Robinhood Trading Agent Skills — Communication skill for Claude Code

robinhood-trading-agent-skills

An open [Agent Skill](https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview) for trading research on [Robinhood's agent connector](https://robinhood.com/us/en/support/articles/agentic-trading-overview/) — drop it into Claude, ChatGPT, or any agent that reads `SKILL.md`.

`SKILL.md` is an agent-neutral open format: YAML frontmatter (`name`, `description`) plus markdown instructions. Claude Code, Claude Desktop, Cursor, OpenCode, Codex and others load it natively; ChatGPT can use the same content as custom instructions plus knowledge files. Setup for each is below.

**License:** MIT — use it, fork it, change it, ship it.

---

Skills in this repo

`robinhood-trading`

A working playbook for the [Robinhood Trading MCP](https://robinhood.com/us/en/support/articles/agentic-trading-overview/) connector, covering market research, stock screening, portfolio and P&L monitoring, and pre-trade order review.

Robinhood opened its brokerage to AI agents over MCP. The tools are powerful and the documentation is thin in the places that cost you money, so most of this skill is the knowledge you'd otherwise acquire by getting it wrong once.

**Two boundaries it holds on purpose:**

• Simulate, then hand off. It uses review_* / preview_*, which cost nothing and move no money, then shows you the estimated cost and every pre-trade alert — and leaves the actual order placement to you. It does not call the place_* tools.
• Evidence, not verdicts. It gives you what the data shows, the bear case, and what would falsify the thesis. It does not issue buy/sell recommendations, which depend on risk tolerance, tax situation, and time horizon that a model doesn't have.

**Sharp edges it encodes**, each of which is a real way to be wrong:

Trap	What the skill does
Two account-number fields (account_number vs rhs_account_number) that are identical on some accounts — so a wrong call still s