Educational resource. Not financial advice. Example parameters are illustrative desk defaults, not a performance guarantee. Agentic trading can lose the entire funded wallet.

Companion: [Robinhood MCP for AI Agents](/resources/robinhood-mcp)

## Goal

Positive-expectancy **small equity scalps** on a Robinhood **Agentic** account using the Trading MCP — with a written screen, trigger, size, take-profit, and cut **before** every entry. Most of the session is [sitting on your hands](/resources/sitting-on-your-hands); see the [refusal list](/resources/when-the-agent-should-sit) and [Evidence](/evidence) field notes.

This playbook is for **desk-side agentic sessions**. It is not a substitute for a production multi-broker automation ledger.

## Hard rules

1. **Account:** place/cancel only when `agentic_allowed=true`.
2. **Default window:** prefer **14:30–15:55 ET** continuation / EOD. Avoid new risk before **14:00** unless a Tier-A exception clears.
3. **Max live ideas:** **2** concurrent (prefer **1** when the first name has extreme relative volume).
4. **Size:** example default **$500–750** notional per name; size up only via the checklist below. Example daily loss stop **−$40**; daily win soft stop **+$40** (stop adding). Scale these to your wallet.
5. **No overnight** except an intentional after-hours penny take-profit a few cents above cost — cancel unfilled stretch targets by evening.
6. **Every long needs both** profit limit(s) **and** a cut (time and/or structural invalidation beyond the sweep). Ladders alone are invalid.
7. **No hard stops parked in liquidity sweeps.** Obvious stop clusters (opening-range low, prior 5m swing low, round-number shelves) are **entry / add zones**, not MAE parking spots. Protective cuts go **beyond** the swept low.
8. **Buy the sweep, not the mid-reclaim.** If entry distance to the true swept low is much larger than the take-profit, skip or wait for a retest — do not invent a nearer shelf to fake R:R.
9. **Review → confirm → place.** Disclose quotes on review. Prefer marketable limits over blind markets.

## Setup filter

### Step 0 — Trend of day

Classify from SPY + QQQ vs prior close and session VWAP:

- **Bull day** — Both green vs prior close and at least one above VWAP → long-only continuation / late push
- **Bear day** — Both red vs prior close and at least one below VWAP → relative-strength Regime D/A longs only, or flat
- **Mixed / chop** — Split signs or oscillating around VWAP → prefer flat after 15:00 unless Tier-A

### Step 1 — Liquid prefilter (scanner)

On bull / mixed days, refresh a liquid gainer scan:

- Common stock · day change **> ~3%** · relative volume **> ~1.5**
- Average volume (10) **> ~1M** · last roughly **$8–80** · market cap **> ~$1B**

On bear days, mirror with liquid losers. Prefer **relative volume** as the primary sort after the liquid gate — raw % change overweight already-extended names.

### Step 2 — Time-of-day post-filter (after 14:30)

- **14:30–15:00** — Prefer pullback/reclaim (Regime A); price > VWAP; 5m RSI ~50–68; MACD hist ≥ 0 or improving. Demote RSI>70 extending; day>12% without Tier-A.
- **15:00–15:25** — Prefer late push (Regime B): fresh 5m RVOL, reclaim prior 5m high, price > VWAP. Demote soft volume / MACD flip / no reclaim by 15:25.
- **15:25–15:55** — Manage opens only; no fresh idea scans.

**Hard rejects (any window):** earnings today / next AM; chaotic spreads; rank-chase without trigger; same-day re-entry after a stop-out; day move >12% unless RVOL extreme and RSI not extended.

## Entry triggers (must pass one)

- **A — Continuation** — Pullback ≤ ~0.4× 5m ATR then hold / reclaim last 5m high. Skip if RSI(5m)>70 and still extending.
- **B — Late push** — Fresh RVOL expansion after 15:00 with price > session VWAP. Skip into resistance / tired RSI.
- **C — Tier-A (pre-14:00)** — Extreme RVOL, strong day %, RSI mid-band, tight spread. Otherwise flat until 14:30.
- **D — Sweep reclaim** — Liquidity grab through OR / prior 5m swing / round shelf, then hold & reclaim — **buy the sweep**. Skip if sweep fails reclaim.

No trigger → **no trade**, even if the scanner ranks the name highly.

### Liquidity-sweep placement

- Map before entry: opening-range high/low (first 15m), last 1–2 swing lows on 5m, nearby round numbers, session VWAP.
- Prefer **limit buys** resting at / just above the sweep after a reclaim print — not a chase into the spike.
- Do **not** put the working cut at that same level. Sweeps exist to take those stops.
- Morning ~09:45 window: default Regime **D** (or A after a clean reclaim).

## Size

### Default

Example desk default: **$500–750** notional for any name that clears gates (including soft MACD miss, micro-ATR, midday relative-strength). Soft signals → stay at default or skip.

