# Waiting, assistance and completion

Wait-for polls a read query until its path equals a typed literal. The operations recipe uses equals type="boolean" value="true"; omitting the type would compare a string. path is relative to response data. intervalSeconds and timeoutSeconds configure polling; omitted values default to 3 and 120 respectively.

Timeout and query failure present a failed stage with retry. Leaving or suspending a run must stop its active work; returning can restart it. Polling is not a durable background job. Do not rely on it continuing after a page closes. A wait stage has equals, not capture children; fetch details separately when needed.

Agent steps require a configured LLM in AiAdapters for AI execution. Configure the LLM before offering an AI-assisted Journey; an agent element or hint does not provision a model. The runner also needs an assistant integration to use that configured LLM.

Agent declares a goal and may provide manual-fallback containing body and checks. The CoCoCo app currently shows that fallback. Without it, assistance is shown as unavailable. A hint supplies assistant context and canSuggest metadata; it does not create an AI integration by itself.

Done is the terminal stage. It can contain body and checks, using the same safe text interpolation as info. Completion clears the user's saved progress. The Journey preview does not save progress.

Knowledge-capture is a root-level text template evaluated with run data. The CoCoCo app offers an editable local note and copying. There is currently no persistence endpoint behind that control. Copy it to the intended system if you need to retain it.

Progress is saved in the current browser only. It does not follow the user to another device and is not a durable server-side workflow. Keep journey and step IDs stable, and check that in-progress runs still resume correctly after changing a published journey. Progress saved by the older JSON-based Journeys does not carry over.