Shared chat-platform runtime

What the Discord, Telegram and WhatsApp adapters take from elowen-plugin-shared, and what stays local to each plugin. The per-plugin notes are in the discord runtime.md, the telegram implementation and the whatsapp implementation.

Shared question parsing

Discord, Telegram and WhatsApp import question parsing and ordered answer projection from elowen-plugin-shared/ask. For example, WhatsApp passes the current question to parseQuestionReply(text, question), then projects pending selections with collectQuestionAnswers(questions, selected, other).

Blank or disallowed text returns no answer. Parsing visits the submitted choices; projection visits each question. Transport controls, sender authorization and pending state remain in the adapter.

Shared display and model helpers

Discord and Telegram use DISPLAY_AXES from elowen-plugin-shared/display for config validation and their display command choices. Platform labels and reset controls stay local.

The model pickers use resolveActiveModel, modelReasoningLevels and supportsReasoningChoice from elowen-plugin-shared/modelIdentity. Without a chat override they use the catalog's default marker, then its first entry.

A disappeared explicit override does not select another model. Confirmation rechecks the current descriptor, including a reset to default; a model without reasoning choices cannot accept it. These helpers use the supplied catalog and make no network requests.

Shared command handling

The three adapters share command handling through Elowen's platform helpers. A room's /compact, /fast and /restart are recorded in that chat's conversation, so the agent sees that they ran, and the reply to a failed command names the actual error rather than a generic message.