SEEDING.md

SkillX Database Seeding

Quick reference for seeding the SkillX database with AI agent skills.

Prerequisites

1. Set ADMIN_SECRET environment variable
2. Ensure web app is running (local) or deployed (production)

Commands

Local Development

# Start dev server
pnpm dev

# Seed database (in another terminal)
ADMIN_SECRET=your-secret-here pnpm seed

Production

API_URL=https://skillx.sh ADMIN_SECRET=your-prod-secret pnpm seed

What Gets Seeded

• 30 curated AI agent skills
• 5 categories: devops, implementation, testing, security, planning
• 23 free + 7 paid skills ($24.99-$79.99)
• ~127 vectors (avg 4.2 chunks per skill)

Pipeline

seed-skills.mjs
  ↓
POST /api/admin/seed (X-Admin-Secret: xxx)
  ↓
For each skill:
  1. Upsert to D1 database (Drizzle ORM)
  2. Chunk text (512 tokens, 10% overlap)
  3. Generate embeddings (Workers AI)
  4. Index vectors (Vectorize)

Response

{
  "skills": 30,
  "vectors": 127
}

Idempotency

Safe to run multiple times:

• D1 upserts on slug conflict
• Vectorize replaces vectors by ID

Files

• scripts/seed-data.json - 30 curated skills
• scripts/seed-skills.mjs - Seed script
• scripts/README.md - Full documentation

Troubleshooting

401 Unauthorized

• Check ADMIN_SECRET matches wrangler.jsonc

Connection refused

• Ensure pnpm dev is running (local)
• Check API_URL is correct (production)

No data returned from AI model

• Workers AI binding not configured
• Check wrangler.jsonc has ai.binding = "AI"

For detailed docs, see scripts/README.md