agent-foundry

用于 Android / Material 3 平台规范问题：当用户要做 Android 界面规则、导航模式、window size classes、dynamic color、edge-to-edge、无障碍或 Material 合规审查时使用。Prefer `compose-skill` when the main task is Compose/KMP 的状态、架构、DI、网络或实现细节；Combine with it when Android / Material 规则需要落到 Compose 实现。

用于需要“顶级专家视角”的任务：当用户要专家判断、方案对比、trade-off 分析、优先级决策、审美判断或复杂方法论，或者会说“从战略上看”“你别给我常规建议”“像最懂这件事的人那样判断”“从高手视角看”“按高手打法判断一下”“像世界级专家那样看这件事”，or English asks like “give me the expert take”, “think like a world-class strategist”, “don't give me the default answer” 时使用。Do not use for 简单事实查询、纯资料罗列或常规执行型任务。

用于图标搜索、选型与同步：当用户要找 icon、替换 emoji 占位、找相似图标、导出 SVG 或把图标接入项目时使用。Prefer this over 手写 SVG 猜测；Combine with `frontend-design` or `ui-ux-pro-max` when 图标只是界面方案的一部分；默认先做查找/选型，不把一句“找个图标”视为已授权 setup、同步或 `global` 变更。

用于需求还模糊的开放式产品/设计任务：当用户要新功能、新组件、新交互或行为变更，但目标、约束、形态还没想清楚，或者会说“先别写代码”“先别急着做”“先帮我想清楚”“先梳理一下方案”“先拆一下需求”“先把目标和范围捋清”“先定方向”，or English asks like “don't write code yet”, “help me think this through first”, “let's clarify the scope first” 时使用。Combine with `planning-with-files` when 任务会跨多轮推进；Do not use for 常规 bugfix、纯 review、最终验证或只改措辞。

Build, scaffold, refactor, and troubleshoot ChatGPT Apps SDK applications that combine an MCP server and widget UI. Use when the agent needs to design tools, register UI resources, wire the MCP Apps bridge or ChatGPT compatibility APIs, apply Apps SDK metadata or CSP or domain settings, or produce a docs-aligned project scaffold. Prefer a docs-first workflow by invoking the openai-docs skill or an OpenAI developer docs MCP integration before generating code.

用于 Chrome 插件 / 浏览器扩展架构设计：当用户要规划 Manifest V3、service worker、content script、popup、options page、权限、消息通信、安全边界或跨端方案时使用。Combine with `frontend-design` when 插件界面本身也要重做；Do not use for 纯网页改版或与扩展架构无关的常规前端任务。

">"

用于会话内上下文压缩：当用户要压缩长对话、做运行摘要、做 compaction、减少 token 消耗但保持任务连续性时使用。Prefer `memory-systems` for 跨会话长期记忆，Prefer `context-fundamentals` for 上下文工程基础概念；Do not use for 需要跨会话持久化事实或偏好存储的问题。

用于上下文工程基础问题：当用户要理解 context window、注意力行为、progressive disclosure、context budgeting 或高层 agent 上下文架构，或者会问“为什么 agent 一长对话就失忆”“为什么 agent 聊久了就不靠谱”“上下文到底怎么设计”“context budget 怎么分”“上下文预算怎么分”“为什么 prompt 很长反而更差”“为什么 prompt 越堆越差”，or English asks like “why does the agent get worse in long chats”, “how should I budget context”, “why does a longer prompt perform worse” 时使用。Prefer `context-compression` for 会话内压缩，Prefer `memory-systems` for 跨会话长期记忆；Do not use for 直接实现具体压缩流程或持久化记忆系统。

用于创建或重构 skill 本身：当主任务是编写 `SKILL.md`、frontmatter、目录结构、命名、bundled resources，或把 skill 迁移到当前 Codex / 本地 skills 约定时使用。Prefer `skill-creator` for eval 循环、benchmark 驱动迭代或 trigger 准确率优化。

Use when building or iterating on a web game and you need a tight implement-test-observe loop; prefer the bundled Playwright client, inspect screenshots and text state, and validate every meaningful change before moving on.

