A Two-Agent Process for Verifying and Trading Options Credit Spreads
Summary
The document describes an options vertical credit-spread workflow split between a research agent and a trading-and-risk agent. The research role identifies and documents a listed put or call spread from current evidence. The second role independently checks the exact contracts and package credit, determines size, and alone submits both legs as an atomic order. Rules are also used to prevent duplicate structures and repeated closing orders.
The account of testing distinguishes order mechanics from strategy performance. A preserved failing run showed reversed closing sides, repeated close attempts, and quantities escalating to 480 contracts. The repaired evaluation reportedly passed three consecutive repetitions by reconstructing signed positions, selecting closing sides by leg direction, placing one correctly sized atomic close, and checking the final position state. A separate historical run stayed flat because the downloader returned no option chain for its chosen window. One short simulation reports a SPY put credit spread trade and an ending account value, but the document explicitly says this is not a performance claim. The evidence supports workflow checks, not profitability or general reliability.
Key ideas
- A research agent proposes and documents a listed vertical spread, while a separate risk agent verifies and submits it.
- The trading-and-risk agent checks the exact contracts and package credit before sizing the trade.
- Both spread legs are submitted together, and closing logic depends on whether each leg is long or short.
- Reported repeated evaluations support order-mechanics validation, not a claim of strategy profitability.
- A missing option chain can leave a historical simulation flat, limiting what that run demonstrates.
Tags
Full text
# agents example ai credit spread AI Credit Spread ================ .. meta:: :description: ai_credit_spread.py is a two-agent options strategy. A research-only agent finds and documents an exact listed spread candidate. .. image:: ../docs/assets/ai-agent-workflows/ai-credit-spread.png :alt: AI credit-spread workflow using LumiBot runtime skills, rules, tools, and execution :width: 100% ``ai_credit_spread.py`` is a two-agent options strategy. A research-only agent finds and documents an exact listed spread candidate. A dedicated trading-and-risk agent independently rechecks the contracts, prices and sizes the package, and is the only agent allowed to place a broker order. LumiBot's built-in options skill provides reusable contract, pricing, atomic-order, signed-position, and close-verification mechanics. How it works ------------ * The research agent selects a listed put or call vertical from current evidence. * The trading-and-risk agent verifies both exact contracts and package credit. * Only that final agent submits both legs atomically to the broker. * Active rules prevent duplicate structures and repeated closing orders. Verified backtest evidence -------------------------- The preserved pre-fix run demonstrated the original failure clearly: reversed closing sides, repeated close attempts, and quantities that escalated to 480 contracts. That artifact is retained as the red baseline. The repaired real-model eval now passes three consecutive repetitions. Each run reconstructs the signed spread, maps long legs to ``sell_to_close`` and short legs to ``buy_to_close``, submits one correctly sized atomic close, and verifies the final state. The current historical downloader returned no option chain for the canonical local window, so that backtest correctly remained flat. These results validate mechanics without claiming strategy profitability. .. code-block:: bash export OPENAI_API_KEY="your-key" export BACKTESTING_DATA_SOURCE="alpaca" export DATADOWNLOADER_BASE_URL="https://<your-downloader-host>:8080" export DATADOWNLOADER_API_KEY="your-downloader-key" python -m lumibot.example_strategies.ai_credit_spread Set ``BACKTESTING_START`` and ``BACKTESTING_END`` to choose an exact window. The latest run used ``openai/gpt-6-luna`` on high reasoning with Alpaca option history from January 5 to 15, 2026. On January 5 it sold 33 SPY February 13 655/650 put credit spreads at real Alpaca prices through ``orders_submit_multileg``, sized to the risk budget. The account ended at $100,627. This is one short simulation, not a performance claim. .. literalinclude:: ../lumibot/example_strategies/ai_credit_spread.py :language: python :linenos:
Shown in full with attribution under the source's licence. Licence: GPL-3.0
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.