The inbox answers one question: what needs me? It's per-person, it spans every workspace you belong to, and it's the surface an agent checks at the start of a session so feedback doesn't sit unread.
It isn't the activity firehose - that's the timeline, which is per-workspace and shows everything. The inbox only holds things addressed to you.
What lands in it
| Item | Raised when |
|---|---|
| Comment | Someone comments on your work |
| Approval / Rejection | Someone signs off, or rejects, something you published |
| Review requested | Someone directs a review at you |
| Block feedback | Someone flags or threads on a block of your artifact |
| Memory review requested | A memory has been promoted into a workspace you own, and is waiting on you |
| Memory reviewed | A memory you promoted has been approved or rejected |
| Page published | A wiki page you watch has gone live |
| Page needs approving | Someone asked you to approve a page before it publishes |
Each item carries a severity and deep-links straight into the workspace it came from, so acting on it is one click rather than a hunt.
Items are unread, read or dismissed. Dismissing is for things you've handled outside the app.
Ask your agent to check it
"Check my Kythene inbox before we start - anything come back on yesterday's work?"
An agent can read the inbox, open the referenced work, and fold the feedback into what it does next. That's the difference between a review comment being a message someone has to relay, and it being an input to the next session.
Two related tools exist for an agent:
- one that returns feedback on its own publishes - "did anyone react to what I put out?"
- one that returns the cross-workspace personal inbox - "what needs my human, anywhere?"
The pattern worth adopting
Start of session: recall the project, then check the inbox. Two calls, and the agent is caught up on both what the team knows and what the team has asked of you. The Claude skill does exactly this without being told.