decision-evaluation（决策评估）

Decision Evaluation

Overview

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.

When to Use

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

Constraint Gathering

When the decision involves technology selection (language, framework, library, or tooling), perform this step before the formal comparison.

Ask the user to confirm:

1. Hard constraints: existing tech stack, license requirements, budget limits, deployment environment restrictions.
2. Preferences or already-excluded options.
3. Priority of evaluation dimensions (e.g., performance > ecosystem > learning curve).
4. 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.

Core Responsibilities

Your responsibilities are:

1. Frame the exact decision question. For technology selections, frame the question based on constraints confirmed in the Constraint Gathering step.
2. Keep the work bounded to one explicit decision node.
3. Present 3 to 4 real options when real options exist.
4. Compare trade-offs clearly and concretely.
5. Recommend one option or a justified combination when appropriate.
6. Explain why the rejected options are not preferred in this context.
7. Record the result in a reusable decision record.

Comparison Standard

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.

Expected Outputs

Produce or update a design_state that includes:

• design_target_type
• decision_nodes
• decisions
• risks
• optional downstream impact notes for the design tree

Diagram Conventions

Use character diagrams inside code blocks (no language tag) to visualize trade-offs and architecture topology.

Star Ratings

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

Architecture Topology

Use when the decision changes component layout or system topology.

┌──────────────────┐
│  Load Balancer   │
└────────┬─────────┘
         │
    ┌────┴────┐
    ▼         ▼
 [API x2] [API x2]
    │         │
    └────┬────┘
         ▼
  ┌────────────┐
  │  Postgres  │
  └────────────┘

• Boxes use ┌ ┐ └ ┘ ─ │ for infrastructure components
• [brackets] for application components
• Vertical flow primary; ──→ for horizontal connections

When to Add Diagrams

• 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

Entry and Exit Criteria

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

Handoff Rules

• Hand back to design-refinement when the chosen option affects unresolved design branches.
• Hand to design-readiness-check when this was the last major unresolved design issue.
• Hand back to design-orchestrator if routing should be reconsidered after the decision lands.
• Do not force broad design discovery into a bounded decision just to make progress.
• Do not take over general design clarification work.