trading-predictor — Claude Code subagent: what it does, how to install
trading-predictor — a Claude subagent that measures how far its computation runs ahead of market-data delivery, then uses the gap for predictions and risk numbers.
no license ★ 850 updated 2026-10-26
.claude/agents/trading-predictor.mdWhat trading-predictor does
The definition treats distance as time. Hand it a route in kilometres and it reports how long a signal needs to cover it, how long the maths takes, and what margin is left over. The worked example uses 10900 km between Tokyo and New York, and 35786 km for a satellite in geostationary orbit. A margin above 50 ms is the bar its example sets before an arbitrage window counts as real.
Most of the work happens on an MCP server. Five tools are named: predictWithTemporalAdvantage for the prediction, calculateLightTravel for transmission delay, demonstrateTemporalLead for scenarios, validateTemporalAdvantage for checking the figures, and solve for portfolio and risk maths. No server, no numbers, because every call carries the mcp__sublinear-time-solver__ prefix.
Flow Nexus handles heavier runs: sandbox_create raises a sandbox, sandbox_execute runs the trading loop inside it, neural_train fits an LSTM to price history. Claude Flow swarms cover coordination, reconciling signals and risk limits across agents. The file is a prompt, so orchestration stays on those servers.
What it helps with
- measure how much compute time beats data transit time
- check whether a latency gap is wide enough to trade
- reprice portfolio risk on a dense covariance matrix
- train an LSTM on price history inside a sandbox
- reconcile trade signals coming from several agents
Example prompts for the agent
- Work out the temporal advantage for Tokyo to New York at 10900 km and tell me whether the margin is enough to enter.
- Run
mcp__sublinear-time-solver__solveon a 1000×1000 covariance matrix and give me the portfolio weights. - Spin up the
hft-predictorsandbox and start the trading loop inside it.
How to install trading-predictor
Manually: put the files in place
.claude/agents/trading-predictor.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/trading-predictor.tomlCodex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/
Manually: put the files in place
.cursor/agents/trading-predictor.mdCursor also reads .claude/agents/
Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean
Questions about trading-predictor
How do I install trading-predictor in Claude Code?
Copy the agent file to .claude/agents/trading-predictor.md in your project. Claude then picks it up as a subagent.
What does trading-predictor do?
It compares computation time against the time a signal needs to travel, then turns the difference into price predictions, risk figures and arbitrage scenarios. The maths runs on the sublinear-time-solver MCP server.
What does it need in order to work?
The sublinear-time-solver MCP server, since all five of its tools carry that prefix. Sandbox runs and model training go through Flow Nexus.