> ## Documentation Index
> Fetch the complete documentation index at: https://docs.abacuslabs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Guidance for agents

> How an agent should use Abacus Labs' output with a person.

* Show the person the options and their trade-offs. Do not pick for them.

* Abacus Labs places no orders. Place them at the person's broker only after the person approves one option, and apply the lot instructions on each sell.

* Commit the approved option, so Abacus Labs remembers its loss sales for later wash-sale checks.

* Send every account the person has, with lots, cost and dates when the broker gives them, and recent trades. Wash-sale checks only see what you send or saved.

* Send what you have and leave out what you do not. Read `warnings`: they say which numbers rest on missing data.

* If the outcome is `clarify`, ask the person `understood.question`. If it is `infeasible`, tell them which rule could not be met.

* Options expire after 15 minutes. Right before trading, run [`POST /v1/trades/check`](https://docs.abacuslabs.ai/concepts/trade-checks) on the chosen trades with fresh holdings.

* A monitoring event is a prompt to look, not an instruction to trade.

* A cap on gains is `tax.max_gain`; a cap on tax is `tax.max_tax`. Both are net: losses the option realizes offset its gains (each sell's own gain is in `trades[].realized_gain`). If the person may mean gross gains, show both. Always read `understood.policy` back to the person: it is exactly what was solved, and a rule Abacus Labs could not read is not in it.

* Abacus Labs gives no investment advice. Say that the options come from rules the person set, not from a recommendation.

The onboarding note at [`https://api.abacuslabs.ai/agent.md`](https://api.abacuslabs.ai/agent.md) is written for agents to read directly. [`/llms.txt`](https://api.abacuslabs.ai/llms.txt) lists every page.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.