decision-evaluation
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseDecision Evaluation
决策评估
Overview
概述
This skill evaluates one explicit design decision at a time.
It does not discover the whole design. It does not validate overall readiness. Its job is to compare real options for a bounded decision node and produce a recommendation with reasons.
本Skill每次评估一个明确的设计决策。
它不负责探索整体设计,也不验证整体就绪状态。其职责是针对有限范围的决策节点对比实际选项,并给出带有理由的推荐。
When to Use
使用场景
Use this skill when:
- the design already contains a concrete decision node
- the user is effectively asking "which of these approaches should we choose?"
- the decision will materially affect architecture, implementation path, compatibility, cost, or operational risk
Typical examples include:
- auth model
- coordination pattern
- storage strategy
- sync model
- state management approach
- retry or execution model
- workflow rollback policy
- workflow approval strategy
- methodology stop condition
- methodology evidence threshold
- framework routing rule
- framework handoff contract
- language or runtime selection
- UI framework selection
- data-layer library (ORM, query builder, validation) selection
- tooling selection (bundler, linter, test runner, CI platform)
Do not use this skill when:
- the design tree is not formed yet
- the task is still broad design discovery
- the main need is deeper refinement rather than bounded comparison
- the task is a final readiness gate
在以下场景中使用本Skill:
- 设计中已包含具体的决策节点
- 用户实际询问的是“我们应该选择这些方案中的哪一个?”
- 该决策将对架构、实现路径、兼容性、成本或运营风险产生实质性影响
典型示例包括:
- 认证模型
- 协调模式
- 存储策略
- 同步模型
- 状态管理方案
- 重试或执行模型
- 工作流回滚策略
- 工作流审批策略
- 方法论终止条件
- 方法论证据阈值
- 框架路由规则
- 框架切换契约
- 语言或运行时选择
- UI框架选择
- 数据层库(ORM、查询构建器、验证工具)选择
- 工具选型(打包器、代码检查器、测试运行器、CI平台)
请勿在以下场景中使用本Skill:
- 设计树尚未形成
- 任务仍处于宽泛的设计探索阶段
- 主要需求是深度细化而非有限范围对比
- 任务是最终就绪检查
Constraint Gathering
约束收集
When the decision involves technology selection (language, framework, library, or tooling), perform this step before the formal comparison.
Ask the user to confirm:
- Hard constraints: existing tech stack, license requirements, budget limits, deployment environment restrictions.
- Preferences or already-excluded options.
- Priority of evaluation dimensions (e.g., performance > ecosystem > learning curve).
- Must-have requirements (TypeScript support, SSR capability, plugin system, etc.).
Filter the candidate list based on confirmed constraints. Only feasible options proceed to formal comparison.
Output: confirmed constraints + filtered candidate list.
When the decision is non-technical but still bounded, do not force technology-selection questions.
Instead confirm the decision boundary, real alternatives, and the dimensions that matter for this workflow, methodology, or framework choice.
当决策涉及技术选型(语言、框架、库或工具)时,在正式对比前执行此步骤。
请用户确认:
- 硬性约束:现有技术栈、许可证要求、预算限制、部署环境限制。
- 偏好或已排除的选项。
- 评估维度的优先级(例如:性能 > 生态系统 > 学习曲线)。
- 必备需求(TypeScript支持、SSR能力、插件系统等)。
根据确认的约束过滤候选列表。仅保留可行选项进入正式对比。
输出:确认的约束 + 过滤后的候选列表。
当决策为非技术性但仍属于有限范围时,请勿强制询问技术选型相关问题。
而是确认决策边界、实际替代方案,以及对该工作流、方法论或框架选择至关重要的评估维度。
Core Responsibilities
核心职责
Your responsibilities are:
- Frame the exact decision question. For technology selections, frame the question based on constraints confirmed in the Constraint Gathering step.
- Keep the work bounded to one explicit decision node.
- Present 3 to 4 real options when real options exist.
- Compare trade-offs clearly and concretely.
- Recommend one option or a justified combination when appropriate.
- Explain why the rejected options are not preferred in this context.
- Record the result in a reusable decision record.
你的职责包括:
- 明确决策问题。对于技术选型,基于约束收集步骤中确认的约束来构建问题。
- 确保工作范围限定在单个明确的决策节点内。
- 当存在实际选项时,列出3至4个真实选项。
- 清晰、具体地对比各选项的权衡关系。
- 推荐一个选项,或在合适情况下推荐合理的组合方案。
- 解释为何在当前场景下不优先考虑被否决的选项。
- 将结果记录在可复用的决策记录中。
Comparison Standard
对比标准
Compare options using the dimensions that matter to the decision. These often include:
- implementation complexity
- compatibility with the current system
- operational risk
- performance
- maintainability
- future flexibility
- debugging and observability implications
Do not force every dimension into every decision. Use only the ones that actually matter.
使用与决策相关的维度对比选项,这些维度通常包括:
- 实现复杂度
- 与当前系统的兼容性
- 运营风险
- 性能
- 可维护性
- 未来灵活性
- 调试与可观测性影响
无需在每个决策中都覆盖所有维度,仅使用实际相关的维度即可。
Expected Outputs
预期输出
Produce or update a that includes:
design_statedesign_target_typedecision_nodesdecisionsrisks- optional downstream impact notes for the design tree
生成或更新包含以下内容的:
design_statedesign_target_typedecision_nodesdecisionsrisks- 可选的设计树下游影响说明
Diagram Conventions
图表规范
Use character diagrams inside code blocks (no language tag) to visualize trade-offs and architecture topology.