用于搜索与评估外部 skill：当用户问“有没有能做 X 的 skill”“帮我找 skill”“看看值不值得装”时使用。先用 `npx skills find` 做候选搜索、边界说明和风险提示；只有在用户明确同意后才进入安装。Do not use for 直接替代目标领域 skill 或默认自动安装第三方内容。

用于无障碍审查与修复：当用户要检查或修 HTML 可访问性、ARIA、键盘导航、焦点管理、颜色对比、表单错误、dialog 或 WCAG 合规时使用。Combine with `frontend-design` or `ui-ux-pro-max` when 先做界面方案再做无障碍验收；Do not use for 纯视觉改版或一般性能优化。

用于动画性能审查与修复：当用户说动画卡顿、transition 发抖、scroll-linked motion 不顺、blur 太重，或要审查 CSS/JS 动画性能时使用。Combine with `frontend-design` or `ui-ux-pro-max` when 先做动效设计再做性能收口；Do not use for 纯视觉方向选择或非动效性能问题。

用于设计和实现新的 Web / 前端界面：当用户要新页面、新组件、改版、重设计，且核心目标是视觉方向、布局、交互手感和整体 UI 气质时使用。Combine with `fixing-accessibility` / `fixing-motion-performance` when 需要专项验收；Prefer `react-best-practices` when the main goal is 现有 React/Next 的性能、缓存、bundle 或代码健康优化；Do not use for 纯平台规范或只做工程性能收口的问题。

Help address review/issue comments on the open GitHub PR for the current branch using gh CLI; verify gh auth first and prompt the user to authenticate if not logged in.

Use when a user asks to debug or fix failing GitHub PR checks that run in GitHub Actions; use `gh` to inspect checks and logs, summarize failure context, draft a fix plan, and implement only after explicit approval. Treat external providers (for example Buildkite) as out of scope and report only the details URL.

用于 git 提交：当用户要 commit、提交代码、保存当前改动、stage 并提交时使用。按 Conventional Commits 生成提交信息；Combine with `verification-before-completion` when 提交前需要新鲜验证证据；Prefer `github-pr-creation` when 下一步是直接开 PR。

用于创建 GitHub PR：当用户要开 PR、提交评审、检查是否 ready for PR、生成 PR 标题/描述或整理待提交项时使用。Combine with `git-commit` and `verification-before-completion` when 提交前还需要收口；Prefer `github-pr-merge` when the task is 合并已有 PR。

用于合并 GitHub PR：当用户要 merge PR、finalize PR、approve and merge、完成合并并做合并前检查时使用。Combine with `verification-before-completion` when 需要合并前再确认检查状态；Do not use for 创建新 PR 或只讨论 review 建议。

用于 GitHub PR review 工作流：当任务是拉取 PR 评论、分拣 review threads、修当前 PR 的 reviewer 反馈、回复 reviewer、检查 review readiness 时使用。Prefer `receiving-code-review` when the feedback 已贴在聊天里，且主任务是先判断建议是否成立；Combine with `verification-before-completion` before 宣称 PR 已 ready。

Use when the user wants to generate, edit, or batch-produce images via the OpenAI Image API; prefer the bundled `scripts/image_gen.py`, require `OPENAI_API_KEY`, and make the generate-vs-edit intent explicit before running.

用于 iOS / iPhone 平台规范问题：当用户要做 iOS 界面规则、导航、touch targets、safe area、Dynamic Type、Dark Mode、无障碍或 HIG 合规审查时使用。Prefer `swiftui-expert-skill` when the task is SwiftUI 代码架构、API 使用或性能优化；Combine with it when 平台规则需要落到 SwiftUI 实现。

Use when the user wants to read, create, triage, or update issues, projects, labels, cycles, or team workflows in Linear; use the current host's Linear connector or MCP integration, read first for context, then batch the writes.

用于 macOS 平台规范问题：当用户要做桌面端菜单栏、工具栏、窗口、多窗口、快捷键、指针行为、Mac Catalyst 或原生桌面交互规则时使用。Prefer `swiftui-expert-skill` when the task is SwiftUI 实现细节而不是 Mac 平台行为；Combine with it when Mac 平台规则需要落到 SwiftUI 代码实现。

">"

Use when the user wants to deploy, publish, preview, or link a web project on Netlify; prefer the Netlify CLI, verify auth and site linkage first, and default to preview deploys unless production is explicitly requested.

