money-os

Security Screener — Development Roadmap

Version: 1.0 Date: 2026-03-22 Parent system: Money OS v3.1+

---

Overview

Three phases. Each produces something usable. The first phase delivers the core value (portfolio-aware scanning) in 3 weeks. Phase 2 adds crypto and fundamentals. Phase 3 adds the intelligence that makes this more than a chart tool.

Phase 1 │████████│ Core Screening Engine         Weeks 1–3
Phase 2 │████████│ Crypto + Fundamentals         Weeks 4–6
Phase 3 │████████│ Portfolio Intelligence Layer   Weeks 7–9
Phase 4 │░░░░░░░░│ Future Enhancements           Weeks 10+

Difference from Lighthouse’s roadmap: Lighthouse builds a full web app with 9 tabs over 14 weeks. We build a data service + Claude skill in 9 weeks. No Kanban board, no notes app, no hourly bars, no human drawing tools. Every week ships something that makes /screen smarter.

---

Phase 1 — Core Screening Engine

Duration: Weeks 1–3 Goal: Scanner finds stocks near trendlines, Claude presents them with portfolio context

This phase is the minimum viable loop: data in → trendlines computed → scan results out → Claude filters by portfolio.

Milestone 1.1 — Data Service Scaffold (Week 1)

• Initialize Next.js 15 project with TypeScript
• Configure Supabase PostgreSQL + Drizzle ORM
• Define schema: bars, trendlines, scan_results, trading_signals, watched_tickers, pipeline_status
• Run migrations
• Configure Vercel deployment + environment variables
• Implement Alpaca API client (lib/fetchers/alpaca.ts)
  • fetchBars(ticker, timeframe, from, to) with pagination
• Build fetchDailyBars cron job — iterate watched tickers, upsert to bars
• Build fetchWeeklyBars cron job
• Build fetchVix cron job
• Seed watched_tickers with S&P 500
• Configure Vercel Cron in vercel.json
• Build GET /api/pipeline/status endpoint

Deliverable: Data service deployed, market data flowing daily, pipeline status visible.

---

Milestone 1.2 — Trendline + Scanner Engines (Week 2)

• Implement lib/indicators/atr.ts — ATR(14)
• Implement lib/indicators/rsi.ts — RSI(14)
• Implement lib/indicators/macd.ts — MACD(12,26,9)
• Implement lib/indicators/zigzag.ts — pivot detection
• Write unit tests for all indicators
• Implement lib/engine/trendlineEngine.ts:
  • Zigzag pivot detection → line candidates → scoring → top-K → classify S/R
• Build runTrendlineEngine cron job (processes all watched tickers, weekly + daily)
• Implement lib/engine/scannerEngine.ts:
  • VIX-adjusted ATR zones (Entry/Alert)
  • Intersection detection (weekly + daily convergence)
• Build runScannerEngine cron job (runs after trendline engine)
• Build API endpoints:
  • GET /api/bars?ticker=&timeframe=
  • GET /api/trendlines?ticker=
  • GET /api/sr-levels?ticker=
  • GET /api/scanner?filter=
  • GET /api/vix

Deliverable: Trendlines and scan results computed daily for all watched tickers. API serving results.

---

Milestone 1.3 — Claude Skill + Portfolio Integration (Week 3)

• Create skills/security-screener/SKILL.md with:
  • /screen command: fetch scan results → read profile → apply portfolio filter → format output
  • /watchlist command: fetch approaching-zone stocks → overlay portfolio context
  • /signals command: fetch recent signals → prioritize by portfolio relevance
• Create commands/screen.md, commands/watchlist-scan.md, commands/signals.md
• Implement portfolio filtering logic in skill instructions:
  • Sector allocation gap scoring
  • Position sizing (1-3% of portfolio)
  • Concentration warnings (sector >5% over target)
• Implement lib/engine/signalEngine.ts:
  • RSI oversold/overbought
  • MACD crossover
  • Divergence detection
  • Proximity signal
• Build runSignalGenerator cron job
• Build GET /api/signals?limit=50&ticker= endpoint
• Build POST /api/universe/sync endpoint (accepts ticker list from Claude)
• Integration test: full flow from /screen command to portfolio-filtered output
• Update commands/money-os.md router to include new commands

Deliverable: User can say /screen and get portfolio-aware investment candidates. Full loop working.

