做什么
评估现有移动产品是否以及如何迁移到
技能库 智客分类:Agent 工作流 assess-react-native-migration
评估现有移动产品是否以及如何迁移到本地反应。 用于审计一个或多个用于迁移准备状态的产品寄存器,包括其iOS,Android等客户端分别生活在目录或寄存器中的产品;选择褐地,绿地,或基于检查站的路径;定义具有代表性的试验;或在执行前准备基线和ROI决定. 在没有产品范围或物证的情况下,每个回合向利益攸关方提出一个问题,而不是发出调查问卷.
官方网址:skills.sh
先看中文介绍;官方 description 原文单独保留,不改写 SKILL.md。
评估现有移动产品是否以及如何迁移到
审计一个或多个用于迁移准备状态的产品储存库,包括iOS、Android和其他客户居住在单独的目录或储存库中的产品;选择棕地、绿地或基于检查站的路径;确定具有代表性的试验;或编写b
按 Agent Skills 渐进披露:启动时只加载 name 与 description(约 100 token);任务匹配后才读入整份 SKILL.md 正文;scripts/、references/、assets/ 仅在需要时再读。 本文件正文结构:Assess React Native Migration、Establish Product Scope、First Response Gate、Rules、Select the Evidence Mode、1. Collect Decision Evidence。
文件分析:这是一份仅含 SKILL.md 的指令型技能,代理激活后整份正文进入上下文。
Assesses whether and how an existing mobile product should migrate to React Native. Use when auditing one or more product repositories for migration readiness, including products whose iOS, Android, and other clients live in separate directories or repositories; choosing brownfield, greenfield, or a checkpoint-based path; defining a representative trial; or preparing a baseline and ROI decision before implementation. When product scope or material evidence is unavailable, grills the stakeholder with exactly one question per turn instead of sending a questionnaire.
Assess React Native MigrationEstablish Product ScopeFirst Response GateRulesSelect the Evidence Mode1. Collect Decision Evidence2. Choose a Path3. Define a Representative Checkpoint4. Test the EconomicsOutput Contract
· 许可:MIT
来源分类:skills.sh agent-skill
nameassess-react-native-migrationdescription具体调用语法与可用工具以目标 Agent 客户端为准。 查看调用机制说明 ↗
先选择目标 Agent 和安装范围,保留技能包的附属文件,安装后检查客户端能否发现该技能。
复制安装指令给支持 Agent Skills 的代理,确认其中的目标目录与客户端匹配。
把 Agent Skill「assess-react-native-migration」安装到我的项目:SKILL.md 原文与官方 description 见 https://zicq.com/zh/skills/skl-4799dfd91f3bb36a-%E8%AF%84%E4%BC%B0%E5%9C%9F%E8%91%97%E7%A7%BB%E5%BE%99%E7%9A%84%E5%8F%8D%E5%BA%94.html 请存为 .cursor/skills/assess-react-native-migration/SKILL.md 或 .claude/skills/assess-react-native-migration/SKILL.md,frontmatter 的 name 与 description 保持原样,不要改写。
需要 Node.js 与 npx。先查看仓库技能列表,确认实际名称。
npx skills add 'https://github.com/callstackincubator/agent-skills' --list
npx skills add 'https://github.com/callstackincubator/agent-skills' --skill 'assess-react-native-migration'
CLI 会交互选择目标 Agent,默认安装到项目;用户级安装使用 -g。先通过查看命令核对仓库内容,再用 npx skills list 检查已安装技能。
Produce a read-only migration decision. Diagnose the product and delivery system; do not execute the migration.
Run the assessment from a workspace that exposes as many production client codebases as possible. The current checkout is not evidence that it contains the whole product.
Before assessing readiness:
When iOS and Android are both supported, inspect both native codebases before recommending a path. If a codebase remains unavailable, mark its evidence unknown, state that the assessment covers only the accessible platforms, and lower confidence accordingly. Do not infer that a platform is unsupported merely because its project is absent from the current repository.
Scope gate: every supported production client is listed, and each codebase is accessible, explicitly unavailable, or confirmed not to exist.
When the scope gate has not passed, the first response must be exactly:
**Question:** Where can I access the production codebase for each client platform this product supports, including iOS and Android if both exist?
**Why it matters:** A migration path based on only one platform can miss native dependencies, product behavior, and delivery constraints that change the decision.
After the scope gate passes, grill rather than survey when repository evidence is unavailable.
If the measurable migration driver is unknown, the first response must be exactly:
**Question:** What measurable delivery or business problem should a React Native migration solve?
**Why it matters:** This determines whether migration is relevant and which outcomes the assessment must test.
If the driver is already known, ask only the next highest-impact unknown using the same two-line shape. End the turn immediately after the question and reason. Do not add a preamble, questionnaire, recommendation, or implementation guidance.
observed, measured, reported, assumed, or unknown.Use repository-backed assessment when source code or delivery artifacts are available:
Use interview assessment when the repository is unavailable or material evidence remains missing:
Before the evidence gate passes, every response must contain only:
**Question:** [one question]
**Why it matters:** [one sentence]
Do not include a questionnaire, path recommendation, checkpoint, or implementation guidance during these turns. If the user pauses the interview, return the current evidence state and the single highest-impact unknown without pretending the assessment is complete.
Interview turn gate: one answer has been requested, its decision impact is explicit, and no second question appears.
State the decision, deadline, current alternative, and measurable driver. A framework preference is not a driver.
