Design Ia

Information Architecture Auto-Design

Core Principles

1. Information finds the path, not the path finds information: IA design starts from user information needs, not from feature lists
2. Layer restraint: 3 levels max, 3-7 items per category, following Miller's Law
3. Batch generation, human filtering: AI generates classification proposals in bulk, humans make final selection and judgment
4. Validation-driven: Key classification nodes must be marked for user validation

AI->Human AI suggests -> Human approves

Interaction Mode

AI->Human AI suggests, human approves

Input

| Input Item | Type | Required | Source | Description | |------------|------|----------|--------|-------------| | PRD | markdown | Yes | output/pm-design/design-prd/prd.md | Product requirements document | | PRD Structured Data | JSON | O | output/pm-design/design-prd/prd.json | Machine-consumable PRD version containing pages[], for IA design to align page routes | | Existing Product IA | JSON | O | User provided | Existing information architecture (if any) | | User Research Data | JSON | O | output/pm-discovery/user-research-voice-analysis / output/pm-discovery/user-research-behavior-analysis | User behavior patterns, content preferences |

Execution Steps

Step 1: Content Inventory Generation [Core]

Extract all feature points and content items from PRD:

• Feature module inventory
• Content type list
• Core business entities
• User-reachable information nodes

Step 2: Auto-Classification [Core]

Generate classification suggestions based on semantic similarity:

AI generates classification suggestions based on semantic similarity of feature names and descriptions:

1. Extract core keywords for each feature point
2. Group by keyword semantic proximity
3. Check if each group quantity meets the 3-7 item constraint
4. Suggest splitting or merging for groups exceeding the constraint
5. Mark classification confidence, low-confidence groups marked with needshumanvalidation

• Constraints:
• Existing classifications are preserved first
• Each category contains 3-7 items
• Hierarchy does not exceed 3 levels

Step 3: Navigation Needs Definition [Core]

Based on content characteristics and user scenarios, define navigation needs (do not define specific navigation patterns, UI Skill decides implementation):

| Content Characteristic | Navigation Need | |-----------------------|-----------------| | Flat structure | 3-5 peer entries need to be simultaneously visible | | Clear hierarchy | Depth 4 levels | Mark "Hierarchy too deep", suggest flattening | | Same-level node count > 7 | Mark "Cognitive overload", suggest grouping | | Key task path click count > 3 | Mark "Path too deep", suggest elevating level | | IA confidence < 0.5 | Escalate to human validation, mark "Low IA inference reliability" | | Mismatch with PRD feature modules | Mark "Feature coverage gap", list uncovered feature modules | | Navigation path has cycles | Must fix, eliminate circular references |

Quality Checks

P0 Checks (must pass for quick/standard/deep)

• [ ] Content inventory completeness (Covers all PRD feature points)
• [ ] Classification reasonableness (Follows Miller's Law (3-7 items per category))

P1 Checks (must pass for standard/deep)

• [ ] Navigation needs definition (Has clear navigation needs matching content characteristics)
• [ ] Validation node marking (All key classification nodes marked for user validation)

P2 Checks (must pass for deep only)

• [ ] Extended analysis complete (deep simulation and roadmap generated)
• [ ] Decision records complete (key decisions have rationale and alternatives)

---

Degradation Strategy

| Missing Upstream Input | Degradation Plan | Output Impact | Data Acquisition Instructions | |------------------------|-----------------|---------------|-------------------------------| | PRD document missing | User provides feature list, design IA directly | Lacks PRD structured data, classification may be less complete | Request user to provide feature list and product description, or upload prd.json file | | Existing IA data missing | Design IA from scratch, no reference baseline | Lacks existing IA reference, may miss established structures | Request user to describe current page structure or navigation hierarchy | | User research data missing | Derive classification based on PRD features | Lacks user research data, classification may deviate from user mental model | Request user to provide user research conclusions or upload persona.json file | | All upstream files missing | Prompt user to execute prior stages first, or design IA based on user-provided feature list | Overall confidence reduced | Request user to provide feature list, target users, and core usage scenarios |

Output Validation Rules

| Field Path | Type | Required | Description | |------------|------|----------|-------------| | iaproposals | array | Yes | IA candidate proposal list, at least 2 | | iaproposals[].name | string | Yes | Proposal name | | iaproposals[].structure | object | Yes | Hierarchy structure definition | | iaproposals[].navigationneeds | string | Yes | Navigation needs description (does not define specific navigation pattern) | | iaproposals[].routes | array | Yes | Route list | | iaproposals[].routes[].path | string | Yes | Route path (must be consistent with prd.json.pages[].route) | | iaproposals[].routes[].page | string | Yes | Page name | | iaproposals[].routes[].depth | integer | Yes | Hierarchy depth | | iaproposals[].avgclickstocore | number | Yes | Average clicks to core features | | iaproposals[].alignmentwithusermodel | string | Yes | Alignment with user mental model | | iaproposals[].needsuservalidation | array | Yes | Nodes requiring user validation |

Upstream Change Response

Upstream Change Impact

| Upstream Change | Impact Scope | Response Strategy | |-----------------|-------------|-------------------| | PRD feature module addition/removal | Content inventory, classification structure, route definition | Mark affected feature points and classification nodes, suggest human confirmation on whether to re-cluster | | PRD priority adjustment | IA hierarchy structure, navigation patterns | Mark affected hierarchy relationships, suggest human confirmation on whether to adjust depth | | User research data update | Classification proposals, user mental model alignment | Mark affected classification judgments, suggest human confirmation on whether to adjust classification proposals | | Existing IA structure adjustment | Route mapping, navigation patterns | Mark affected routes and navigation, suggest human confirmation on whether to redesign |

Downstream Notification Mechanism

| IA Change Type | Notification Scope | Notification Method | |---------------|-------------------|---------------------| | Route structure change | design-userflow, design-prototype, design-handoff-spec | Mark route changes, trigger user flow and prototype redesign | | Navigation needs change | design-prototype, interaction-spec | Mark navigation needs changes, trigger prototype and interaction spec updates | | Hierarchy depth change | design-userflow, design-handoff-spec | Mark hierarchy changes, trigger flow and handoff document updates | | Classification node change | design-userflow, design-prototype | Mark classification changes, trigger flow and prototype updates |

Alignment with prd.json Data Contract

| This Skill's Output Field | prd.json Corresponding Field | Alignment Rule | |--------------------------|----------------------------|----------------| | iaproposals[].routes[].path | prd.json.pages[].route | Route paths must be consistent; after IA proposal is confirmed, prd.json is updated synchronously | | iaproposals[].routes[].page | prd.json.pages[].name | Page names must be consistent | | ia_proposals[].structure | prd.json.pages[] hierarchy | IA hierarchy determines parent-child relationships of pages |

Data Acquisition Instructions

This Skill requires PRD, existing IA, and user research data. Please provide via one of the following methods:

1. Directly describe feature list and user needs
2. Upload PRD document / persona.json / voice-analysis.json files
3. Provide data file paths

• AI is not responsible for external data collection, only for analysis

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: LuckyOneTwoThree
• Source: LuckyOneTwoThree/vibe-skill
• License: MIT
• Homepage: https://luckyonetwothree.github.io/all-skill-html/

Install and usage instructions live in the source repository linked above.