在代码块(无语言标签)中使用字符图表可视化权衡关系和架构拓扑。
Star Ratings
星级评分
Use when comparing 3+ options across 2+ quantitative dimensions.
Overall Rating:
Auth0: ★★★★☆ 4/5
Keycloak: ★★★☆☆ 3/5
Self-build: ★☆☆☆☆ 1/5
Dimension Breakdown:
Security: Auth0 ★★★★★ Keycloak ★★★★☆ Self ★★☆☆☆
Speed-to-MVP: Auth0 ★★★★★ Keycloak ★★★☆☆ Self ★☆☆☆☆
Cost-at-scale: Auth0 ★★★☆☆ Keycloak ★★★★☆ Self ★★★★★Rules:
- (filled, U+2605) +
★(empty, U+2606), 5-star scale (max = ★★★★★)☆ - Format:
label ★★★★☆ 4/5 - Always show all 5 stars (e.g. , never
★★☆☆☆)★★ - Overall rating section first, then Dimension Breakdown
- Breakdown section: one dimension per row, all options on the same row
- Always state the scale explicitly
当在2个以上量化维度对比3个以上选项时使用。
Overall Rating:
Auth0: ★★★★☆ 4/5
Keycloak: ★★★☆☆ 3/5
Self-build: ★☆☆☆☆ 1/5
Dimension Breakdown:
Security: Auth0 ★★★★★ Keycloak ★★★★☆ Self ★★☆☆☆
Speed-to-MVP: Auth0 ★★★★★ Keycloak ★★★☆☆ Self ★☆☆☆☆
Cost-at-scale: Auth0 ★★★☆☆ Keycloak ★★★★☆ Self ★★★★★规则:
- 使用(实心,U+2605)和
★(空心,U+2606),5星制(满分=★★★★★)☆ - 格式:
label ★★★★☆ 4/5 - 始终显示全部5颗星(例如,切勿只写
★★☆☆☆)★★ - 先显示整体评分部分,再显示维度细分
- 细分部分:每个维度占一行,所有选项在同一行
- 始终明确说明评分标准
Architecture Topology
架构拓扑
Use when the decision changes component layout or system topology.
┌──────────────────┐
│ Load Balancer │
└────────┬─────────┘
│
┌────┴────┐
▼ ▼
[API x2] [API x2]
│ │
└────┬────┘
▼
┌────────────┐
│ Postgres │
└────────────┘- Boxes use for infrastructure components
┌ ┐ └ ┘ ─ │ - for application components
[brackets] - Vertical flow primary; for horizontal connections
──→
当决策会改变组件布局或系统拓扑时使用。
┌──────────────────┐
│ Load Balancer │
└────────┬─────────┘
│
┌────┴────┐
▼ ▼
[API x2] [API x2]
│ │
└────┬────┘
▼
┌────────────┐
│ Postgres │
└────────────┘- 基础设施组件使用绘制方框
┌ ┐ └ ┘ ─ │ - 应用组件使用表示
[方括号] - 主要采用垂直流向;水平连接使用
──→
When to Add Diagrams
何时添加图表
- Star ratings: 3+ options × 2+ quantitative dimensions
- Topology: the decision changes the physical or logical component layout
- Do NOT add diagrams for simple yes/no or qualitative comparisons — use markdown tables
- Max width: 78 characters
- 星级评分:3个以上选项 × 2个以上量化维度
- 拓扑图:决策会改变物理或逻辑组件布局
- 简单的是/否或定性对比请勿添加图表——使用Markdown表格
- 最大宽度:78字符
Entry and Exit Criteria
进入与退出标准
Enter when:
- the decision node is explicit and bounded
- the decision has at least two meaningful alternatives
Exit when:
- there is a clear recommended option
- the consequences of that decision are clear enough to feed back into the design
进入条件:
- 决策节点明确且范围有限
- 决策至少有两个有意义的替代方案
退出条件:
- 有明确的推荐选项
- 该决策的影响足够清晰,可反馈到设计中
Handoff Rules
交接规则
- Hand back to when the chosen option affects unresolved design branches.
design-refinement - Hand to when this was the last major unresolved design issue.
design-readiness-check - Hand back to if routing should be reconsidered after the decision lands.
design-orchestrator - Do not force broad design discovery into a bounded decision just to make progress.
- Do not take over general design clarification work.
- 当所选选项影响未解决的设计分支时,交接回。
design-refinement - 当这是最后一个未解决的重大设计问题时,交接至。
design-readiness-check - 若决策落地后需要重新考虑路由,交接回。
design-orchestrator - 请勿为了推进进度而将宽泛的设计探索强行纳入有限范围决策。
- 请勿接管一般性的设计澄清工作。
Journal Integration
日志集成
When operating on a task tracked under , append a journal entry at this skill's milestone.
.agents/tasks/<task-id>/- Trigger: after a bounded decision is recommended and recorded
- Reserved key(s):
decision - Entry shape:
## <ISO8601-timestamp> — decision-evaluation decision: <decision-node-name> [optional ≤ 15-line body; longer content goes to artifacts/]
Resolve the task id from the explicit caller argument or . If neither resolves, skip the append; do not guess.
.agents/tasks/.currentSee for the full convention.
task-journal当处理下跟踪的任务时,在本Skill的里程碑处追加日志条目。
.agents/tasks/<task-id>/- 触发时机: 有限范围决策已被推荐并记录后
- 预留键:
decision - 条目格式:
## <ISO8601-timestamp> — decision-evaluation decision: <decision-node-name> [可选 ≤15行正文;更长内容存入artifacts/]
从明确的调用者参数或解析任务ID。若两者均无法解析,则跳过追加操作;请勿猜测。
.agents/tasks/.current有关完整规范,请参考。
task-journal