Inspect these dimensions:
| Dimension | Minimum evidence | | --- | --- | | Product | Supported platforms, app variants, shared versus platform-specific roadmap, critical flows, accessibility, analytics, and edge cases | | Native surface | SDKs, modules, permissions, background work, app extensions, payments, hardware APIs, custom rendering, and viable React Native paths | | Continuity | Auth and sessions, secure and persisted storage, push tokens, deep links, subscriptions, installed-user update, legal, security, and offline constraints | | Verification | Reproducible builds, test accounts, manual and automated QA, device control, native-reference evidence, performance baselines, and independent review | | Release | Current cadence and recovery, internal distribution, flags, experiments, store rollout, and desired binary plus optional OTA lanes | | Ownership | Decision authority and owners for artifacts, parity, native boundaries, shared foundations, verification, and releases | | Agent governance | Approved model and source boundary, protected secrets and test data, least-privilege access, evidence retention, audit trail, and human architecture and release approval | | Delivery baseline | Duplicate implementation and review, waiting and handoffs, parity gap, two-platform verification, release metrics, defects, rework, and maintenance cost |
For an OTA-dependent plan, require an owner plus runtime compatibility, rollout, observability, rollback or republish, and audit policy. OTA availability alone is not a migration benefit.
Use a small migration core that combines existing product and native knowledge with React Native migration expertise. Ask only for missing facts that could change the decision; expose the rest as assumptions.
Gate: every dimension has evidence or an explicit unknown, and every path-blocking unknown is named.
Choose one outcome and state why the alternatives lose.
| Outcome | Recommend when | | --- | --- | | Path A: brownfield | Release or installed-user continuity dominates, native coupling is deep, flows can move independently, or whole-app cutover risk is unacceptable. Include the cost of host boundaries and dual architecture. | | Path B: greenfield | Behavior is recoverable, native dependencies have credible replacements, continuity can be proven, verification is strong, and legacy scope can be controlled until replacement. | | Path C: greenfield-first checkpoint with brownfield fallback | Greenfield offers a simpler target but material uncertainty remains, and completed React Native work can be proven inside the native hosts before scaling. | | Defer | The business case is plausible, but evidence, verification, ownership, budget, or release readiness is missing. Name the smallest readiness work and reopening condition. | | Do not migrate | The native system meets the desired outcomes, duplicated mobile delivery is not material, the roadmap is asymmetric, platform-specific work dominates, or risk-adjusted return is not credible. |
Treat Path C as Callstack's emerging post-2025 operating model, not an industry benchmark. Agent access makes behavioral porting more viable; only a measured checkpoint on this product establishes speed and quality.
After Path A is accepted, hand implementation planning to react-native-brownfield-migration. Do not repeat its Expo, XCFramework, AAR, or host-integration guidance.
Gate: one outcome is supported by decisive evidence, rejected alternatives have reasons, and confidence reflects evidence quality.
Use a checkpoint for Path C and whenever one uncertainty could invalidate the recommended path. Set a fixed calendar and effort budget supplied by the organization. Select two or three vertical flows:
Tie each flow to native source references, runtime evidence, owners, and parity scenarios. Do not select only easy screens.
Define measurable acceptance criteria against the existing product:
Run two passes on at least one flow:
Before scaling, assign owners for MIGRATION.md, SCREENS.tsv, STATE_AND_STORAGE.tsv, DEPENDENCIES.tsv, EVENTS.tsv, and PARITY_CHECKS.md. During assessment, identify missing artifacts rather than creating full inventories unless asked.
End with one decision: continue greenfield, continue Path C, move completed work into Path A, defer, or stop. Do not extend the checkpoint merely because its result is inconvenient.
Gate: flows, budget, criteria, evidence, owners, and terminal decisions are explicit.
Map one representative change from ready-for-implementation to verified availability on both platforms. Capture two clocks:
Count potential return from removed duplicate implementation and business-logic review, lower parity coordination, engineer mobility, and measured release or recovery improvements. Keep iOS and Android device verification in the model. Keep web convergence outside the base case unless separately funded and assessed.
Count full investment: checkpoint, dual maintenance, native modules, training, idiomatic second pass, testing and device automation, observability, release infrastructure, platform ownership, cutover, and native-surface retirement.
Do not invent payback inputs. Use ranges when supplied data is uncertain. Revise, defer, or stop when verification and boundary work consume expected savings, native surfaces keep growing, dual maintenance has no retirement bound, agents add rework without verified throughput, parity improves without delivery improvement, or released capacity has no planned use.
Gate: marginal benefits and full investment can be compared without treating two-platform verification as removed work.
Use this contract only after the evidence gate passes or when the user ends the interview.
Return a concise report in this order:
Keep the conclusion diagnostic. If execution comes up, state only that orchestration, task contracts, prompts, retry rules, and private agents are engagement-specific. Never expose or invent private implementation material.
Agent 工作流
创造有效技能指南。 当用户想创造出新的技能(或更新现有的技能),以专业知识,工作流程,或工具集成来扩展克洛德的能力时,应该使用这种技能.
Agent 工作流
使用ClawdHub CLI搜索,安装,更新并发布从taladhub.com的代理技能. 需要获取苍蝇上的新技能时使用,将安装的技能同步到最新版本或特定版本,或者发布 npm-instainddhub CLI 的新/更新的技能文件夹.
Agent 工作流
管弦乐团多代理团队,任务设定周期,交接协议,审查工作流程. 使用时间: (1)建立2+特派员队伍,具有不同专业,(2)确定任务路线和生命周期(收录框_ spec_建设_审查_完成),(3)在特派员之间制定交接协议,(4)建立审查和质量关口,(5)管理特派员之间的交流和文物共享.
Agent 工作流
Spec-first,TDD,子代理驱动的软件开发工作流程. 当:(1)构建任何新功能或应用——触发脑暴_计划_子代理执行回路,(2)调试出一个bug或测试失败——触发系统性的根起过程,(3)用户说"让我们构建","帮助我计划","我想添加X",或"这个被打破",(4)完成一个功…