Development

在开发方案已经讨论清楚、需要固化保存时使用。也适用于用户明确要求保存方案、编写规格文档，或希望在开发前明确目标、边界、约束和验收标准的场景。将已确认的方案整理成可供后续任务拆解、开发和验证共同使用的规格文档。

x-spec3 的任务拆解 skill。读取 `docs/spec/{spec-name}/spec.md` 的目标、边界与不变量、判断依据、验收清单和直接 GWT Scenarios，生成 `docs/spec/{spec-name}/tasks/{task-name}/dev-checklist.md`，并以 Scenario ID 精确回指实现范围。用户提到 x-req3、要求把 spec3 拆成开发任务，或目标规格含 `spec_version: 3` 时使用。spec2 包继续使用 x-req2。

verify 通过后的质量审查。risk Q2 由一个 reviewer 一次完成 q1-intent、q2-correctness、q3-evidence；risk Q3 依次使用三个独立 reviewer：q1-intent 检查用户意图、需求、公开契约与实现是否对齐，q2-correctness 检查边界、失败路径、状态与并发是否正确，q3-evidence 检查测试和 verify 证据是否真实有效。发现 P0/P1 后登记 issue 并交 x-fix 批量修复，修复后增量复审。

Bug 修复执行 skill。分三种入口： 1. 用户直接报告 Bug → 定位根因 → 修复 → 产出 fix-report-*.md 或 fix-note-*.md（无需 CR 报告） 2. 有 x-cr 的 CR 报告 → 按报告逐条修复 → 回写同一份 `reports/cr/cr-report-*.md` 主档并产出修复记录 3. 有 x-verify / x-qa-gate fail 报告 → 按 issue 清单一次批量修复，产出逐条处置表，交回 gate 增量复审 触发方式："x-fix"、"修一下这个 bug"、"这个功能坏了"、 "按 CR 报告修复"、"把 CR 问题修了"。

开发任务执行 skill。读取单个 task 的 dev-checklist，按 `spec:` 指针跟读归属 spec 包，按依赖实现、写 verify 证据，并由 checklist 头部 risk 驱动交付或 Gate ②。 触发：`x-dev <功能名称>` 或现有 task 目录。

软件正确性调查 skill。用于用户说“XX 不太对”“这个功能有 bug”“结果和预期不一致”“帮我查原因”，也用于用户要求 review 某个模块、文件、diff 或 PR 的正确性。遇到已知异常或模块 correctness review 时必须优先使用本 skill。 本 skill 使用贝叶斯根因调查：先列候选原因 H，再按日志、代码路径、测试、diff、spec 等证据 E 更新置信度，最后判断根因属于原始 spec 不一致、实现过程偏移、spec 缺口、环境/数据问题或证据不足。 x-cr 独立于 x-verify / x-qa-gate 自动门禁，产出 `reports/cr/cr-report-*.md`，x-fix 可按该 CR 报告继续修复。

系统方案规划 skill。产出 docs/spec/<spec-name>/ 下的独立需求包（目标与 DoD + 组件设计 + 接口/数据结构 + 核心流程时序 + 验证策略 + task 映射）。 当用户给的是模糊想法而非具体功能需求时使用。触发场景： "这个东西怎么设计"、"帮我梳理整体方案"、"先别写代码把方案理清"、 "这个系统该怎么拆"、"x-spec"、架构级改造、 一个系统切面会派生多个 task 需要先建 spec 需求包。上下文不足时先进入头脑风暴模式，收敛后再保存 spec 文档。 入口选择：小功能小修复由 x-req 定级 Q0/Q1；已有明确需求走 x-req。

Spec 驱动 task 拆解 skill：只服务已有 docs/spec/<spec-name>/ 归属的 task，读 spec.md 与 modules.md 直接拆 dev-checklist.md（新头部 spec:/risk:，新表头 任务说明/Requirement/风险），不产出 README，不复述需求/模块/验收内容，risk 依托 spec 包既有信号定级（不自造判据）。x-req2 是 openspec/changes/xreq-spec-driven 变更的过渡期命名，该变更单轨替换旧版 x-req（旧结构已删除、不再维护），实现完成后会改名替换 skills/x-req。 调用时机（命中任一即优先使用）： - 用户提到 x-req2。 - 用户给出的 task 明确归属某个已建好的 docs/spec/<spec-name>/ 包（该 spec 所属模块状态已是"可进入 x-req"）。 - 已跑过 x-spec2 建模，需要把稳定 spec 拆成可执行 task。 不要用：历史 `dev-pipeline/tasks/` task 的更新（旧结构...

