Agent Factory

End-to-end automation for building and deploying paid Lucid Agents.

Composed Skills

This meta-skill orchestrates:

Prerequisites

Requires: Bun, Railway CLI, GitHub CLI (gh), Bird CLI (optional for X/Twitter).

export RAILWAY_TOKEN="<your-token>"
export PAYMENTS_RECEIVABLE_ADDRESS="<your-wallet>"
export AUTH_TOKEN="<twitter-auth-token>"  # optional
export CT0="<twitter-ct0>"  # optional

Pipeline Execution

Step 1: Discover Topic

Use skill: trend-discovery

Search for trending topics and evaluate monetization potential:

bird search "need API for" --limit 50

Output needed:

• Topic name and description
• Score ≥ 7/10
• Target audience
• Potential data sources

Step 2: Check Skills Market for Duplicates

Before building, review the skills-market repo to confirm you aren't recreating an agent that already exists.

1. List all existing plugins:

``bash ls plugins/ ``

1. For each plugin, read its description:

``bash for d in plugins/*/; do echo "=== $(basename $d) ==="; head -10 "$d/skills/SKILL.md" 2>/dev/null; echo; done ``

1. Compare your chosen topic against existing skills. Check for overlap in:

- Topic/domain — does a skill already cover this subject area? - Data sources — does an existing agent already consume the same APIs? - Endpoints — would your planned endpoints duplicate what's already deployed?

If a match is found:

• If the existing skill fully covers your topic → go back to Step 1 and pick a different topic.
• If there's partial overlap → narrow your agent's scope to cover only the gap, or pick a different angle that adds unique value.

Output needed:

• List of existing skills reviewed
• Confirmation that no duplicate exists (or description of how your agent differs)

Step 3: Research APIs

Use skill: api-research

Find and validate live data sources:

web_search "<topic> free API"
curl -s "<api_endpoint>" | head -c 500

Output needed:

• 1+ validated API endpoints
• 6 endpoint designs (1 free + 5 paid)
• Data mapping

Step 4: Build Agent

Use skill: lucid-agents-sdk

Create the agent with 6 endpoints (1 free + 5 paid). Test locally before deploying.

Output needed:

• All 6 endpoints tested
• All return real data
• All return status: "succeeded"

Step 5: Deploy

Use skill: railway-deploy

Ship to production:

gh repo create <user>/<agent-name> --public --source=. --push
railway add -s <agent-name> -v "PAYMENTS_RECEIVABLE_ADDRESS=..."
railway up --detach --service <agent-name>
railway domain --service <agent-name>

Output needed:

• GitHub URL
• Railway URL
• Health check passing

Step 6: Announce

Update portfolio and compose tweet:

🚀 Just shipped: <Agent Name>

<One-line description>

✅ 1 free endpoint
💰 5 paid via x402

Built with @daydreamsagents

<url>

Complete Checklist

Discovery

• Topic selected with score ≥ 7
• Target audience identified
• Data sources proposed

Dedup Check

• Existing skills-market plugins reviewed
• No topic/API/endpoint overlap with existing skills

Research

• API endpoints validated (curl tested)
• 6 endpoints designed
• No auth required (or free tier available)

Build

• package.json with Zod v4
• .gitignore includes node_modules/
• All 6 endpoints implemented
• Local tests pass
• Real data returned (no hardcoded)

Deploy

• GitHub repo created
• Railway service created
• Env vars set (3 required)
• Domain configured
• Health check passes
• Live endpoints return data

Announce

• Portfolio updated
• Tweet composed

Error Recovery

Example Run

Topic: Space Weather
Score: 8.6/10
APIs: NOAA SWPC (free, no auth)
Endpoints:
  - overview (free): Current conditions
  - kp-index ($0.001): Geomagnetic index
  - solar-cycle ($0.002): Sunspot data
  - alerts ($0.002): Active warnings
  - aurora ($0.003): Visibility forecast
  - report ($0.005): Full analysis

Result:
  GitHub: github.com/user/solar-storm-agent
  Railway: solar-storm-agent-production.up.railway.app
  Status: LIVE ✅

Scheduling

Run automatically via cron:

0 * * * *  # Every hour

Each run:

1. Picks NEW topic (avoid repeats)
2. Builds and deploys agent
3. Reports results to operator