miscellany

指导编码智能体以 capability-runtime 为业务落地入口，交付基于 capability-runtime 的 skills / agents / workflows，并在 Greenfield 或 Legacy Convergence 场景下优先使用 Runtime public surface、structured output、NodeReport、host summary 与 service/session surfaces。只要任务目标是用 capability-runtime / capability_runtime 落地业务代码、收敛下游 runtime boundary，或涉及 Runtime.run / Runtime.run_stream / run_structured / run_structured_stream / AgentSpec / PromptRenderMode / prompt_render_mode / _runtime_prompt / precomposed_messages / multimodal / vision / image input / 多图输入 / 视频抽帧输入 / OpenAI-compatible messages / image_url content parts / WorkflowSpec / NodeReport / RuntimeServiceFacade / describe_capability / summarize_host_run，就应优先使用本技能。不要用于普通通用编码、prompt-only 任务、直接学习上游原生框架 API，或任何明确要求“直接用 skills-runtime-sdk / Agently / provider SDK，不走 capability-runtime”的任务；若已触发但随后识别出这是反目标，必须立即退出，并停止提供任何上游实现细节、伪代码或 API 猜测。

Turn vague ideas into a validated design/spec through structured brainstorming. Use before any creative work - new features, UI/components, behavior changes, refactors, architecture decisions. Trigger whenever a user asks to brainstorm, define requirements, propose approaches, write a design doc, or says something like 'I want to build X' or 'how should we approach Y'. Even seemingly simple tasks benefit from a quick design pass.

用 Skills Runtime SDK（Python）开发复杂业务 agent、skills、workflow 的编码智能体指南。用户一旦提到 skills_runtime、Skills Runtime SDK、overlay YAML、FakeChatBackend、AgentBuilder、Coordinator、skill_ref_read、skill_exec、approvals/sandbox、WAL/replay、exec sessions、spawn_agent/send_input/wait、waiting_human/resume、examples/apps/workflows，或要在本仓上落地复杂业务开发/修复/回归，就应优先使用本技能。不要用于与本框架无关的通用编码或纯文案任务。

Trace and document the complete call chain / data flow of any feature or process in a codebase. Use this skill whenever the user asks to understand how a feature works end-to-end, trace a call chain, analyze code flow, map data flow across layers, reverse-engineer a process, or asks questions like "how does X call Y", "what happens when the user clicks Z", "trace the request from frontend to database". Also use it when the user wants the analysis exported to a file for offline reading.

Use only when the user explicitly wants to build with the Agently framework (mentions Agently/agently/OpenAICompatible/TriggerFlow/ToolExtension/ChromaCollection, or says “用 Agently 做/用 agently 做”). Deliver runnable code plus regression tests validating schema/ensure_keys and streaming (delta/instant/streaming_parse), with optional tools (Search/Browse/MCP), TriggerFlow orchestration, KB (ChromaDB), and serviceization (SSE/WS/HTTP). Do not use for generic streaming/testing questions that are not about Agently, or for prompt-only writing without tests/structure.

用 tmux 稳定驱动交互式 CLI：启动 session、发送按键、等待输出就绪，并支持 worker 向 controller pane 回传结果（backchannel）。

Send a critical mobile push through Day.app (Bark) with one GET request. Use when a task finishes, fails, is blocked, or needs immediate alerting, and you should summarize task name and summary from current task context before sending.

DOCX 文档离线读写：提取/分析、OOXML 解包编辑回包、批注与修订（tracked changes/redlining）。适用于合同/制度/论文等需要保留格式与修订痕迹的场景（依赖安装可能需要网络）。

Render and view webpages using a headless browser (Playwright) to fetch JS-rendered HTML, extract visible text, and optionally save full-page screenshots. Use when a user asks to “无头浏览器打开/查看网页”, needs the rendered DOM instead of raw curl HTML, or wants a screenshot of a page.

Loopback 迭代开发循环 - 基于 Ralph Loop 思想实现的 Codex 版本。用于创建自引用的迭代开发循环，让 AI 在多次迭代中逐步改进代码，直到任务完成。Use when: (1) 需要多次迭代改进的任务, (2) 复杂功能需要分步实现, (3) 需要自我修正的代码生成, (4) 迭代优化已有代码.

PDF 文档离线读写与表单处理：提取文本/表格、合并拆分、生成 PDF、填写表单。适用于“本地处理/读取/生成 PDF 文件”（依赖安装可能需要网络）。

PPTX 文档离线读写：解析/替换/重排/缩略图、OOXML 解包编辑回包，以及 html2pptx（HTML→PPT）工作流。适用于生成与维护演示文稿（依赖安装可能需要网络）。

Transform PRD (Product Requirements Document) into actionable engineering specifications. Creates detailed technical specs that developers can implement step-by-step without ambiguity. Covers data modeling, API design, business logic, security architecture, deployment, and agent system design. Use when: converting product requirements to technical specs, validating PRD completeness, planning technical implementation, creating task breakdowns, or defining test specifications. Triggers: 'PRD to spec', 'convert requirements', 'technical spec from PRD', 'engineering doc from requirements', 'validate PRD'.

