geogons/skill-financial-analyst/skill-financial-analyst

Financial Analysis Skill

A multi-source financial analysis engine that aggregates data from 14+ free APIs and 20+ RSS feeds to produce professional-grade stock analysis with actionable buy/sell ratings and price targets.

First-Time Setup

Before using this skill, read the API setup guide and configure API keys:

Read references/api_setup_guide.md

Then initialize the config:

python scripts/api_config.py init
python scripts/api_config.py status

Install Python dependencies:

pip install yfinance feedparser pandas pandas-ta requests --break-system-packages

Architecture Overview

financial-analysis/
├── SKILL.md                          # This file — main instructions
├── scripts/
│   ├── api_config.py                 # API registry, keys, fallback chains
│   ├── usage_tracker.py              # Rate limit enforcement, logging, reports
│   ├── api_caller.py                 # Resilient caller with auto-fallback
│   └── rss_feeds.py                  # RSS feed catalog and parser
└── references/
    └── api_setup_guide.md            # Step-by-step key setup + rate limits

How the API System Works

Every API call flows through scripts/api_caller.py which:

1. Checks rate limits before calling (per-minute, per-day, per-month)
2. Enforces delays between calls to avoid IP blocks
3. Logs every call to ~/.financial-analysis/logs/api_usage.jsonl
4. On failure: logs to error_log.jsonl and tries the next API in the fallback chain
5. On all failures: returns an error with details of every attempt

The tracker at scripts/usage_tracker.py aggregates usage and generates:

• Daily usage reports with % of free limit consumed per API
• Error reports showing which APIs failed and which fallbacks worked
• Paid tier recommendations when any API exceeds 70% of its free limit

Fallback Chains

Every data category has a prioritized chain of APIs. If the primary fails or is rate-limited, the next one is tried automatically. The chains are defined in scripts/api_config.py → FALLBACK_CHAINS. Key chains:

Three Use Cases

This skill supports three distinct workflows. The user will ask for one of them (or you can suggest the appropriate one based on context).

Use Case 1: Weekly Portfolio Review

Trigger: User provides a PDF, CSV, or list of current holdings and asks for a review, analysis, or buy/hold/sell recommendations.

Input: Portfolio holdings (tickers + optional quantities/cost basis) Output: Per-ticker rating (Buy / Hold / Trim / Sell) with confidence score, 3 entry levels, 3 sell targets, and key reasons. Frequency: Once per week (Sunday/Monday)

Workflow

Execute these phases in order. Use call_with_fallback() from scripts/api_caller.py for every API call to get automatic fallback + logging.

Phase A — Data Collection (for each ticker in portfolio):

1. Fetch current price + 1yr daily OHLCV history (yfinance)
2. Pull quarterly financial statements: income, balance sheet, cash flow (yfinance + SEC EDGAR)
3. Get analyst ratings consensus: buy/hold/sell counts (Finnhub)
4. Scan insider activity last 90 days: Form 4 filings (SEC EDGAR)
5. Check insider sentiment score: MSPR (Finnhub)
6. Pull front-month options chain (yfinance)
7. Get news sentiment (Finnhub + Alpha Vantage)
8. Single call: Reddit trending tickers + mention velocity (ApeWisdom)
9. Single call: recent Congress trades (Mboum)
10. Single call: upcoming earnings calendar (Finnhub)

Phase B — Technical Analysis (computed locally with pandas-ta, no API calls):

• SMA/EMA suite (10/20/50/100/200)
• RSI (14), MACD (12,26,9), Bollinger Bands (20,2)
• Stochastic Oscillator, ADX
• Volume analysis + OBV
• Pivot points (Classic + Fibonacci)
• Fibonacci retracement + extensions from recent swing
• ATR for position sizing

Phase C — Scoring (all local compute):

• Fundamental score (0-100): revenue growth, margins, FCF, ratios, peer comparison
• Technical score (0-100): trend, momentum, volatility, S/R positioning
• Sentiment score (0-100): analyst + insider + Congress + news + social
• Composite score: 40% fundamental + 30% technical + 30% sentiment
• Rating: 75-100 = Buy · 40-74 = Hold · 0-39 = Trim/Sell

Phase D — Output Generation (per ticker):

After all tickers are analyzed, run the usage report:

from scripts.usage_tracker import get_tracker
get_tracker().print_daily_report()
get_tracker().save_daily_summary()

Use Case 2: Daily Opportunity Scanner

Trigger: User asks to find new stock ideas, scan for opportunities, or identify swing trades.

Input: User's portfolio focus (e.g., "growth", "value", "income") and optional current holdings to avoid duplicates. Output: Top 5 Portfolio Adds + Top 5 Swing Trades, each with score, 3 entries, 3 exits. Frequency: Daily (before market open)

Workflow

Phase A — Source Aggregation (cast a wide net):

1. Scan all Tier 1+2 RSS feeds using scripts/rss_feeds.py → scan_all_feeds(). Extract tickers mentioned across all feeds.
2. Pull Seeking Alpha per-ticker RSS for top 5 tickers from step 1.
3. Get Reddit trending tickers (ApeWisdom) — cross-reference with RSS mentions.
4. StockTwits sentiment scan for top 10 overlapping tickers.
5. Check new Congress trades (Mboum).
6. Pull upcoming earnings in next 5 days (Finnhub).

