trading-agent.md

gistfile1.txt

---
name: daily-trading-agent
description: >
  Runs a daily non-tech equity trading session using the Robinhood MCP. Reads memory first, then scans market trends, identifies 1–3 small-cap or mid-cap opportunities outside the technology sector (GICS sector 45), evaluates each candidate, places $5–$10 fractional trades, and logs a session summary. Use this skill whenever the user asks to run the morning trading agent, execute the daily trade scan, or do the non-tech portfolio session. Also trigger when the user says things like "run the agent", "do the morning trades", "what should I buy today (non-tech)", "run the daily scan", or "execute today's trades". Trigger proactively on any message that implies running the automated brokerage workflow.
compatibility: "Robinhood MCP, WebSearch"
---

# Daily Non-Tech Trading Agent

A repeatable morning session that builds a diversified, non-tech equity portfolio through small daily trades. Runs once per trading day, ideally at or after market open (9:30 AM ET).

---

## Goal

Identify and execute 1–3 small trades per session in sectors **outside technology** (GICS sector 45). Target small-cap to mid-cap companies with clear near-term catalysts. Position size: **$5–$10 per trade**. Over time, this builds a portfolio that diversifies away from existing tech-heavy holdings.

**Priority sectors:** Pharmaceuticals · Healthcare · Industrials · Consumer Staples · Energy · Utilities · Financials · Materials

**Excluded sectors:** Technology (GICS 45) · Semiconductors · Software · Cloud infrastructure · Tech-adjacent ETFs (QQQ, XLK, etc.)

---

## Step 0 — Read Memory

Before doing anything else, read all memory files to load prior session context, known blockers, and open position watch items. This is critical — the memory system captures things the brokerage API alone cannot tell you.

```
Read: memory/MEMORY.md   → load the index
Read: each linked .md file listed in the index
```

Key things to look for in memory:
- **Account blockers** (e.g., investor profile incomplete) — if a known blocker exists, attempt a single `review_equity_order` early in the session to test whether it's been resolved before proceeding
- **Open position watches** — check flagged positions (e.g., a high-beta holding below a key support level) against current prices
- **Watchlist entries** — previously identified tickers approved for entry at specific price levels; check these before doing a fresh market scan

---

## Step 1 — Pre-flight Check

Verify account state and available buying power.

```
Robinhood:get_accounts           → get account_number where agentic_allowed=true
Robinhood:get_portfolio          → check buying power
Robinhood:get_equity_positions   → review all current positions (check memory watch items vs current prices)
Robinhood:get_equity_orders      → filter placed_agent="agentic", created_at_gte=today — check for duplicate orders
```

**Abort conditions:**
- Buying power < $5 → log reason, stop
- 3+ agentic orders already placed today → session limit reached, log and stop
- Known account blocker from memory still active (confirmed by a failed review attempt) → log, stop, remind user of the resolution step

---

## Step 2 — Macro Check

Before scanning, confirm no major macro event is scheduled for today that would warrant sitting out.

```
WebSearch: "FOMC CPI NFP scheduled [today's date]"
```

If FOMC decision day, CPI release, or NFP release is today → **abort session**. Log "macro event day — no trade" and stop. Waiting one session for signal clarity is almost always the right call.

---

## Step 3 — Market Scan

Check memory watchlist entries first — if a previously identified ticker is at or near its approved entry price, that's a candidate without needing a fresh scan. Then run 2–3 targeted web searches:

```
WebSearch: "non-tech stock movers today [date]"
WebSearch: "pharma small cap catalyst today [date]"
WebSearch: "[sector] stocks momentum [date]"
```

Look for:
- Pre-market or early session price movement (>2% on volume)
- News catalysts: FDA decisions, earnings beats, contract wins, analyst upgrades
- Sector rotation signals into defensive or cyclical names
- ISM / jobs data beats that lift specific sectors (e.g., strong ISM Manufacturing → industrials, materials)

**Filter out immediately:**
- Any ticker with primary revenue from tech infrastructure, semiconductors, or software
- Market cap > $10B (unless thesis is unusually compelling)
- ETFs or index funds — individual equities only
- Average daily volume < $1M

