seed

Seed

Auto-generate KB docs from your existing codebase. The cold start killer.

Purpose

The #1 reason agentic engineering fails is an empty knowledge base. Engineers are told "write docs for the KB" and either procrastinate or write docs so generic they're useless.

This skill analyzes your codebase and generates draft KB documents that capture how your team actually builds software — extracted from real code, not imagination. Senior engineers review and refine the drafts, starting from 80% instead of zero.

Usage

/seed ~/code/my-app

Or run from within a repo:

/seed .

Process

Step 1: Analyze Codebase Structure

Scan the repo to understand what's here:

- Language(s) and framework(s) (detect from package.json, *.csproj, pyproject.toml, go.mod, etc.)
- Directory structure and organization conventions
- Number and types of source files
- Test location and naming patterns
- Config files and their purposes

Output: Brief summary to the user:

Found: TypeScript/React frontend (src/app/), Python/FastAPI backend (src/api/),
       847 source files, 234 tests, PostgreSQL migrations in db/migrations/

Step 2: Detect Patterns

For each major area, analyze 5-10 representative files to extract patterns:

API/Routes:

• Read 5 route handlers → extract common patterns (validation, error handling, response shape)
• Read API tests → extract testing patterns
• Note: endpoint naming, middleware usage, auth patterns

Components/Pages (frontend):

• Read 5 page components → extract layout patterns, data fetching, state management
• Read 5 shared components → extract prop patterns, composition style
• Note: styling approach (CSS modules, Tailwind, styled-components)

Data Layer:

• Read models/schemas → extract naming conventions, relationship patterns
• Read migrations → extract migration patterns
• Read database access code → extract query patterns (ORM style, raw SQL, etc.)

Testing:

• Read 5 test files → extract testing patterns (setup, assertions, mocking)
• Note: test runner, assertion library, fixture patterns