已废弃：保留旧版 x-req 的任务包规划规则与模板，供历史 task 和迁移核对使用。新需求使用 x-req2。

Turn a vague performance requirement into arithmetic before the architecture is chosen, size concurrency with Little's Law, and label every input as measured, derived, or assumed. Use when a design calls itself "scalable" or "fast" without a number, when choosing between one machine and a distributed system, when sizing a thread pool, connection pool, queue, or instance count, or when someone asks how much traffic, storage, or bandwidth a system will need.

Use when a new Godot system or feature has open design decisions — interrogates them in batched rounds, scope first, each question with a recommended answer, and records the answers before any design or code. Triggers on "grill me", "ask me first", "question me on the design", "what do I need to decide before building". Not for choosing between nodes or APIs, and not for bug fixes.

Vitest fast unit testing framework powered by Vite with Jest-compatible API. Use when writing tests, mocking, configuring coverage, or working with test filtering and fixtures.

Record captioned product walkthrough videos of any web app with Playwright + ffmpeg — proof videos for stakeholders, feature demos, bug-fix evidence

Keyword & SERP research via the DataForSEO API — search volume, competition, CPC, keyword ideas for naming/SEO decisions

Offload large / multi-file reads to a cheap worker model so raw files never enter Claude's context (token savings)

Work the C++/Blueprint continuum with Nanite, Lumen and the Gameplay Ability System. Use when developing Unreal Engine gameplay in C++, Blueprints, or GAS.

Set up Git LFS for binary assets, handle unmergeable scene and prefab conflicts, and know when Perforce beats Git. Use when configuring Perforce, Git LFS, or binary asset branching.

Build Blender Python add-ons, asset validators, exporters and pipeline automation. Use when writing Python scripts, add-ons, or batch tools for Blender.

Use when the user makes or revisits a meaningful product, technical, or organizational decision and wants it captured ADR-style. Triggers: "we decided to…", "let's go with X over Y", "write up the call we made on…", or reconstructing a past decision. Especially AI-product decisions — model choice, eval threshold, fallback design — whose rationale is easy to lose.

Use when the user wants to analyze whether non-promotable (glue) work is fairly distributed across their team, with attention to demographic patterns. Triggers: "is our glue work fair", "who does the non-promotable work on our team", "audit workload distribution", "are women doing more glue work", "workload equity analysis", "non-promotable work allocation". Produces a diagnosis with bias signals and redistribution recommendations

Use when a team must revisit their Goals/Roles/Norms foundation to surface drift and realign. Triggers include "team check-in", "are we still aligned", "team retro on our operating model", "review our team charter", "3X3 check-in". Structured conversation (30–90 min): still committed to our goals/roles/norms — if not, what's the gap? Second step of Moussa's 3X3. Works from a charter or de novo.

Use when a team lacks psychological safety — members don't speak up, withhold concerns, or fear making mistakes. Triggers include "build psychological safety", "team is afraid to speak up", "create safe environment", "people are withholding their opinions", "no one challenges decisions", "fear-based culture". Prescriptive practices from Amy Edmondson's research — not just diagnostic like `team-diagnosis`, but what to do this week.

Use when designing or revisiting a 12–18 month growth plan for a direct report. Trigger phrases include "build a growth plan for [report]", "what should [name]'s next year look like", "I want to architect [report]'s path to senior", "I'm prepping for promo cycle and [report] needs a plan", "let's think about [report]'s development". Distinct from one-on-one-prep and coaching-mode.

Use when preparing for a 1:1 — manager with a report, IC with their manager, or peer sync. Trigger phrases include "I have a 1:1 with X tomorrow", "help me prep for my 1:1", "what should I bring up in my next 1:1?". Builds a relationship-matched agenda; if the other person is a registered stakeholder, reads their reflection file first.