---

Phase 2 — Crypto + Fundamentals

Duration: Weeks 4–6 Goal: Crypto assets in the scanner, fundamental data enriching every recommendation

Milestone 2.1 — Crypto Support (Week 4)

• Implement CoinGecko API client (lib/fetchers/coingecko.ts)
  • fetchDailyBars(coinId) — OHLCV for top 20 coins
  • fetchWeeklyBars(coinId) — aggregate from daily
• Build cryptoBarsFetcher cron job (daily midnight UTC)
• Seed watched_tickers with top 20 crypto by market cap
• Extend trendline engine to handle crypto tickers (24/7 market, UTC daily close)
• Extend scanner engine for crypto (no VIX adjustment for crypto — use BTC volatility index or fixed factor)
• Update skill to handle crypto in /screen output (no P/E for crypto, show network metrics where available)

Deliverable: /screen returns crypto candidates alongside stocks.

---

Milestone 2.2 — Fundamental Data (Week 5)

• Implement FMP API client (lib/fetchers/fmp.ts)
  • fetchFundamentals(ticker) — quarterly key ratios
  • fetchEarningsCalendar(days) — upcoming earnings
• Create fundamentals table, run migration
• Build fundamentalsFetcher cron job (daily 6:30 AM)
• Build earningsFetcher cron job (daily 6:00 AM)
• Extend scan_results to include fundamental overlay (P/E, rev growth, sector)
• Build API endpoints:
  • GET /api/fundamentals?ticker=
  • GET /api/earnings?days=7
• Update scanner engine to attach fundamental data to scan results

Deliverable: Scan results include fundamental quality metrics. Earnings calendar available.

---

Milestone 2.3 — Earnings + Sectors (Week 6)

• Create skills/security-screener/ reference docs for earnings interpretation
• Add /earnings command to skill:
  • Fetch earnings calendar → enrich with trendline proximity → filter by portfolio
  • Flag stocks the user owns that are reporting
  • Note trendline levels as “lines in the sand” for earnings reactions
• Add /sectors command to skill:
  • Fetch sector performance from category price data
  • Map to user’s allocation targets
  • Generate rotation signals (lagging + underweight = opportunity)
• Create sectors and sector_tickers tables
• Seed with 10 default sectors (same as Lighthouse’s categories but with allocation context)
• Build GET /api/sectors and GET /api/sectors/:slug/prices
• Update commands/money-os.md router

Deliverable: Full feature set live: /screen, /watchlist, /signals, /earnings, /sectors.

---

Phase 3 — Portfolio Intelligence Layer

Duration: Weeks 7–9 Goal: The screener becomes truly personal — it learns your thesis, suggests based on conviction, and connects to the full Money OS skill chain

Milestone 3.1 — Thesis-Aware Screening (Week 7)

• Extend /screen to accept thesis context:
  • If profile/history.md contains a prior /thesis-to-trades analysis, use its categories and conviction levels to weight scan results
  • “Aligned with thesis” filter: only show candidates that map to thesis categories
  • Score candidates by thesis alignment × technical quality × fundamental quality × portfolio fit
• Build thesis-ticker mapping: when a thesis names sectors/themes, map to specific tickers in the universe
• Add “thesis score” to /screen output (how well each candidate fits the stated thesis)

Deliverable: /screen results are weighted by thesis conviction, not just technical proximity.

---

Milestone 3.2 — Tax-Aware Recommendations (Week 8)

• Extend /screen position sizing to consider funding sources:
  • When recommending a buy, suggest which existing position to trim
  • Check holding periods from profile/holdings.md — flag short-term vs long-term gains
  • Estimate tax impact: “Trimming MSFT (long-term, 15% rate) frees $2,400. Trimming AMZN (short-term, 32% rate) costs $380 more in taxes.”
• Integrate with /rebalance skill: /screen output includes a “generate trade plan” CTA that feeds directly into rebalancing
• Add wash-sale detection: if user recently sold a ticker at a loss, warn before recommending re-entry within 30 days

Deliverable: Recommendations include tax-optimized funding sources and wash-sale warnings.

