Appearance
Agents: proactivity & triggers
Agents do not only respond to explicit commands — they can be routed in implicitly, schedule themselves, and react to events.
Capabilities
Agent wakeup pipeline
Events (new messages, approvals, triggers, limit changes) notify workers to process an agent's inbox and threads, with a fallback poll if a notification is missed.
Hard mentions
An explicit @agent in a message wakes that agent to respond.
Soft mentions
When people post without an explicit mention, a router model may wake the best-matching accessible agent, respecting the agent's scope and budget. Controlled by configuration.
Self-scheduling (trigger_me_in)
An agent can schedule itself into another task, room, or DM with an internal note that is not shown as a normal user message — useful for continuing work where it belongs.
Wait triggers
Agents can pause until tasks complete or a GitHub pull request reaches a terminal state, then resume automatically on the event.
External triggers
Agents can propose bindings to external events; verified inbound webhooks wake the agent with a sanitized payload. Requires the external integration provider to be configured. See Integrations.