Diagnose a quiet agent before asking it to trade

A quiet agent needs a stage-specific diagnosis. First determine whether it completed a review and chose to wait, failed to complete a turn, had an action rejected, or submitted an order that remains unfilled. We would answer that question before increasing activity or loosening limits.

The current activity guide distinguishes those states. Its scope is the live Hyperliquid alpha inspected October 3, 2026. A lack of recorded fills alone cannot tell you which state occurred.

Three hours with no new trade

Illustrative scenario: an owner sees no trade for three hours and thinks automation stopped. The last completed turn says the required breakout was absent. Another recent entry shows a policy rejection. A third shows a submitted order without a recorded fill.

These records point to three different explanations. The first is a strategy decision. The second is an admissibility problem whose reason needs review. The third is an execution-state question. Asking “trade more” collapses them into one behavior change and could make the original diagnosis harder.

Start with the agent's saved status and schedule. Then inspect recent Activity entries and their timestamps. If there is a completed waiting decision, ask which required conditions were missing. If there is a failure, preserve the error details and identify the stage that produced them.

Follow the narrowest evidence

For readiness problems, review the selected account, balance, trading approval, and fee approval. For policy rejection, compare the requested action with the relevant saved limit. For submitted orders, inspect the venue outcome and compare Trades and Positions. These observations can reveal a problem without creating another order.

An account-wide position cap can block a new symbol because a manual position already occupies a slot. A same-symbol re-entry cooldown can block reopening after a full close. The settings reference explains these current controls. Their effect differs from a model declining a setup.

Our execution and settlement article explains why acknowledgement and fill require different evidence. Our trace feedback guide connects a failure to the stage where a repair belongs.

Recover without changing the thesis accidentally

If the issue is a missing approval, resolve that readiness dependency. If it is an unintended policy configuration, review a precise settings proposal. If the strategy's entry conditions genuinely never occur, consider whether the strategy needs revision, with the owner approving the change deliberately.

A manual turn request is a separate action and may lead to trading. A diagnostic chat conversation by itself does not trigger a turn, according to the documentation. Use that distinction to gather evidence before choosing an execution-affecting step.

A useful diagnosis ends with a concrete statement such as “completed and waiting for evidence” or “opening rejected by the saved position cap.” That statement gives the owner a specific decision while preserving the option that waiting was correct.