踪：
找到 N 个 Ticket：
1. <任务名> [依赖: 无]
2. <任务名> [依赖: Ticket 1]
...
将所有 Ticket 写入 todo 列表（全部 pending），第一个设为 in_progress。
1.1 上下文准备
1.2 实现
1.3 验证
1.4 对齐检查
1.5 发现记录
| D-N | <发现> | <原因> | <日期> |1.6 更新进度
todo_write：当前 Ticket 标记 completed，下一个设为 in_progress报告：
✅ Ticket N: <任务名> — 完成
变更文件：...
复用情况：<调用了哪些现有代码，或"无复用对象，新建">
验证结果：...
下一个：Ticket N+1
🎉 所有 Ticket 实现完成。
总计：N 个 Ticket，M 个文件变更
建议：运行完整测试套件确认无回归
可选：用 /handoff 保存上下文
每个升级都有明确的触发信号和收益判断——不是"能用就用"，而是"不用会出问题才用"。
触发信号：修改了公共接口、类型复杂的代码、用户反馈类型错误
动作：Ticket 完成后调 lsp diagnostics 检查当前文件
收益：在运行测试前捕获类型错误
不触发：纯脚本、配置文件、文档变更
触发信号：Ticket 涉及破坏性变更（改接口签名、重构核心模块）、或用户要求并行实现多个 Ticket
动作：enter_worktree 创建隔离分支，在 worktree 中实现，完成后展示 diff
收益：失败不影响主分支，可安全回滚
不触发：常规功能开发、小改动
触发信号：Ticket 独立性很强（无上下文依赖）、且实现复杂度高（>100 行、涉及多文件） 动作：派发「后端架构师」或对应专业子智能体，在子会话中独立完成 收益：隔离上下文，主会话不被污染，可并行 不触发：Ticket 依赖前序上下文、简单改动、需要频繁和用户确认的场景
触发信号：测试运行时间 >30 秒、需要同时跑 dev server 和测试
动作：run_shell_command 以 is_background: true 运行长任务
收益：不阻塞主流程
不触发：快速测试（<10s）
触发信号：所有 Ticket 完成后、或用户要求 review 动作：派发「代码审查员」子智能体审查变更 收益：捕获实现阶段遗漏的问题 不触发：用户明确说不需要 review
/sdd-spec