Use when the user wants to turn chats, decisions, notes, or learnings into structured Notion pages; search existing context first, choose the right destination, then create or update pages with links, owners, and reusable structure.

Use when the user wants to prepare agendas, pre-reads, or decision notes from Notion context; gather prior docs first, tailor the structure to the meeting type, and create a page that is ready to share.

Use when the user wants to research across Notion pages and synthesize the results into a brief, comparison, or report; search first, fetch the key sources, then write a cited and structured document.

Use when the user wants to turn a Notion PRD or feature spec into an implementation plan, tasks, and progress tracking; read the spec first, capture assumptions, then create linked plan and task artifacts.

Use when the user asks how to build with OpenAI products or APIs and needs up-to-date official documentation with citations, help choosing the latest model for a use case, or explicit GPT-5.4 upgrade and prompt-upgrade guidance; prioritize an OpenAI docs MCP integration when available, use bundled references only as helper context, and restrict any fallback browsing to official OpenAI domains.

用于需要文件化计划的长任务：当任务够大，需要 task plan、进度跟踪、研究记录、迁移方案或 `/clear` 后恢复时使用。Best for 多步骤项目、研究、迁移和明显会超过少量 tool calls 的任务；Combine with the domain skill that owns the actual work；Do not use for 很小的一次性任务或只需口头澄清的问题。

Use when the task needs real browser automation from the terminal for navigation, form filling, snapshots, screenshots, extraction, or UI-flow debugging; prefer the bundled `scripts/playwright_cli.sh` wrapper over ad-hoc browser code.

用于 React / Next.js 现有代码优化：当用户要检查渲染性能、data fetching、bundle size、缓存、架构健康或性能回退时使用。Combine with `frontend-design` when 先改界面再做工程收口；Prefer `frontend-design` when the main goal is 新界面设计或视觉重构；Do not use for 与 React/Next 无关的问题。

用于判断 review 建议该不该改：当收到 reviewer 反馈，需要先评估建议是否技术上正确、是否值得实现、是否存在误判时使用。Prefer `github-pr-review` when the main task is 操作 live GitHub PR review threads；Do not use for 直接拉取/回复 PR 线程或创建/合并 PR。

Use when the user wants to deploy, publish, or set up an application on Render; prefer the Git-backed Blueprint flow with `render.yaml`, validate when the Render CLI is available, and return a dashboard deeplink plus next steps.

用于在实现过程中主动发起一次中间代码审查：当第一版已经写出来、准备开 PR 前先自审，或者用户说“先帮我 review 一下这版”“先站在 reviewer 角度挑问题”“先别 merge，先审一轮”时使用。Prefer `verification-before-completion` when 还没完成基本验证；Prefer `github-pr-review` when 已经有 live PR 评论线程；Combine with `receiving-code-review` when 审查后要判断哪些建议值得改；Do not use for 需求还没成形或根因未明的 bug 诊断。

用于做 SaaS 官网 / 营销站 / 首页的研究、策略和报告：当用户说“联网看看别人家怎么做官网”“研究 3-5 个同类产品首页”“别人首屏一句话、CTA、SEO 文案怎么写”“先别改页面，先给我一份报告”“判断首页该做品牌入口还是转化页”这类需求时使用。聚焦竞品官网研究、首页结构、价值表达、CTA、品牌入口、proof、SEO / 营销表达与动效策略，并结合当前项目输出可落地建议。Prefer `frontend-design` when the user wants to 直接改页面、产出具体前端实现或视觉稿；Prefer `brainstorming` when 业务目标、受众或约束还没定清；Combine with `planning-with-files` when 研究范围大、需要分阶段沉淀或持续跟踪。

Perform language and framework specific security best-practice reviews and suggest improvements. Trigger only when the user explicitly requests security best practices guidance, a security review/report, or secure-by-default coding help. Trigger only for supported languages (python, javascript/typescript, go). Do not trigger for general code review, debugging, or non-security tasks.