Security: Ticker accepts A–Z letters only (1–6 chars). API key must match
sk-ant-... format. Your key is never stored — sent only to api.anthropic.com via CSP-restricted connection. No third-party scripts.
📊Pre-Trade Analysis
A–Z only, 1–6 letters
Must start with sk-ant-
NVDA
Act as a professional trader and give me a concise pre-trade
analysis for NVDA for a potential LONG position.
1. Trend & Momentum · 2. Bollinger Bands
3. Key Levels · 4. ATR & Risk
5. Volume & Structure · 6. Catalysts
7. Trade Plan · 8. Multi-Day Outlook
Verdict: “Good entry now” OR “Wait for pullback to $___” OR “Avoid — reason”
SCORES (each 1–10, 10 = most favorable for long entry):
Trend Direction Score: X/10
RSI Score: X/10
Bollinger Position Score: X/10
Bollinger Volatility Score: X/10
Breakout vs Pullback Score: X/10
Volume Confirmation Score: X/10
Key Levels Score: X/10
ATR Risk Score: X/10
Catalyst Score: X/10
Trade Setup Score: X/10
Outlook Score: X/10
Overall Score: X/10
Claude is building analysis + 12-score scorecard…
📈Trade Setup Scorecard
—/10
Context
Why build a Claude-powered trade analyzer?
Most retail traders spend 20–30 minutes pulling together trend data, Bollinger Bands, RSI, ATR, and catalyst checks before every trade. This tool collapses that workflow into a single structured API call — returning a full 8-section analysis and 12-dimension scorecard in ~10 seconds.
8
Analysis sections
12
Scores generated
~10s
Time to full output
0
Backend servers
Understanding the Analysis
Trading terms, how Claude agents work, and the prompt structure — all explained.
RSI score
Rates RSI position 1–10. RSI 40–60 = neutral (5–6). RSI <40 from oversold bounce = bullish (8–9). RSI >70 approaching overbought = bearish (2–3). Divergence from price adds weight.
Bollinger position score
Rates where price sits within the bands. Near lower band = high score (potential discount entry). Near upper band = low score (extended/overbought). Mid-band = neutral 5.
Bollinger volatility score
Rates band width and direction. Bands squeezing (narrowing) before a breakout = high score. Bands already wide and expanding = lower score — most of the move may already be priced in.
Breakout vs pullback score
Rates the structural entry quality. Clean pullback to support in uptrend = high score (favorable R/R). Chasing a breakout extension = low score. Range-bound with no clear setup = 5.
Volume confirmation score
Rates volume behavior. Declining volume on pullback (healthy retracement) = high score. Rising volume on breakout = high score. Breakout on low volume = low score — conviction is absent.
ATR & risk score
Rates the ATR-implied daily risk. Small ATR relative to entry distance = high score (tight stop viable). Large ATR requiring wide stop = lower score. A 14-day ATR of $6 means $9 stop minimum (1.5×).
Key levels score
Rates proximity and clarity of support/resistance. Entry near a clean, untested support = high score. Entry in open air with no nearby level = low score. Multiple confluent levels = highest score.
Catalyst score
Rates macro and event risk. No earnings, no FOMC in the next 2 weeks = high score. Earnings next week = low score (binary risk). Positive recent news flow = +1 bonus to score.
Trade setup score
Overall entry quality combining R/R ratio, trend alignment, and entry zone. 8–10 = high conviction long. 5–7 = proceed cautiously. Below 5 = wait or avoid entirely.
Overall score
Average of all 11 dimension scores. Above 7 = favorable setup. 5–7 = proceed cautiously. Below 5 = wait for better conditions. Use as a go/no-go gauge — not a guarantee.
Trend direction score
Rates the strength and clarity of trend on Daily and 1H charts. Strong uptrend with higher highs/lows on both timeframes = 9–10. Mixed or consolidating = 5. Downtrend = 1–3.
Outlook score
Rates the 3–10 day expected scenario quality. Strong bullish catalyst + clear path to target = 8–10. Uncertain macro backdrop = 5. Bearish macro + sector weakness = 2–4.
Step 01 — Input
You enter a ticker
JavaScript validates: ticker must match
/^[A-Z]{1,6}$/, API key must match /^sk-ant-[A-Za-z0-9\-_]{10,}$/. Nothing is sent until both pass. Invalid characters are stripped on each keystroke.↓
Step 02 — Prompt
Ticker injected into the expert prompt
The validated ticker replaces [TICKER] in a pre-written prompt with 8 structured sections plus a 12-score block. This is prompt engineering — forcing consistent, machine-parseable output from the model.
↓
Step 03 — API call
Browser → api.anthropic.com only
POST to
api.anthropic.com/v1/messages. CSP header restricts connect-src to Anthropic exclusively. No third-party scripts, no proxy, no data logging on any intermediary server.↓
Step 04 — Inference
Claude generates 8 sections + 12 scores
claude-sonnet-4 reads the full structured prompt, applies domain knowledge about technical analysis, and outputs all 8 analysis sections followed by a SCORES block with 12 numeric 1–10 ratings.
↓
Step 05 — Render
Parse → grouped scorecard + cards
JavaScript extracts all 12 score values via regex, groups them into 4 themed sections (Trend, Bollinger, Structure, Risk), builds DOM nodes with
textContent (no innerHTML — XSS-safe), colors bars green/amber/red.What makes this an AI Agent pattern?
