knowledge-builder

name	knowledge-builder
description	Domain knowledge builder skill - triggered when working with unfamiliar domains, researching new technologies, or when the user asks to build knowledge about a topic. Researches deeply from the web and official docs, organizes into structured knowledge files, supports importing documentation files, and distinguishes between GLOBAL knowledge (cross-project) and PROJECT-specific knowledge.
allowed-tools	Read, Write, Edit, Glob, Grep, Bash, WebFetch, WebSearch, Agent

Knowledge Builder Skill

You are building domain-specific knowledge. Knowledge has TWO scopes:

Scope	Location	When to Use
Global	.claude/knowledge/domains/<topic>.md	Reusable across ALL projects
Project	projects/<project>/knowledge/<topic>/	Specific to ONE project (client-specific)

When This Skill Activates

• User runs /knowledge build, /knowledge update, /knowledge add
• User asks to learn about a new domain (robotics, PLC, medical, finance, etc.)
• User provides documentation files (PDF, URLs) to import
• During /design when the domain is unfamiliar

Scope Selection (IMPORTANT)

When building knowledge, ALWAYS determine the scope:

1. If the user passed --global or --project: use that.
2. Otherwise, ASK:

   Where should I save this knowledge?
     1. Global   - For ALL projects (standard tech/protocols like HL7, OPC-UA, JWT)
     2. Project  - For this project only (client-specific APIs, internal systems)
   

3. Defaults if no active project: Global.
4. Heuristics to suggest the right scope:
   • Names containing "client", "legacy", "internal", "this-project" → Project
   • Standard protocols, industry standards, well-known frameworks → Global

Knowledge Building Process

Step 1: Research Deeply

For any new topic, run multiple searches:

1. WebSearch "<topic> official documentation"
2. WebSearch "<topic> programming guide tutorial"
3. WebSearch "<topic> architecture overview"
4. WebSearch "<topic> API reference"
5. WebSearch "<topic> integration with other systems"
6. WebSearch "<topic> best practices 2026"
7. WebSearch "<topic> common problems solutions"

Step 2: Read Official Docs

For each important URL found:

1. WebFetch <url> - read the page
2. Extract: concepts, APIs, protocols, data formats
3. Note: code examples, configuration patterns
4. Identify: integration points, constraints, security

Step 3: Organize Knowledge

Structure into the standard format:

• Overview (what is it, why it matters)
• Key Concepts (glossary with domain-specific terms)
• Architecture (how it works, components, flow) — include a Mermaid diagram
• Programming (languages, tools, APIs, code examples)
• Communication Protocols (how systems talk)
• Data Formats (what data looks like)
• Integration Patterns (how to connect with our system)
• Constraints & Limitations
• Security Considerations
• Common Mistakes

Step 4: Save at Correct Scope

Global scope:

.claude/knowledge/domains/<topic>.md

Project scope:

projects/<project>/knowledge/<topic>/
  <topic>.md              # Structured knowledge (auto-generated)
  docs/                   # User-provided documentation
    <file>.pdf            # Original PDFs, datasheets
    <file>.md             # API docs, notes
  urls.md                 # Bookmarked URLs for reference

Always include the scope in the knowledge file's frontmatter:

# Knowledge: [Topic]

**Built**: [date]
**Scope**: Global | Project (<project-name>)
**Sources**: [list of URLs consulted]

Step 5: Render in Chat

After saving, draw any diagrams IN THE CHAT as ASCII / Unicode box-drawing (see Rule 20). Mermaid stays in the .md file. so the user can see the architecture without opening files.

Step 6: Import User-Provided Docs

When user provides documentation files:

• PDF files: Read, extract text, organize
• URLs: WebFetch, extract relevant technical content
• Manual notes: Append to knowledge under "Manual Notes" section
• For Project scope: save in projects/<project>/knowledge/<topic>/docs/
• For Global scope: reference URLs in the knowledge file (don't copy large PDFs globally)

Documentation Import (project scope)

projects/<project>/
  knowledge/
    <topic>/
      <topic>.md              # Structured knowledge (auto-generated)
      docs/                   # User-provided documentation
        kuka-krl-manual.pdf   # PDF manuals
        plc-datasheet.pdf     # Datasheets
        api-reference.md      # API docs
        notes.md              # User notes
      urls.md                 # Bookmarked URLs for reference

When user says "here's the documentation for X" or provides a file:

1. Determine scope (project is usually right for user-provided docs)
2. Save/copy to the right folder
3. Read the file content
4. Extract key information
5. Update the topic's knowledge file with new findings
6. Add the source to the knowledge file's "Sources" list

Knowledge Usage

Once knowledge exists, it's used automatically:

• During /design: architecture decisions reference domain knowledge (project-specific first, then global)
• During /implement: use APIs and code patterns from knowledge
• During /schema: domain-specific data structures
• During discovery: answer domain questions from knowledge
• During planning: include domain constraints in plans

Promote / Copy Operations

Promote (project → global): when you built something as project-specific but realize it's reusable, run /knowledge promote <topic>. The skill will:

1. Copy the file to .claude/knowledge/domains/
2. Update Scope to Global
3. Ask the user whether to remove from project

Copy (global → project): when you want to customize a global file for a specific project, run /knowledge copy <topic> --to <project>. The skill will:

1. Copy global file into project folder
2. Update Scope to Project
3. User can edit freely without affecting global

Mermaid Diagrams in Knowledge

Generate diagrams to visualize domain concepts. Always use proper formatting with a title comment and clear node labels:

%% Industrial Automation - Communication Architecture
graph TB
    subgraph Field["Field Layer"]
        Sensors[IoT Sensors]
        Robots[KUKA Robots]
    end
    subgraph Control["Control Layer"]
        PLC[Siemens S7 PLC]
        SCADA[SCADA System]
    end
    subgraph Enterprise["Enterprise Layer"]
        MES[MES System]
        ERP[ERP System]
    end

    Sensors -->|MQTT| PLC
    Robots -->|EtherNet/IP| PLC
    PLC -->|OPC-UA| SCADA
    SCADA -->|REST API| MES
    MES -->|Message Queue| ERP

Render these IN THE CHAT as well as saving to the file.