---

## Step 4 — Candidate Evaluation

For each candidate (target 3–5 before narrowing to 1–3):

```
Robinhood:search                 → confirm ticker symbol and instrument_id
Robinhood:get_equity_quotes      → get current price, bid/ask spread, check volume
Robinhood:get_equity_tradability → confirm tradeable and fractional_tradability="tradable"
```

Score each on:

| Factor | What to assess |
|---|---|
| Trend strength | Price moving with volume, or just noise? |
| Catalyst clarity | Specific event driving the move? |
| Sector fit | Clearly outside tech? Low tech-revenue exposure? |
| Downside | Clean setup, or obvious risks (dilution, litigation, commodity collapse)? |
| Portfolio fit | Already held? Too similar to an existing position? |
| Memory watchlist | Previously approved entry? At or near the approved level? |

Aim for **1–3 trades**. If no clean candidate exists, skip the session — do not force a trade.

---

## Step 5 — Pre-Trade Review

For each selected trade, run a dry-run review before placing:

```
Robinhood:review_equity_order
  account_number: [from Step 1]
  symbol:         [ticker]
  side:           "buy"
  type:           "market"
  dollar_amount:  "5.00" to "10.00"
  market_hours:   "regular_hours"
  time_in_force:  "gfd"
```

Check the response for:
- Pre-trade alerts (buying power, PDT flag, instrument halt)
- **Investor profile error** (`non_field_errors` about answering investing goal questions) → if this appears, abort all trades, log the blocker, remind user to resolve at: `https://applink.robinhood.com/investment_profile?account_number=[account]&context=second_trade`
- Any other warnings → resolve or skip that ticker

---

## Step 6 — Execute

Place each approved trade:

```
Robinhood:place_equity_order
  account_number: [from Step 1]
  symbol:         [ticker]
  side:           "buy"
  type:           "market"
  dollar_amount:  "5.00" to "10.00"
  market_hours:   "regular_hours"
  time_in_force:  "gfd"
  ref_id:         [fresh UUID per order — reuse same UUID on retry only]
```

Generate a fresh UUID for each logical order. On network retry, reuse the same `ref_id` to prevent duplicates.

---

## Step 7 — Position Review (Weekly)

Once per week, after placing trades, pull full position data and assess the portfolio:

```
Robinhood:get_equity_positions   → all current holdings
Robinhood:get_equity_quotes      → current prices for all held symbols
```

For each position, compute unrealized P&L. Flag any position:
- Down >10% from cost basis → note in session summary, consider trimming
- Below a memory-noted support level → escalate the watch
- Oversized vs the $5–$10 session sizing (implies manually placed or legacy position) → flag for user review

---

## Step 8 — Session Summary

After all trades are placed (or session is skipped/aborted), save a structured summary to the workspace:

**File:** `session-YYYY-MM-DD.md` in the project folder

```markdown
## Morning Session — [Date]

**Scanned:** [N] candidates across [sectors]
**Traded:** [N] positions

| Ticker | Sector | $ Invested | Thesis |
|--------|--------|------------|--------|
| XYZ    | Pharma | $7.50      | FDA catalyst expected Q3 |

**Skipped candidates:**
- [TICKER]: [reason]

**Session skipped/aborted:** [if applicable — reason + any resolution steps for user]

**Open position watch:**
- [TICKER]: [current price vs cost basis, any flag]
```

---

## Guardrails

- **No session on macro event days** (FOMC decision, CPI, NFP release) — wait for clarity
- **Hard tech exclusion** — skip any ticker where >30% of revenue is tied to tech infrastructure, semiconductors, or enterprise software. When in doubt, search the company description
- **No forced trades** — choppy market or no clean setup → log "no trade today" and stop
- **Session cap** — max 3 trades per session, $10 max per position
- **No market orders in extended hours** — `market_hours` must be `regular_hours` for dollar-amount (fractional) orders
- **Idempotency** — always check `get_equity_orders` at Step 1 before placing new ones
- **Memory-first** — Step 0 is not optional; prior session context shapes candidate selection and risk assessment

---

## Notes

- Dollar-amount (notional) orders require `type=market` and `market_hours=regular_hours`. The server computes fractional share count from last trade price.
- If `review_equity_order` returns a PDT (Pattern Day Trader) alert, skip that ticker for the day.
- This agent is designed for learning signal and portfolio diversification — not aggressive returns. Small size is intentional.
- Session logs are saved to the project folder as `session-YYYY-MM-DD.md`.