Prompt formula
[Intent] + [Asset / strategy] + [Constraints] + [Where]
Say what you want done, name the coin or strategy, add any limits (risk, model, budget), and say where it should run — Radar, Arena, analysis only, or leave it undeployed.
This is the same shape TrueNorth-style ops prompts use: outcome first, then the target, then the rails, then the surface.
Dissecting the slots
Intent
The verb and the outcome. Prefer one clear job per message when you are exploring; chain only when you already trust the flow.
| Strong | Weak |
|---|---|
| “Build a 4h compression breakout playbook” | “Do something with volatility” |
| “Halt Falcon — leave Radar alone” | “Fix my agents” |
| “Why isn’t Radar firing on SOL?” | “Check stuff” |
Asset / strategy
Name the coin, the strategy, and/or the agent. If Commander has several in play, ambiguity freezes the wrong one.
- Coin:
SOL,ETH, … - Strategy: title or “the one we just built”
- Agent: display name
Constraints
Rails that should land on the first save — not “we’ll tweak later.”
| Kind | Examples |
|---|---|
| Risk | max exposure $100, 1% sizing, cautious posture |
| Brain | measured / aggressive; approved model only if you care |
| Strategy shape | soft stop, raise aggregate gate, range not trend |
| Schedule | Radar hours / regime floors (when deploying) |
Where
| Where | Meaning |
|---|---|
| Analysis only | Read market / account — no writes |
| Strategy | Compile / apply the playbook; no deploy |
| Agent | Commission or rebind; still not on duty |
| Radar | Standing duty for a coin → Radar |
| Arena | Session / preset deployment → Arena |
| Terminal / approval | You (or a pending entry) take the wheel |
Omit Where when you only want a draft. Say it explicitly when money or standing authority is involved.
Starter shapes
Commander also offers starters (lander + host-specific). They are the same formula pre-filled — market, strategies, agents, trades, and “now” rows keyed to approvals, halts, open positions, and recent closes.
Example prompts
Build a strategy
“Build me something that buys SOL breakouts on rising volume”
Drafts a playbook for that setup — entries, filters, and exits you review before apply.
Tip: Swap the coin. Add “in a range, not a trend” or “with a soft stop” to steer shape.
Market read
“What is the market doing right now?”
Live structure, heat, and what is moving — not a generic essay.
Tip: Add a coin (“on SOL”) or timeframe (“on the 4h”).
Account check
“How am I actually doing?”
Agents, exposure, and recent results in plain language.
Tip: Name an agent for one runner’s scorecard.
Commission an agent
“Commission a new agent on that strategy”
Creates a runner bound to the strategy you just built (or name one).
Tip: Bake brain and risk in — “cautious brain, max $50 exposure.”
Deploy on Radar
“Put that agent on duty for SOL”
Saves a Radar deployment so the agent watches that coin under its playbook.
Tip: Name agent and coin when more than one is in play. Confirm — this is standing authority.
Enter Arena
“Enter me into the next open session”
Builds and submits a grid for the next Arena session. This can spend real money — confirm the fee.
Tip: Only when you mean to play and wager permission is on.
Halt an agent
“Halt Falcon — leave its Radar slots alone for now”
Stops new entries without tearing down deployments you still want.
Tip: Halt = pause capital commitment; archive is permanent and should refuse while something is still on duty.
Constrain risk in the ask
“Commission an agent on SOL Breakout with max exposure $100 and 1% sizing”
Risk Management lands on the first save.
Tip: Wider hard stops size smaller notional for the same dollars at risk — say both if you care.
Coverage check
“What’s on duty right now, and where am I blocked?”
Radar / Arena coverage and gate blocks — live vs stuck.
Tip: Run this before commissioning into a full slot list.
Strategy + deploy in one breath
“Build a mean-reversion playbook for ETH, commission a measured agent on it, and put it on Radar for ETH”
Chains author → agent → deploy. Prefer confirmations at each money or authority step.
Tip: One clear outcome per message when testing; chain when you already trust the flow.
Doctor a quiet agent
“Why hasn’t Falcon traded, and what would make Radar fire on SOL?”
Pairs strategy-doctor with Radar activity — qualification, idle reason, edge spent, gates.
Outside the UI (MCP)
Same intents work over MCP from Cursor or other MCP clients. Keep the formula; map Where to tools and scopes. See MCP.
Confirmation hygiene
- Reads — no confirm.
- Strategy apply / agent config — confirm when the surface asks; revision-aware.
- Wager / standing authority / closes — always confirm; say the coin, agent, and dollar impact in the ask so the card matches intent.
← Commander overview · Next: MCP