xyz-pi-extensions

完成 worktree 的完整合并流程：本地验证 → PR CI → merge → post-merge CI → 发布准备 → Release Notes → 创建 Release → 清理。 使用 git merge --no-ff 保留完整分支历史。 支持项目级钩子（.bare/custom-hooks/）实现个性化发布流程。 触发词："合并worktree"、"merge-worktree"、"合并PR"、"发布"、"release"、"上线"。

Reference for writing workflow JS scripts. Auto-loaded when using workflow-generate or writing/editing workflow scripts for Pi. Covers runtime environment, injected globals, constraints, and script patterns. Not for general coding or subagent usage.

Use when the user says "轻量执行", "lite execute", "按 Wave 执行", "goal 模式执行 plan", "执行 plan", or has a completed plan.md (from lite-plan) or execution-plan.md (from mid-detail-plan) and needs to execute the Waves. ADR-029 后阶段 A+B 由 workflow run execute-full-workflow 机器接管（worktree-setup → dev waves → test+review → cleanup），主 agent 不再直接派 subagent。 对应 CW action: dev (workflow 内 implementer 渐进式提交) + test (workflow 内 test-runner 渐进式提交). Not for planning (lite-plan / mid-detail-plan). Not for retrospect (coding-retrospect).

Use when the user says "issue拆分", "问题分解", "issue decomposition", "方案对比", "tradeoff analysis", "优先级排序", or has finished system-architecture.md and needs to break it into prioritized issues with solution comparisons. Produces issues.md as a fog-of-war decision map. Step 3 of 6. Not for business requirements (Step 1) or architecture modeling (Step 2). Not for non-functional risk analysis (Step 4) or code-level design (Step 5).

Use when the user says "mid 详细计划", "中等功能实施设计", "issues+nfr+架构+计划一起做", "批量设计实施", or has finished mid-plan and needs issues.md + non-functional-design.md + code-architecture.md + execution-plan.md (+ code-skeleton) + detail.json (CW detail action 入参). 纯设计 skill，不写实现代码。对应 CW action: detail. Not for L3 heavy (use full-*). Not for requirements/architecture (that is mid-plan).

[internal] Shared reference files for the lite workflow (lite-plan / coding-execute / coding-retrospect). Not invoked directly — sibling lite-* skills resolve paths via ../lite-shared/references/{file}.md. Kept hidden from model invocation.

[internal] Shared reference files for the mid workflow (mid-plan / mid-detail-plan). Not invoked directly — sibling mid-* skills resolve paths via ../mid-shared/references/{file}.md. Kept hidden from model invocation.

Use when the user says "编码流程", "coding workflow", "开始编码", "走 CW", "建 topic", "create topic", "开发功能", or wants to start/advance a structured coding task through the coding-workflow orchestrator (CW) state machine. 唯一入口：先调 coding-workflow tool 的 action=create 建 topic（锁 tier），之后按 CW 返回的 nextAction 选下一个 action。 Not for pure planning without CW (lite-plan / mid-plan). Not for design-only (design-status).

Use when the user says "轻量计划", "lite plan", "小功能计划", "写测试计划", "Wave 拆分", "plan this feature", or is in plan mode brainstorming a small feature (no architecture change). Produces plan.md + plan.json (CW plan action 入参). 对应 CW action: plan. Not for architecture-level changes (use full-* workflow). Not for execution (that is coding-execute).

Use when the user says "设计收尾", "design closeout", "沉淀设计", "归档主题", "归档 topic", "archive topic", or after coding-execute + 测试验收全绿，需要把稳定结论 沉淀进长期文档 (ARCHITECTURE/PRODUCT/NFR/ADR/TEST-STRATEGY) 并归档 .xyz-harness/{topic}. 设计→实施→沉淀闭环的最后一步。对应 CW action: closeout (status 流转到 closed 终态). Not for full-* phases themselves, not for 回顾执行质量 (那是 harness-retrospect), not for 需求澄清或编码实现。

Use when the user says "初始化项目", "init project", "准备项目文档", "新建 AGENTS.md", "项目基建", "project bootstrap", "start full workflow", or when the project lacks AGENTS.md/CONTEXT.md/ARCHITECTURE.md and needs documentation bootstrapped before full-* design work. Not for business requirements (Step 1), architecture modeling (Step 2), or coding. Not for overwriting existing docs — only scans, reports, and fills gaps with minimal skeletons on user confirmation.

