Using robinhood agent

Using the Robinhood Agent

Overview

This is the entry point for the pack. It explains the trading lifecycle, routes a request to the right skill, and states the rules every skill must follow. The agent proposes trades with a strategy swarm, but an independent risk guardian has the final say. The default execution mode is paper (simulated) trading.

When to Use

• At the start of any trading or portfolio request.
• When unsure which strategy or stage applies.

Lifecycle

Run the stages in order. Each stage has a dedicated skill.

1. Intake — portfolio-intake: validate and normalize the account, positions, goals, and market snapshot.
2. Enrichment — market-enrichment and company-news: add derived metrics and a deterministic news-sentiment summary.
3. Macro regime — macro-regime: classify the US tape as risk-on, neutral, or risk-off (advisory; never changes a decision).
4. Strategy swarm — run every enabled strategy independently:
   • rebalancing-strategy
   • concentration-analysis
   • momentum-strategy
   • mean-reversion-strategy
   • thematic-strategy
   • dca-strategy
   • company-news
   • relative-strength-strategy
5. Risk — risk-guardian (plus market-hours-guard and wash-sale-guard): evaluate every proposal against the risk limits.
6. Decisioning — trade-decisioning: allow, require approval, or block.
7. Execution — order-execution: place allowed orders (paper by default).
8. Reporting — trade-reporting: produce the audit-ready report with the regime and a one-day alpha estimate versus SPY.

Process

1. Identify the request type and locate the run input (account, positions, goals, market snapshot).
2. Run intake and enrichment to build the trading context.
3. Run every enabled strategy; each emits explainable signals and proposals.
4. Evaluate each proposal with the risk guardian, apply the decision policy, place allowed orders, and produce the report.

Operating Rules

1. Every decision must be explainable and cite its risk findings.
2. Scoring and decisions are deterministic. A language model may enrich the narrative only; it must never change a score or a decision.
3. The risk guardian has final say. A blocking finding vetoes the order and cannot be overridden by a strategy's conviction.
4. Use pseudonymous account identifiers and synthetic market data. Never request or store real brokerage credentials in inputs.
5. Paper mode is the default. Live trading requires BOTH EXECUTION_MODE=robinhood_mcp AND LIVE_TRADING_ENABLED=true, and orders still pass the risk guardian and any human-approval gate.
6. Collective memory is optional and off by default. When enabled, recall and feedback remain deterministic, the memory signal is advisory, and threshold calibration is advisory only.
7. The macro-regime read is advisory. It frames strategy proposals (for example, it suppresses new relative-strength buys in a risk-off tape) but never changes a risk finding or a decision.

Outputs

A routed workflow and a shared rule set that every downstream skill honors.

Verification

• The correct lifecycle stage and skill are selected for the request.
• The operating rules above are applied throughout the run.