从产品 PRD 产出“复刻级可落地”的 UI/UX 研发规格文档包（目录同构骨架、公共基座、组件/页面契约、覆盖映射、索引与 worklog）。适用于需要把 PRD 转成前端可复刻实现的规格文档、UI/UX 研发规格、界面契约与验收标准的场景；避免用于只要视觉灵感/纯 UI 赏析或直接写代码实现的请求。

Write complete, unambiguous PRDs that development teams can implement without guesswork. Includes requirement discovery framework, structured documentation methodology, completeness checklists, and common pitfall avoidance. Use when: writing new PRDs, reviewing PRD drafts, validating requirement completeness, preparing for engineering handoff. Triggers: 'write PRD', '写PRD', '产品需求文档', '需求文档', '需求规格', '需求评审', '完善需求', 'create requirements doc', 'product requirements', 'feature spec', 'requirements document'. Anti-triggers: 'technical design doc', 'architecture design', 'implementation plan', 'API design', '架构设计', '技术方案', '实现方案', '接口设计'.

对任意代码仓库进行合规审计并生成可取证报告（Markdown + JSON findings），覆盖“是否遵循 AGENTS.md/仓库规则/用户指令”“文档索引/规格/工作记录/任务总结”“TDD 与离线回归证据”“可复现性（.env.example 等）”“潜在密钥泄露与仓库卫生”等；并支持在**人类勾选 finding.id** 后执行选择性低风险整改（默认不改业务逻辑）。触发场景：仓库交付前自检、接手陌生仓库、需要合规审计报告、需要把整改条目做成可选择的执行清单。

End-to-end source code repository deep dive and review. Use when you need to understand a repo’s architecture, core abstractions, entrypoints and execution flow, extension points, and operational/developer setup; and when you must deliver a structured report (Markdown + standalone HTML) with diagrams (Mermaid), actionable recommendations, and a scored evaluation.

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, update or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

Use when a user asks to review, interpret, or audit an AI agent skill (SKILL.md plus bundled scripts/references/assets) for capabilities, triggering behavior, tool/command usage, safety & privacy risk, supply-chain provenance, quality gaps, and improvement recommendations; also use when validating a skill before installing or deploying it.

A portable, reproducible UI/UX spec standard: scan a frontend repo for UI sources and scaffold a ui-ux-spec documentation bundle (tokens, global styles, components, patterns, pages, a11y). Also supports plan-driven UI-only refactors based on an existing ui-ux-spec. Excludes business logic and domain workflows.

Package an existing React UI/UX demo into a single self-contained .jsx file with default-export root component, zero third-party runtime dependencies (no react-router/lucide/echarts/etc.), in-file styles (style tag or inline style objects), inline SVG icons, embedded or placeholder images, and state-based navigation. Use when asked to “合并为单文件 JSX/单文件打包/one-file React/零外部依赖/内联 CSS/替换图标库/用 state 做路由/把 demo 打包成独立 JSX 文件”.

Excel 表格离线读写与公式校验：创建/修改 xlsx，保持公式可复算，输出必须零公式错误；附带 LibreOffice 重算与错误扫描脚本（依赖安装可能需要网络）。

Write comprehensive PRDs for AI Agent products—covering agent identity, capability architecture (skills, tools, memory, RAG, workflows), behavior specifications, safety guardrails, and evaluation frameworks. Use when: designing conversational agents, autonomous agents, copilots, multi-agent systems, or any LLM-powered agentic application. Triggers: 'AI agent PRD', 'agent product requirements', 'design AI agent', 'agent capability spec', 'LLM agent requirements', '智能体PRD', '智能体需求文档', '对话机器人PRD', '多智能体系统需求'. Anti-triggers: '传统PRD（非智能体）', '只润色提示词/只写Prompt', '只写用户故事/验收标准但不涉及工具调用、记忆或RAG'.

Extract complete, replicable engineering specifications from existing codebases. Produces documentation detailed enough to fully replicate a project without seeing the original source code—even using a different tech stack. Use when: analyzing existing projects, documenting legacy systems, creating technical specs from code, preparing for system migration, or onboarding new teams. Triggers: 'extract spec from code', 'document codebase', 'analyze project architecture', 'create spec from existing system'.

Create new high-quality agent skills with a standalone, repeatable workflow (no dependency on other skills). Use when you want to go from a vague skill idea → narrow scope → extract expert frameworks → write SKILL.md + examples/evals/index/changelog artifacts → self-validate with test prompts.