Use when the user says "轻量复盘", "lite retrospect", "做个复盘", "复盘一下", or after a coding-execute run has completed (goal completed) and wants a quick self-check retrospect on the dev/test process. Produces a lightweight checklist report (changes/retrospect.md, CW retrospect action 入参). 对应 CW action: retrospect. Not for design closeout (that is coding-closeout). Not for planning or execution.

Use when the user says "代码架构", "code architecture", "详细设计", "接口契约", "时序图", "工程目录", "API 设计", or has finished non-functional-design.md and needs concrete code-level architecture. Produces code-architecture.md + code-skeleton/ (可编译骨架代码). Step 5 of 6. Not for system-level architecture (Step 2) or issue decomposition (Step 3). This designs the code structure and contracts AND validates them via a compilable skeleton (Step 7) — it does not write the implementation bodies, which belong to the execution/coding phase.

Use when the user says "执行计划", "execution plan", "wave编排", "任务拆分", "交付计划", "sprint 规划", or has finished code-architecture.md and needs to split the work into executable waves. Produces execution-plan.md. Step 6 of 6. Not for code architecture design (Step 5). Not for writing code itself — this plans the execution sequence, it does not execute it.

Use when the user says "mid 计划", "中等需求设计", "需求架构一起做", "批量设计需求架构", or has a L2-complexity feature (multi-module single system, 3-5 Waves, 2-3 NFR dimensions). Produces requirements.md + system-architecture.md (+ .html) + clarify.json (CW clarify action 入参). 对应 CW action: clarify (skill 名 mid-plan 对应 CW 的 clarify 状态). Not for L3 heavy (use full-*). Not for L1 small feature (use lite-plan). Not for issues/nfr/code-arch/execution (that is mid-detail-plan).

Use when the user says "系统设计", "架构设计", "architecture design", "领域建模", "domain modeling", "模块划分", "状态机", or has finished requirements.md and needs to model the system. Produces system-architecture.md. Step 2 of 6. Not for business requirements/use-case work (Step 1), issue decomposition with priorities (Step 3), or code-level API/sequence design (Step 5). Not for writing code.

Use when the user says "澄清需求", "clarify requirements", "需求分析", "业务建模", "use case", "start design", or is at the beginning of the full workflow and needs to define what to build before how to build it. Produces requirements.md. Step 1 of 6. Not for system/architecture design, API definition, tech-stack selection, database schema, or coding — those belong to Steps 2-6. Not for pure technical refactors with no business change.

Use when the user says "非功能性设计", "NFR", "副作用分析", "风险评估", "non-functional design", "安全性分析", "性能分析", or has finished issues.md and needs to analyze side effects of the chosen solutions. Produces non-functional-design.md. Step 4 of 6. Not for issue decomposition/priorities (Step 3) or code architecture (Step 5). Not for writing code. Not for defining new features — only analyzing existing solutions' side effects.

[internal] Shared reference files for the full-* workflow phases. Not invoked directly — sibling full-* skills resolve paths via ../full-shared/references/{file}.md. Kept hidden from model invocation.

Use when a full-* phase Step 5b needs a finalized .md rendered to a self-contained .html, or when the user says "可视化", "渲染", "画图", "render html", "visualize". Produces {deliverable-name}.html with the phase's mandated hero diagram. Serves full-clarity / full-architecture / full-issues / full-nfr / full-code-arch / full-execution-plan. Delegates complex architecture to drawio (built-in autolayout scripts) when available; Mermaid for simple diagrams; hand-built HTML/CSS for rich-card and table-heavy deliverables. Not for design decisions — only visualizes finalized content. Not for writing code.

Use when users want to merge multiple development branches into one test branch for integration testing, or mention lightmerge, lm, lightmerge-branch, 合并测试, or need to quickly deploy and verify multiple features together. Also use when users want to add or remove branches from a test merge configuration, or manage per-project branch merging setups.

