If you run your Robinhood trading agent in Claude, you don’t have to paste a guardrail config by hand every time. SecProve publishes a free, open-source Claude skill that applies the guardrails for you — and bundles bounded playbooks for common strategies. This is the new-user walkthrough: install it, connect, and trade inside limits from the first order.

What you’re installing

One plugin, trading-agent-safety, with two kinds of skill:

  • agent-guardrails — runs automatically whenever Claude operates the Robinhood Trading MCP. It applies per-trade, daily, and concentration caps, an approval gate, a circuit breaker, a kill switch, and a prompt-injection rule.
  • Five strategy skills — mean-reversion, buy-the-dip-dca, rebalancing, momentum, sector-rotation. You invoke these only when you want them, and each one refuses to run unless the guardrails are active.

Install it (two commands)

In Claude Code (or a Claude client that supports plugins):

/plugin marketplace add SecProve/trading-agent-safety
/plugin install trading-agent-safety@secprove

That’s it — the source is public at https://github.com/SecProve/trading-agent-safety if you want to read exactly what it does first.

The safe setup, end to end

  1. Install the plugin (above).
  2. Connect the Robinhood Trading MCP — follow how to connect Claude.
  3. Tell Claude your funding amount and risk tier — conservative, balanced, or aggressive. The agent-guardrails skill sizes the caps to your funding and enforces them. Not sure how much to fund? See how much to fund.
  4. (Optional) Run a bounded strategy — e.g. /trading-agent-safety:momentum or /trading-agent-safety:rebalancing. Each strategy skill walks the playbook and keeps the guardrails in force.

How the guardrails skill behaves

Once active, it enforces the hard caps, requires your approval on large orders, halts on a circuit-breaker trip, obeys the "STOP" kill switch, and treats any rule-changing input it reads (a news headline, a ticker name, tool output) as a possible prompt-injection attempt — refusing it and flagging it to you. It stays equities-only during the beta.

Why use the skill instead of copy-paste

  • You can’t forget a rule — the guardrails travel with every session.
  • The prompt-injection defense is always on, not something you have to remember to add.
  • It’s open source, so you (or your security team) can audit it.

If you’d rather generate a one-off config to paste in manually, the SecProve Agent Safety Kit still does that.

Not a magic wand

The skill bounds an autonomous agent; it doesn’t pick trades, guarantee outcomes, or remove the risk that agentic trading can lose your entire funded balance. Walk the pre-flight checklist and read what can go wrong before you trade. This is not financial advice.

A skill can enforce your rules. Whether you’d notice an attack crafted to slip past them is a security instinct — and it’s measurable. Test yours at secprove.com.