SHARPTRADER AI AGENTS (MCP) · FREE AGENT SKILLS
Nine free skills that teach Claude, ChatGPT, Codex and other AI agents to plan SharpTrader non-toxic orders well: what your broker sees of an account, which ordinary-looking trades to add, how many, when, at what cost, and how to check the result.
Short answer: the skills are instruction files that your AI agent reads before it works with SharpTrader’s non-toxic orders. They do not give the agent any new power. The agent still trades only through SharpTrader’s AI agents server (MCP), only on the sessions you mark Active and only inside the limits you set. The skills help it plan within what you allow: read the flow profile, choose a plan, estimate the cost, book a small pilot and check the effect.
On this page
Short answer: a broker judges an account by its whole order flow: holding times, the share of very short trades, buys against sells, win rate, hours and symbols. Non-toxic orders are real trades that your agent books through SharpTrader next to your arbitrage strategies, so that the account’s flow as a whole looks more like ordinary trading. The skills teach the agent how to do that carefully.
Always install the core skill. The other eight are playbooks that build on it, one idea each. Every skill has a guide for people on GitHub with example requests and a worked example.
| Skill | The idea |
|---|---|
sharptrader-non-toxic-core |
Ground rules, reading the flow profile, your limits, cost and capacity, and the plan, book and check loop. Always install it. |
sharptrader-non-toxic-holding-time |
Trades held for minutes to hours, so that fewer of the account’s trades are closed within seconds. |
sharptrader-non-toxic-ambient-flow |
Small background trades in quiet hours, so that the account has an ordinary trading day. |
sharptrader-non-toxic-pacing |
Irregular timing within an hourly budget: no bursts, no fixed intervals, no round seconds. |
sharptrader-non-toxic-order-shape |
Balanced sides, varied sizes, split orders and more than one symbol. |
sharptrader-non-toxic-dilution |
Ordinary trades for each quick winner and a normal win rate, at a cost you agree to. |
sharptrader-non-toxic-indicator-entries |
Entries on visible technical setups, with stops and targets sized to volatility. |
sharptrader-non-toxic-exit-styles |
A mix of targets, stops, trailing stops and time exits, with a broker stop as a safety net. |
sharptrader-non-toxic-activity-cover |
Ordinary trades on the same symbol after bursts of arbitrage activity. |
The skills work only with SharpTrader’s AI agents server and its non-toxic order tools. If you have not connected an agent yet, follow MCP for trading first.
1
2
3
4
Download the repository from GitHub. Its pack.ps1 script builds the upload files in a dist folder on Windows: one plugin with all nine skills, and one zip per skill.
sharptrader-non-toxic-plugin.zip with all nine skills (recommended), or the skill zips you want. Always include sharptrader-non-toxic-core.zip.If Claude never reads the skills, check that code execution (“Create and edit files”) is on in Settings › Capabilities.
Copy the folders in skills to %USERPROFILE%\.claude\skills\, or load the whole set as a plugin with claude --plugin-dir <repository folder>.
Copy the folders in skills to %USERPROFILE%\.agents\skills\, or open Skills in the ChatGPT sidebar and upload the skill zips. To give these apps SharpTrader’s tools, add the server to %USERPROFILE%\.codex\config.toml with the address shown on SharpTrader’s AI agents (MCP) page:
[mcp_servers.sharptrader] url = "http://127.0.0.1:47350/mcp" bearer_token_env_var = "SHARPTRADER_MCP_TOKEN"
Put the access token in that environment variable once and restart the app. ChatGPT in a web browser cannot reach SharpTrader, because its custom connectors need an OAuth sign-in that SharpTrader’s token does not use.
Agents that read Agent Skills (SKILL.md folders) take the folders from skills in their own skills folder. For an agent without skills support, paste the text of the core skill, then of the playbooks you want, into its instructions.
You do not call a skill. Ask in plain words and the agent reads the skills that match before it acts. To be sure, name one: “use the sharptrader-non-toxic-holding-time skill”.
Two read-only requests in the ChatGPT desktop app connected to SharpTrader's AI agents server. The skills use the same tools to look at an account before they plan anything.


The strategies make 60 trades a day on Broker A, and 50 of them close within 60 seconds (83 %). The target is 40 %. With the holding-time skill, the agent works out:
Yes. The skills are published on GitHub. To use them you need SharpTrader with the AI Agents (MCP Server) module, because the agent books the orders through SharpTrader’s tools.
No. The skills add no tools and no permissions. The agent trades only on sessions you mark Active and only within your limits, which it can read but never change. Each order is checked against the limits when the plan is booked and again just before it is sent.
Claude Desktop and Claude Code, the ChatGPT desktop app and Codex, and any agent that reads Agent Skills folders. Agents without skills support can use the text of the skills as instructions.
No. Always install the core skill, then the playbooks you want. Installing all nine as one plugin is the simplest.
No one can promise that. Non-toxic orders change the statistics a broker reads, so arbitrage trades stand out less, but a broker that looks at individual short, profitable trades can still find them. Check your broker’s terms before you switch non-toxic orders on.
Trading with leverage carries a high risk of loss. AI agents can make mistakes, and software does not guarantee profit.