Phase B — Screening (narrow down to top 20 candidates):

1. News sentiment on top 20 candidates (Finnhub).
2. SA quant ratings + factor grades for top 5 (SA RapidAPI).
3. AI sentiment score for top 5 (Alpha Vantage — 5 of 25/day).
4. Price + volume screening (yfinance) — 52wk range positioning, volume vs avg.
5. Financial health check for top 10 (yfinance) — revenue growth, margins, PE.
6. Analyst consensus for top 10 (Finnhub).
7. Insider activity for top 10 (SEC EDGAR).

Phase C — Analysis (for top 10 survivors):

1. Compute S/R levels + entry/exit targets (pandas-ta locally).
2. Check unusual options activity for top 5 (Mboum).
3. Multi-factor scoring + final ranking (local compute).

Phase D — Output:

Present two lists:

Portfolio Adds (growth-aligned, longer horizon):

• Ticker, Score, Rating, 3 Entries, 3 Exits, Stop Loss, R:R, Catalysts

Swing Trades (momentum-based, shorter horizon):

• Ticker, Score, Rating, 3 Entries, 3 Exits, Stop Loss, R:R, Catalysts, Time Horizon

Use Case 3: On-Demand Stock Deep Dive

Trigger: User asks to analyze a specific stock, do a deep dive on a ticker, or asks "should I buy X?"

Input: 1-10 tickers Output: Full analyst-grade report per ticker with Buy / Watch·Hold / Sell rating, 3 entries, 3 exits, and comprehensive analysis.

Workflow

This is the most thorough analysis — every available data source is queried. Four phases, ~30 API calls per ticker.

Phase A — Fundamentals (11 steps per ticker): Company profile, 5yr income statement, quarterly income trends, balance sheet, cash flow, XBRL cross-validation from SEC EDGAR, key ratio computation (PE/PB/PS/PEG/EV-EBITDA/ROE/ROA/D-E/FCF yield), peer comparison (5 peers), earnings history + surprises, upcoming earnings date, dividend history.

Phase B — Technicals (17 steps, mostly local via pandas-ta): 2yr daily + 6mo intraday price history, SMA/EMA suite, RSI, MACD, Bollinger Bands, Stochastic, ADX, Volume + OBV, VWAP, Pivot Points, Fibonacci Retracement, Fibonacci Extensions, Volume Profile, ATR, Ichimoku Cloud. Cross-validate 2 indicators via Alpha Vantage API.

Phase C — Sentiment & Alternative Data (16 steps): Analyst ratings consensus (Finnhub), price targets (yfinance), insider Form 4 filings (SEC EDGAR), insider MSPR (Finnhub), Congress trades (Mboum), institutional holders (yfinance), short interest, news sentiment (Finnhub + AV AI), Reddit mentions (ApeWisdom), StockTwits sentiment, SA quant ratings + factor grades (SA RapidAPI), SA per-ticker articles (RSS), options flow (yfinance), unusual options (Mboum), recent SEC filings scan.

Phase D — Scoring & Output (12 steps, all local): Fundamental score (0-100), Technical score (0-100), Sentiment score (0-100), Composite score (weighted 40/30/30), Rating assignment (Buy ≥75 / Watch·Hold 40-74 / Sell <40), 3 Entry levels, 3 Exit targets, Stop Loss, Risk:Reward ratio, Confidence level, Key Catalysts, Key Risks.

Output a comprehensive per-ticker report with all findings. Only recommend trades where R:R ≥ 2:1.

Error Handling

When any API call fails, the system:

1. Logs the error to ~/.financial-analysis/logs/error_log.jsonl with timestamp, API, category, and error message
2. Tries the fallback — next API in the chain for that data category
3. Logs fallback result — records which fallback was used and whether it succeeded
4. If all fallbacks fail, continues with available data and notes the gap in the output

Viewing Errors

python scripts/usage_tracker.py errors      # Last 7 days
python scripts/usage_tracker.py errors 30   # Last 30 days

Common Errors and Fixes

Usage Reports & Paid Tier Recommendations

After each workflow run, the usage tracker saves a daily summary. When any API exceeds 70% of its free limit, it generates a paid tier recommendation:

python scripts/usage_tracker.py daily    # See today's report
python scripts/usage_tracker.py summary  # Save JSON summary

The summary includes per-API: calls today/this month, % of free limit, error count, and a severity rating (SAFE / WARNING / CRITICAL).

The APIs most likely to need upgrading (in order):

1. Alpha Vantage (25/day) — first to hit limits. Upgrade: $29.99/mo
2. Mboum (600/mo) — tight if running all 3 use cases daily. Upgrade: $9.95/mo
3. SA RapidAPI (1K/mo) — tight with heavy deep dive usage. Check RapidAPI pricing.

Important Notes

• This skill provides data-driven analysis, not financial advice. Always do your own due diligence.
• yfinance is an unofficial library — data may be delayed or temporarily unavailable.
• SEC EDGAR data is authoritative but requires parsing.
• All timestamps are in the timezone of the market being analyzed (US Eastern for US stocks).
• The scoring system weights can be adjusted based on user preference.