An AI Agent receives a role, a multi-step task, and a required output contract — then executes autonomously. Here Claude is briefed as “professional trader”, given 8 structured sections plus a 12-score format, and returns a complete trade brief with ratings. Same architecture as Cursor, Perplexity, and enterprise copilots — model has a job, context, and output format. No memory or tool calls needed for this single-call agentic pattern.
An AI Agent receives a role, a multi-step task, and a required output contract — then executes autonomously. Here Claude is briefed as “professional trader”, given 8 structured sections plus a 12-score format, and returns a complete trade brief with ratings. Same architecture as Cursor, Perplexity, and enterprise copilots — model has a job, context, and output format. No memory or tool calls needed for this single-call agentic pattern.
Three prompt engineering techniques: role injection, numbered sections (parseable), and constrained score format (12 labeled scores make automated rendering trivial).
Act as a professional trader · Analyze [TICKER] for LONG position
Sections 1–8: Trend · Bollinger · Key Levels · ATR
Volume · Catalysts · Trade Plan · Outlook
Verdict: “Good entry now” | “Wait for pullback to $___” | “Avoid — reason”
SCORES (1–10):
Trend Direction · RSI · Bollinger Position · Bollinger Volatility
Breakout vs Pullback · Volume Confirmation · Key Levels · ATR Risk
Catalyst · Trade Setup · Outlook · Overall
Why 12 scores instead of 8? The original 8 sections map to broad categories. The 4 new sub-scores — RSI, Bollinger Position, Bollinger Volatility, Breakout vs Pullback, Volume Confirmation — give you granular signal within the two most action-relevant sections (momentum and structure). A Bollinger section that says “near upper band, volatility expanding” is very different from “near lower band, bands squeezing” — the two sub-scores make that distinction visible at a glance.
Summary
Key takeaways
🧠
Prompt structure is everything
Numbered sections with sub-questions force structured expert reasoning. The 12-score format makes automation trivial. This is the most important applied AI skill.
📊
Sub-scores reveal nuance
Bollinger Position and Bollinger Volatility tell different stories. RSI at 68 vs RSI at 45 have opposite implications. Granular scores surface signal that section-level summaries hide.
🔒
Security by design
Input validation, textContent rendering (no XSS), CSP restricting connections to Anthropic only, and zero third-party scripts — non-negotiable even for educational tools.
⚠️
Always verify the live price
Claude has a training cutoff. Treat output as a structured checklist to verify against your live chart — not a Bloomberg terminal substitute or a trading signal.
🤖
One prompt can be an agent
No LangChain, no pipelines. Encoding role + 8-step task + 12-score output contract into a single prompt is the same architecture behind most real-world copilots.
🏗️
The foundation to extend
Add a live market data API (Polygon.io), brokerage execution (Alpaca), and persistent watchlist storage — this prompt + parse + render architecture scales directly.
Interview Prep
Cheat sheet — quick definitions
Define
What does the Bollinger Position score measure?
It rates where price sits relative to the three Bollinger Band lines. Near the lower band = high score (potential discount entry, mean reversion likely). Near the upper band = low score (extended, overbought risk). At the 20-day MA middle band = neutral 5. The Volatility score is separate — it rates whether the bands are expanding (breakout possible) or squeezing (setup forming).
Define
What does the RSI score represent and what are the thresholds?
The RSI score translates the raw RSI reading into a 1–10 entry quality rating. RSI 30–45 bouncing upward = 8–9 (oversold recovery). RSI 45–60 in neutral zone = 5–6. RSI 65–70 approaching overbought = 3–4. RSI above 70 = 1–2 (caution, extended). RSI divergence — price making new high but RSI declining — overrides the raw number and caps the score at 3.
Compare
Breakout vs pullback score — how does each scenario score?
A clean pullback to tested support in an established uptrend scores 8–9: your stop is tight, target is far, R/R is excellent. A fresh breakout above resistance with strong volume scores 7–8 if volume confirms. Chasing a breakout that has already extended 5%+ from the breakout level scores 3–4: your stop must be wide, R/R deteriorates. A range-bound setup with no clear directional bias scores 5.
Define
What does Volume Confirmation score measure?
It rates whether volume behavior supports the trade thesis. Declining volume on a pullback = bullish (sellers are exhausted, healthy retracement) → score 8–9. Rising volume on a breakout = conviction confirmed → score 8–9. Rising volume on a down day during a pullback = distribution risk → score 3–4. Low volume across all bars = no conviction either way → score 5.
Gotcha
Why can’t you blindly trust Claude’s specific price levels?
Claude has a training data cutoff — it does not see live ticks, real-time order books, or today’s chart. Price levels ($105, $112, etc.) are approximations from training patterns, not live market data. Always cross-reference every specific price level against your live charting platform before acting. Use the analysis as a structured reasoning checklist — not a Bloomberg terminal substitute.
Weakness
What are the three hard limits of this tool?
(1) No live data — cannot see today’s candles, volume profile, or order flow; (2) No execution — cannot place orders, set broker alerts, or connect to a trading platform; (3) Stateless — no memory between sessions, no watchlist persistence. To extend: add Polygon.io for live prices, Alpaca for execution, and localStorage or a backend for watchlist persistence.
Built for vijay-gokarn.com · Powered by Claude API · Educational use only — not financial advice