# Inbox

The Inbox is the review gate. Everything an AI hands back, and everything it proposes, waits here for a person. Only what you add to Project Memory feeds later tasks.

It has three tabs: **Messages**, **Drafts** and **Notes**.

## Drafts: results waiting for you

**Drafts** holds results from people and AI, ready for review before anything enters Project Memory. The headline tells you how many items wait for your decision, or that nothing needs you. Filter chips sort them: **New**, **Needs review**, **Accepted**, **In Project Memory**, **Revision requested**, **Rejected**.

Each result shows where it came from (**AI role**, **Outside AI via MCP** or **Manual hand-off**), its text, and the evidence the AI reported with it: commits, branch, test runs, pull request, links. Legevio shows that evidence as reported; it does not check or fetch it.

An owner or admin decides. Others can read results.

| You choose | What happens |
| --- | --- |
| **Add to Project Memory** | The result becomes a record that later tasks receive, filed where the card says (**Files into** …, with **Change**). |
| **Accept result** | The task closes as Approved. The result stays with the task. Project Memory does not change. You can still add it later. |
| **Request revision** | The result stays, and a follow-up task asks for what is missing (**Open follow-up task**). |
| **Reject** | After you confirm, the result leaves review and cannot enter Project Memory. If nothing else on the task is waiting, the task moves to **Blocked** and shows **Result rejected**. A new hand-in starts a new version. |

Results for the same task are grouped, newest first. Once one version is accepted, older versions still waiting close on their own as **Replaced**, so you never review a draft that was already superseded.

A result whose inputs changed while it waited says **Made from an older version of:** and names the record. See [What the AI receives](/concepts/context/#when-the-inputs-change).

At the top of **Drafts**, one line says how results reach Project Memory in this project: only when a person accepts them, or automatically for trusted roles. See [automatic acceptance](/concepts/project-memory/#automatic-acceptance).

### Proposals

An AI connected over MCP can propose instead of acting: a new task, a change to a task, deleting a task, a team role, or a whole team. Each arrives as a card marked as proposed by your MCP client. **Accept** carries it out; **Dismiss** leaves everything as it was. A deletion card warns you before you accept it.

**Handoffs to confirm** holds the next role Legevio proposes when one role finishes. Nothing is assigned or run until you press **Confirm**.

## Messages

One thread per participant, a role or a person, built from the project's activity. You can review a result right in its thread, and a reply there is sent as a revision request.

The **System** thread is read-only: changes to Project Memory, assignments, access, and decisions on hand-ins, such as "accepted your result" or "added your draft to Project Memory".

An AI connected with the **Post messages** permission can post into a thread, optionally about a task. A message changes no task, role or Project Memory.

## Notes

Team notes, with @mentions.

## For your AI

- `submit_result` hands in a result, moves the task to Review and keeps the git references, in one call.
- `list_activity` with `mine: true` shows what was decided about the AI's own hand-ins.
- `post_message` posts into a thread. `propose_task`, `propose_task_update`, `propose_task_delete`, `propose_role` and `propose_composition` create proposal cards.

See the [MCP tools reference](/reference/mcp-tools/).