Shows you, above the prompt, the critical decisions the agent made without being asked: red to rethink, yellow to know, each with a rethink button. Works out of the box for any kind of project; /critical-areas auto lets the agent learn what is critical in yours.
Shows whether the session's prompt cache is warm or cold: a pixel-art fire that shrinks each quarter of the TTL, a countdown, and a sparkling snowflake once it expires.
Watches the chat's context size: shows it from 200k; at 250k asks the model to propose splitting the rest of the work at the next transition point if something big and separable is left; at 400k, to propose a focused /compact, a new chat or carrying on. Each warning fires once per level and resets after /compact. /foco-compact hands you the compact text.
A pane listing, per project, the chats that ended waiting on you (asking for an OK, a test or a decision), with how long ago and whether their cache is still warm; dismiss each one.
A pane that opens whenever the agent or a subagent calls an MCP tool, saying in plain words which MCP was called, what it was asked and whether it finished.