almanac terminal command can appear as readable activity in the product. That activity is a projection of the command result, not its full payload: it may show at most 50 items and does not replace returned JSON, message bodies, or continuation cursors. Shell scripts and pipelines are not reliably projected as one Almanac operation. Run a single command directly when its activity should be visible, then inspect its actual result for evidence.
When the runtime exposes show_widget, an agent can place a compact original-object card in the conversation. For a returned thread, the tool input looks like this (with a returned account and native thread ID substituted):
kind: "calendar" with a viewer-local YYYY-MM-DD date, optional day/week/month/year view, and optional account; or kind: "connections" with an optional provider. The person can click a card to open the existing side reader or connection UI. The card does not open the pane automatically, authorize an account, send mail, or save a source. Its successful creation also does not prove a full thread was read or every calendar was checked. See email, calendar, and connected accounts for the identities to supply.
A chat claim grounded in a registered source can cite its exact source UUID as [@source-uuid], which renders a source bubble. In a saved knowledge record, a shorter [@key] works only when that record’s citation map binds the key to a source. For example, a fictional Meridian Labs thread might be registered as source UUID 00000000-0000-4000-8000-000000000001: the chat uses [@00000000-0000-4000-8000-000000000001]; a saved page can use [@meridian-email] after binding that key. Sources and citations explains the evidence lifecycle. A card and a citation help the person follow an original; neither substitutes for checking what the original actually says.