Overview

Step 3 — Compute technical indicators

For each symbol, pass its candles to the script:

python3 scripts/compute_indicators.py --candles '<json_array>'

The script outputs a JSON object with:

• rsi_14: current RSI value
• macd_line, signal_line, histogram: current MACD values
• ema_20, ema_50, ema_200: current EMA values
• close: latest close price
• signal_summary: one of STRONG_BUY / BUY / NEUTRAL / SELL / STRONG_SELL

Refer to references/trading-strategy.md for how to interpret these values.

Step 4 — Fetch news sentiment

For each symbol in the universe, use WebSearch to search:

"[SYMBOL] NSE stock news site:economictimes.com OR site:moneycontrol.com OR site:livemint.com"

Summarize the top 3 results per symbol. Score sentiment per references/sentiment-scoring.md.

Step 5 — Review past decisions

Run: python3 scripts/log_decision.py --action read --symbol ALL --last 20

Review recent decisions. Note:

• Symbols where a stop-loss was hit (avoid re-entering too soon)
• Positions already entered that are still open (don't double-enter)
• Patterns in what's been working vs not

Refer to references/decision-log-format.md for the log schema.

Step 6 — Score and rank trade ideas

For each symbol, combine:

• Technical signal score (from Step 3, per trading-strategy.md)
• Sentiment score (from Step 4, per sentiment-scoring.md)
• Past decision context (from Step 5)

Apply risk filters from references/risk-management.md:

• Skip any symbol with risk/reward below 1:2
• Skip any symbol where you already hold an open position logged in the decision log
• Flag any symbol that hit a stop-loss in the last 5 trading days

Rank remaining symbols by combined score, highest first.

Step 7 — Present recommendations

Show the top 3–5 trade ideas in a clean table:

SYMBOL | Direction | Entry | Target | Stop-Loss | R/R | Technical Signal | Sentiment | Reasoning

After the table, write 2–3 sentences per trade explaining the setup in plain English.

Step 8 — Ask for approval

Ask the user: "Which trade(s) would you like to place? Type the symbol(s) or say 'none'."

For each approved trade:

• Confirm: symbol, direction (BUY/SELL), quantity (suggest based on risk-management.md), order type (MARKET or LIMIT)
• Ask: "Place this order: BUY [QTY] [SYMBOL] at [PRICE]? (yes / no)"

Step 9 — Place approved orders

For each confirmed order, call:

• mcp__kite__place_order with the confirmed parameters
• Confirm success and show the order ID

Step 10 — Log decisions

For each trade presented (whether placed or skipped), run:

python3 scripts/log_decision.py --action append \
  --symbol SYMBOL \
  --direction BUY/SELL/SKIP \
  --entry_price PRICE \
  --target TARGET \
  --stoploss STOPLOSS \
  --reasoning "brief reasoning" \
  --order_id ORDER_ID_OR_NA

Hard Rules

• NEVER place an order without explicit user confirmation ("yes") on each trade
• NEVER recommend more than 5 trades in a single session
• NEVER skip the stop-loss calculation — every recommendation must have one
• If Kite returns an error at any step, report it clearly and continue with the remaining symbols
• If fewer than 3 stocks pass the risk filter, say so and explain which ones were filtered and why
• Always present the decision log review (Step 5) findings before showing recommendations