### Size-up checklist (all required)

Bump modestly (example **1.25–1.5×**, capped until sample grows) only when **every** gate is green:

1. MACD histogram **≥ 0** (hard; no soft miss)
2. Price **> VWAP** with clear cushion
3. RSI(5m) roughly **52–65**
4. RVOL **≥ 1.5** (prefer ≥ 2.0)
5. Regime **D or A** written: buy at/near sweep; structural cut **beyond** true sweep; risk roughly **≤ 1.5–2×** take-profit
6. Spread acceptable
7. Window **14:30–15:25** **or** morning clean Regime D / Tier-A
8. Take-profit distance is meaningful in dollars (not micro-ATR noise)
9. No earnings today / next AM

**Never size up** to recover a loser or because RVOL alone looks loud (especially earnings movers).

## Exit grammar

### Profit

- Default: single take-profit near **~0.4–0.6× 5m ATR** above fill.
- Optional 2-rung ladder only on Tier-A / extreme RVOL.
- Avoid 3-rung ladders until the session is already green and an MAE plan is live.

### Cut (pick at entry; write it down)

- **Time (default)** — Flat by **T+45m** or **15:55 ET**, whichever first
- **Structural** — 5m **close** beyond the swept trigger / OR low → exit. A wick alone is the sweep, not the cut.
- **MAE (rare)** — Only if a clear shelf exists **under** the sweep. Never park MAE at OR / swing / round liquidity.

If take-profit 1 fills and the residual is red vs entry → do not hold the residual past the time stop.

## Session checklist

- Confirm Agentic buying power; confirm flat book
- Scan + earnings check; no entries before 14:00 unless Tier-A / Regime D
- ~14:25: re-scan; pick ≤2 names with written trigger, size, TP, cut, sweep map
- Review → confirm → place; attach TP; watch cut clocks
- ~15:50 cancel stretch targets; **15:55 hard flat**
- Log fills for next session's lessons

## Decision tree

```
Screen name
  ├─ map sweep levels — buys go there; stops do NOT
  ├─ Tier-A metrics clear at any time? → still need a trigger
  ├─ time < 14:00 and not Tier-A / sweep-reclaim? → SKIP
  ├─ no pullback / reclaim / VWAP / sweep trigger? → SKIP
  ├─ enter (prefer limit at sweep reclaim)
  │     → TP(~0.5×ATR) + time≤45m
  │     + invalidation = close beyond swept low
  └─ at cut → cancel TPs → sell → stop adding if daily loss stop hit
```

## Anti-patterns

- Midday ladder farms without cuts
- “Monitor and hope” into the close after MAE already exceeds TP1 distance
- Size-up to recover a loser
- Size-up on soft MACD, mid-reclaim, or micro-ATR
- Trading non-agentic accounts via MCP
- Hard stops parked in liquidity sweeps
- Chase entries into RSI>70 extension when a sweep limit is available
- Inventing a nearer shelf so paper R:R looks fine when true sweep risk is much larger than TP

## Living lessons (pattern language)

- **MAE in the sweep** — Protective cuts sitting in the liquidity grab get hunted — place orders at the sweep, cuts beyond it
- **Mid-reclaim ≠ Regime D** — Buying halfway up the reclaim with a fake near shelf destroys true R:R
- **Size-up = full stack** — Soft MACD or micro-ATR winners stay at baseline size

## Morning Opening Range variant

Same hard rules with a shorter clock:

- Premarket liquid gapper / gainer screen
- Map opening range **09:30–09:45 ET**
- Prefer Regime **D** sweep-reclaim after 09:45
- Hard flat by **11:00 ET**
- Share the same daily loss / soft win stops across morning and afternoon sessions

## FAQ

### Does this playbook guarantee profits?

No. It encodes risk grammar and filters that improve desk discipline. Markets change; sample sizes stay small.

### Can I run this fully unattended?

Only if you explicitly enable auto-execute **and** enforce hard caps outside the model (wallet size, daily stop, flat clock). Prefer confirm-before-place until the session grammar is stable.

### How does this relate to Robinhood's example prompts?

Robinhood's examples are open-ended. This playbook replaces them with a closed decision tree an agent can follow without inventing risk rules mid-trade.

## Related

- [Why Soyhands: Sitting on Your Hands](/resources/sitting-on-your-hands)
- [When the Agent Should Sit](/resources/when-the-agent-should-sit)
- [Robinhood MCP for AI Agents](/resources/robinhood-mcp)
- [Evidence](/evidence)
- [Resources index](/resources)
- [Robinhood Agentic Trading overview](https://robinhood.com/us/en/support/articles/agentic-trading-overview/)

## Disclaimer

Educational only. Not an offer to buy or sell securities. Past patterns are not future results. AI agents can err. You are responsible for every agent-placed order. This site is not a broker-dealer and is not affiliated with Robinhood.


Canonical: https://www.soyhands.com/resources/agentic-trading-playbook
Markdown: https://www.soyhands.com/resources/agentic-trading-playbook.md