| name | build-feature-tree |
| description | WHEN/WHERE/WHO: [Scheduling: Analysts or agents reviewing flat product backlogs or lengthy PRDs.] HOW: [Structural: Use this SKILL to group unstructured requirements into a hierarchical Feature Tree (L1, L2, L3) rendered as a mindmap.] WHY: [Scheduling: Flat lists hide missing features. A visual hierarchy exposes gaps.]
|
Build Feature Tree
Who
Business Analysts, Product Managers, and AI Agents. The agent uses this skill to restructure and visualize scope.
What
This skill implements the Feature Tree from the Requirements Modeling Language (RML). It takes a flat list of features or user stories and categorizes them into a strict hierarchy:
- L1 (Level 1): Major functional areas (e.g., "Account Management").
- L2 (Level 2): Feature groups (e.g., "User Profile").
- L3 (Level 3): Specific features (e.g., "Update Avatar").
It outputs this structure as a Mermaid.js Mindmap.
When
Invoke this skill during scope definition or when analyzing a lengthy PRD. Trigger phrases include: "build a feature tree," "organize this backlog," "create a mindmap of features," or "check for missing features."
Where
Applies to Product Requirements Documents (PRDs), JIRA backlogs, and feature lists.
Why
A backlog with 150 items is impossible to review for completeness. By organizing features hierarchically on a single page, the human brain can instantly spot missing siblings (e.g., if L2 "Reporting" only has L3 "Export to CSV", someone will ask "What about Export to PDF?").
Inputs
- A flat list of features, requirements, or user stories.
- The overarching Product Concept (the root node).
Output (Logical Evidence)
- A hierarchical list of features.
- A Mermaid.js Mindmap rendering the tree.
- A list of "Potential Missing Features" discovered by analyzing the visual gaps.
- A short "Visualization Quality Notes" section documenting readability and labeling decisions.
Optimization Readiness
- Failure Signals: Technical tasks are mistaken for business features, hierarchy depth becomes unreadable, labels stay verbose, or missing-feature suggestions are arbitrary rather than driven by structural gaps.
- Evidence To Collect: Normalized feature lists, generated trees, visualization notes, and examples of sparse or asymmetric branches that led to meaningful gap discovery.
- Safe Mutation Boundaries: Refine normalization rules, grouping heuristics, Mermaid guidance, and readability checks without changing the core business-feature hierarchy workflow.
- Acceptance Criteria: Accept revisions only if the output stays feature-focused, readable in text form, structurally coherent, and useful for spotting missing siblings or scope gaps.
- Rejected Revision Handling: Record poor grouping patterns, over-deep trees, and misleading visualization choices so they are not repeated.
- Transfer Check: Verify the workflow still works for PRDs, backlogs, and story collections with or without Mermaid rendering available.
- Stop Rule: If the input is primarily technical implementation work rather than business functionality, stop and redirect before drawing the tree.
Constraints (Logical Boundaries)
- Features should be brief noun phrases (e.g., "Shopping Cart," not "The system shall allow the user to add items to a cart").
- Limit the depth to 3 or 4 levels to maintain readability.
- Keep one primary message per rendered mindmap (scope structure), and avoid decorative styling that does not improve interpretation.
- If node coloring is used, limit category colors to a small, distinguishable set and do not rely on color alone to convey hierarchy.
- Prefer explicit node labels and structural grouping over legend-heavy color coding.
- Keep execution guidance text-first; the feature tree can be represented in plain structured lists when diagrams are unavailable.
- Do not require Mermaid rendering to complete analysis or validation.
One More Thing
If the input text is not a list of features but rather a list of technical tasks (e.g., "Setup database," "Configure DNS"), stop and inform the user that a Feature Tree models business functionality, not technical implementation.
How (Structural Workflow)
Phase 1: Extraction and Normalization
- Read the input and extract all functional requirements.
- Convert long sentences into concise 1-3 word noun phrases.
Phase 2: Hierarchical Grouping
- Identify the L1 Features (Major functional areas, usually 3-7 max).
- Group the remaining features under the appropriate L1 buckets to form L2 Features.
- Break down complex L2 features into L3 Features if necessary.
Phase 3: Visualization and Analysis
- Output the structure using Mermaid.js Mindmap syntax.
- Review the tree for symmetry and completeness. If an L2 category feels sparse compared to its siblings, suggest missing L3 features.
Phase 4: Visualization Quality Pass
- Check visual density and simplify if crowded (split into multiple trees if needed).
- Ensure labels remain concise and readable without requiring color interpretation.
- Add "Visualization Quality Notes" summarizing any compromises (for example: split by domain, grouped sparse branches).
Resources
Validation
- Verify that the Mermaid syntax is valid (
mindmap format).
- Ensure no feature from the original list was dropped; every item must have a home in the tree.
- Verify the tree remains readable at the expected viewing size and does not depend on color-only interpretation.