清理 git worktree。默认检查分支是否已合并到 main，确认安全后才删除。 支持参数：--force（跳过合并检查，强制清理）、--skip-sync（跳过同步其他 worktree）。 当用户说\"删除worktree\"、\"remove worktree\"、\"清理worktree\"、\"清理分支\"、\"删除分支\"、 \"remove-worktree\"、\"清理工作区\"时使用此 skill。即使用户只是说\"这个分支不要了\"或 \"帮我清理一下\"，也应考虑触发此 skill。

合并分支并发布。触发词："合并"、"merge"、"发布"、"release"、 "上线"。仅用于 xyz-pi-extensions 项目。

审查代码变更。触发词："review"、"审查代码"、"code review"、 "帮我看看代码"。仅用于 xyz-pi-extensions 项目。

Use when developing Pi extensions locally. Switches between npm-installed and local symlink mode for @zhushanwen/pi-* packages. Triggers: "link local", "dev link", "switch to local", "symlink extension", "unlink extension", "restore npm". Not for installing new packages or managing non-pi extensions.

Phase 1 (spec) of the xyz-harness workflow. Explores user intent, requirements and design before implementation, produces spec.md.

Phase 3 (dev) of the xyz-harness workflow. TDD-based implementation with automated review-gate.

Phase 4 (test) of the xyz-harness workflow. Integration/E2E testing with Test-Fix Loop Workflow.

Phase 2 (plan) of the xyz-harness workflow. Creates implementation plan, E2E test plan, and test case templates.

Use when creating or editing skills, diagnosing trigger failures, or testing trigger accuracy. Not for CLAUDE.md optimization or general rule writing — read references/rule-templates.md instead.

提交 Pull Request。触发词："提交 PR"、"创建 PR"、"push"、"提交代码"、 "pr-worktree"。仅用于 xyz-pi-extensions 项目。

Use when the user says "code review", "review code", "审查代码", "帮我 review", "review changes", "check code quality", "看看代码质量", or wants a multi-dimensional review of worktree/branch changes. Not for single-file edits or non-code changes. Not for single-dimension review — dispatch the specific review agent directly.

Manage evolution suggestion lifecycle: list, apply, skip, and rollback suggestions generated by /evolve. Trigger: /evolve-apply, evolve-apply, 应用建议, /evolve-rollback, evolve rollback.

View evolution daily reports and usage statistics. Shows session data, tool usage, token consumption, and trend analysis. Trigger: /evolve-report, evolve-report, 查看报告, 进化报告, /evolve-stats, evolve-stats.

Analyze session usage data and generate evolution suggestions. Runs Python analyzer data through LLM analysis to produce actionable improvement recommendations for CLAUDE.md and skills. Trigger: /evolve, evolve, 进化分析, 分析使用模式, analyze usage.

Trace code call chains from entry points (HTTP routes, WebSocket messages, IPC channels, class/method names) to all related files. Replaces batch-tracer, code-trace, issue-trace, review-tracer with AST-based graph traversal via code-review-graph. Triggers: trace code, call chain, code-link, find callers, trace route, analyze data flow, linked files. Not for general code review or lint.

Use when debugging web pages or Electron apps with screenshots, element inspection, UI interactions, or style checks. Also use for network monitoring, performance analysis, heap snapshots, or accessibility tree inspection. Not for API testing or backend debugging.

在 bare repo + worktree 结构中创建隔离的工作目录。自动检测 workspace、同步配置、 安装依赖和 git hooks。当用户说"创建 worktree"、"新 worktree"、"create worktree"、 "新建分支"、"开个新分支"时使用此 skill。即使用户只是说"我要做一个新功能"或 "帮我开个分支做 xxx"，也应考虑触发此 skill。

Use when creating or editing agent.md files for Claude Code or Pi subagents. Not for SKILL.md — use meta-sk-skill-writer instead.

图像/视频的专用结构化分析工具（OCR、错误诊断、架构图、数据图表、UI 对比/转代码、视频分析）。当前模型不支持图片输入时，作为 read 的降级替代。智谱 GLM-4.6V 优先，MiniMax VLM fallback。触发词：OCR、错误截图、架构图、数据图表、UI 对比、设计稿转代码、视频分析、模型不支持图片。