Problem
The chat empty state shows three starter-prompt chips (packages/views/chat/components/chat-window.tsx, STARTER_KEYS: list_open / summarize_today / plan_next, resolved from the i18n chat namespace). The set is global: every agent in every workspace gets the same three developer-centric prompts.
In multi-agent workspaces where agents have sharply different roles (a routing front door, a domain conductor, a reviewer), the generic set misleads users about what the selected agent is for - e.g. "List my open tasks by priority" presumes the human executes tasks, which inverts the model in workspaces where agents execute and humans review.
Proposal
Make starter prompts configurable per agent, falling back to the current i18n set when unset:
- an
agent.starter_prompts field (array of up to 3 strings, optionally with an icon), editable in the agent settings UI,
- chips render the agent's set when present; tapping still prefills the composer as today.
Per-agent prompts are user-authored strings persisted to the DB, so they need no i18n plumbing - the same reasoning as helper-starter-prompts.ts, which already persists prompt strings at write time instead of reading the i18n bundle.
Why per agent, not per workspace
The prompt that makes sense depends on the selected agent's role, and the empty state already re-renders per agent ("Hi, I'm {agentName}"). A workspace-level set would inherit the current problem one level down.
Problem
The chat empty state shows three starter-prompt chips (
packages/views/chat/components/chat-window.tsx,STARTER_KEYS: list_open / summarize_today / plan_next, resolved from the i18nchatnamespace). The set is global: every agent in every workspace gets the same three developer-centric prompts.In multi-agent workspaces where agents have sharply different roles (a routing front door, a domain conductor, a reviewer), the generic set misleads users about what the selected agent is for - e.g. "List my open tasks by priority" presumes the human executes tasks, which inverts the model in workspaces where agents execute and humans review.
Proposal
Make starter prompts configurable per agent, falling back to the current i18n set when unset:
agent.starter_promptsfield (array of up to 3 strings, optionally with an icon), editable in the agent settings UI,Per-agent prompts are user-authored strings persisted to the DB, so they need no i18n plumbing - the same reasoning as
helper-starter-prompts.ts, which already persists prompt strings at write time instead of reading the i18n bundle.Why per agent, not per workspace
The prompt that makes sense depends on the selected agent's role, and the empty state already re-renders per agent ("Hi, I'm {agentName}"). A workspace-level set would inherit the current problem one level down.