Knowledge Base Management with Deepfield

Overview

Deepfield is an AI-driven knowledge base builder designed for understanding brownfield projects. It provides a structured approach to documenting legacy codebases, facilitating team onboarding, and capturing institutional knowledge about complex systems.

When to use Deepfield:

• •Exploring unfamiliar legacy codebases
• •Onboarding new team members to existing projects
• •Documenting poorly documented systems
• •Understanding complex architectures
• •Modernizing brownfield applications

Core Workflow

The Deepfield workflow follows three foundational steps:

1. Initialize (df-init)

Create the workspace directory structure.

Command: /df-init

What it creates:

• •deepfield/source/ - Source materials and baseline snapshots
• •deepfield/wip/ - Work in progress (active exploration runs)
• •deepfield/drafts/ - Draft documents and notes
• •deepfield/output/ - Final knowledge base artifacts
• •Template files for configuration and documentation

When to use: First step when starting a new knowledge base project.

2. Configure (df-start)

Run interactive setup to define project context.

Command: /df-start

What it does:

• •Asks questions about project type and goals
• •Collects focus areas (architecture, APIs, security, etc.)
• •Creates project.config.json with settings
• •Pre-fills brief.md template with answers

What to provide:

• •Clear project name
• •Specific goal (e.g., "understand authentication flow")
• •Relevant focus areas
• •Honest assessment of project type

After configuration: Fill out deepfield/brief.md thoroughly. The quality of generated knowledge depends on the context provided in the brief.

3. Check Status (df-status)

View current state and get guidance on next steps.

Command: /df-status

What it shows:

• •Current workflow state
• •Project information (name, goal, focus)
• •Last modification time
• •Suggested next action

Use this command to:

• •Understand where you are in the workflow
• •Get reminded of next steps
• •Check project configuration
• •See exploration history (Phase 2+)

The Brief: Your Exploration Guide

The deepfield/brief.md file is the most important document in your knowledge base. It guides all AI exploration and learning.

Key sections to fill out:

Project Overview:

• •What the project does and why it exists
• •Main value proposition
• •Key stakeholders and users

Technical Overview:

• •Architecture type (monolith, microservices, etc.)
• •Core technologies (languages, frameworks, databases)
• •Critical dependencies (external services, APIs)

Pain Points:

• •Known issues and technical debt
• •Areas of confusion or poor documentation
• •Problematic code sections

Exploration Priorities:

• •What areas to investigate first
• •What questions need answers
• •What would make this knowledge base valuable

The more context you provide, the better the AI can explore and document your codebase.

Workflow States

Understanding workflow states helps you know what to do next:

Check your current state anytime with /df-status.

Directory Structure

Understanding the four-space architecture:

deepfield/
├── source/          # SOURCE: Raw materials
│   └── baseline/    # Initial state snapshots
│       └── repos/   # Repository copies
│
├── wip/            # WIP: Active work
│   └── run-N/      # Individual exploration runs
│
├── drafts/         # DRAFTS: Working documents
│   └── notes/      # Temporary notes and findings
│
└── output/         # OUTPUT: Final artifacts
    └── docs/       # Completed documentation

source/: Immutable baseline. Original code state for reference. wip/: Active exploration. Temporary state during runs. drafts/: Working space. Notes and draft documents. output/: Final results. Polished knowledge base artifacts.

Best Practices

When initializing:

• •Choose a clear, descriptive project name
• •Be specific about your goals
• •Select focus areas that matter most

When configuring:

• •Don't skip filling out the brief
• •Provide architectural context
• •List known pain points honestly
• •Include specific questions

During exploration (Phase 2+):

• •Let the AI work through runs systematically
• •Review generated knowledge incrementally
• •Update the brief as understanding grows

Common Use Cases

Legacy Codebase Understanding

Scenario: You've inherited a 10-year-old codebase with minimal documentation.

Approach:

1. •Run /df-init to create workspace
2. •Run /df-start, select "Legacy codebase (brownfield)"
3. •In brief.md, focus on:
   • •Known pain points (slow queries, unclear modules)
   • •What works vs what's mysterious
   • •Critical workflows to understand first

Team Onboarding

Scenario: New developers need to understand the system quickly.

Approach:

1. •Initialize with focus on "Team onboarding"
2. •In brief.md, emphasize:
   • •Entry points and key flows
   • •Common gotchas and pitfalls
   • •How to run and test locally

Code Modernization

Scenario: Planning to modernize an old system.

Approach:

1. •Select "Code modernization" type
2. •Focus on:
   • •Current architecture and dependencies
   • •Technical debt and outdated patterns
   • •Integration points that might break

Phase 1 Capabilities

Current (Phase 1):

• •✅ Initialize directory structure
• •✅ Interactive project configuration
• •✅ Status checking
• •✅ Template-based documentation framework

Coming (Phase 2+):

• •🔜 Autonomous codebase exploration
• •🔜 Incremental knowledge building
• •🔜 Change detection and updates
• •🔜 Domain decomposition
• •🔜 Learning accumulation

Phase 1 focuses on scaffolding and setup. The AI exploration capabilities come in subsequent phases.

Troubleshooting

"deepfield command not found"

• •Install CLI: npm install -g deepfield
• •Or use plugin's bundled version

"No deepfield/ directory found"

• •Run /df-init first to create structure

"Configuration is missing"

• •Run /df-start to set up project config

"Brief needs filling out"

• •Open deepfield/brief.md and add project context
• •Focus on pain points and priorities

Examples

See the examples/ directory for:

• •Complete workflow example (init → start → status)
• •Sample brief.md with good context
• •Common configuration patterns

This skill provides guidance on using Deepfield for AI-driven knowledge base building. For technical details about the CLI or plugin implementation, refer to project documentation.