---

Milestone 3.3 — Quality Scoring + Backtesting Foundation (Week 9)

• Implement composite quality score for scan results:
  • Technical score (0-100): zone proximity + signal confluence + trendline touches
  • Fundamental score (0-100): valuation vs sector average + growth + margins
  • Portfolio score (0-100): fills allocation gap + thesis alignment + diversification benefit
  • Overall score = weighted average (configurable weights)
• Build historical signal tracking:
  • When a signal fires, record the outcome 5/10/20 days later
  • Store in signal_outcomes table
  • Begin accumulating win-rate data per signal type
• Add signal quality metadata to /signals output:
  • “RSI oversold signals on daily have led to 5-day bounces 68% of the time in your universe”
• Build GET /api/signals/performance endpoint

Deliverable: Scored recommendations with early backtesting data. Foundation for Money OS M5 (Learning System).

---

Phase 4 — Future Enhancements

Timeline: Post Week 9 (backlog, prioritized by user demand)

4.1 Optional Web Dashboard

• Minimal chart viewer for scan results (Lightweight Charts)
• Scanner results table with sorting/filtering
• Not a full Lighthouse clone — just a visual companion to the Claude skill
• Single-page, read-only, pulls from same API

4.2 Alert System

• Email digest: daily summary of entry zone stocks + portfolio-relevant signals
• Telegram bot: real-time alerts for high-conviction entry signals
• Alert preferences stored in user profile

4.3 Advanced Patterns

• W-pattern detection (from Lighthouse — deferred from Phase 1 for simplicity)
• Volume profile analysis
• Relative strength vs. sector (is the stock leading or lagging its peers?)

4.4 Expanded Asset Coverage

• International ADRs (emerging market exposure)
• Bond ETFs (for fixed income allocation)
• Commodity ETFs (gold, oil, agriculture)
• REITs (real estate exposure)

4.5 AI Enhancements

• LLM-generated chart commentary (describe what the chart “looks like” in plain language)
• Earnings preview briefs (pre-earnings research summary from web search)
• Pattern recognition confidence scoring

---

Sprint Structure

Each sprint = 1 week. Single developer.

Mon:    Planning + schema/API design
Tue-Wed: Backend (fetchers, engines, cron jobs)
Thu:     Skill writing + integration
Fri:     Testing + deployment + skill validation with real data

---

Definition of Done

A milestone is done when:

1. All API endpoints return correct data (tested with curl)
2. Cron jobs execute on schedule and pipeline status shows green
3. Claude skill produces correct, portfolio-aware output for test scenarios
4. No TypeScript errors
5. Deployed to Vercel
6. End-to-end: user says /screen → gets relevant, filtered results with position sizing

---

Risk Register

Risk	Likelihood	Impact	Mitigation
Alpaca API rate limits (200 req/min)	High	High	Batch requests, stagger fetches, use bulk endpoints where available
CoinGecko free tier limits (30 req/min)	Medium	Medium	Cache aggressively, fetch only top 20 coins, consider Pro tier
FMP free tier (250 req/day)	High	Medium	Fetch fundamentals weekly not daily, cache in DB, prioritize portfolio tickers
Trendline engine >20 min for full universe	Medium	Medium	Process in batches of 50, parallelize where Vercel allows
Vercel serverless 60s timeout	Medium	High	Split heavy jobs into smaller cron-triggered chunks
Stale data producing false signals	Low	High	Pipeline status monitoring, skill checks data freshness before presenting results
User profile out of date	Medium	Medium	Skill prompts user to verify holdings if last update >30 days

---

Comparison: Our Roadmap vs. Lighthouse’s

Dimension	Lighthouse (14 weeks)	Ours (9 weeks)
Scope	Full standalone web app, 9 tabs	Data service + Claude skill, 5 commands
Assets	549 stocks only	Stocks + crypto + ETFs, dynamic universe
Intelligence	Technical analysis only	Technical + fundamental + portfolio + tax
Personalization	None (same view for everyone)	Filtered by holdings, thesis, allocation targets
Productivity tools	Kanban + Notes (4 weeks)	None (use existing tools)
Hourly data	Yes (expensive)	No (irrelevant for personal investing)
Real-time scanning	Every 60 seconds	Once daily post-market
UI	Full dark-theme dashboard	Claude conversation (optional minimal dashboard later)
Integration	Standalone	Feeds into /thesis-to-trades, /decide, /rebalance

This site is open source. Improve this page.