Proposed cards

When the chat agent thinks something deserves to become an action item, it does not drop it on your action board. Instead, it lands in the Proposals inbox.

This is deliberate. The agent is good at noticing patterns; it is not good at judging your priorities. You stay in the loop on every card.

The lifecycle

A card moves through four states:

• Proposed — drafted by the agent, parked in the inbox.
• Active — promoted by a human; visible on the action board.
• Done — completed.
• Rejected — the human decided not to act on it. Restorable.

Transitions are atomic and guarded by state, so two reviewers acting on the same proposal at the same time get a clean conflict response rather than silently overwriting each other.

Reviewing a proposal

Open any card in the Proposals inbox. The side panel shows:

• Title and description — both editable on the way to promote.
• Tier — the agent's recommendation, also editable.
• Why Facturn proposed this — a short motivation written by the agent.
• Contributing signals — the original messages, tickets, or notes that led to this card. Each one is clickable.

You then have two choices:

1. Promote to action items — sends the card to the action board with any edits you made.
2. Reject — optionally with a reason. We recommend a one-line reason when you have one, even something short like "we already shipped this" or "tier is wrong, this is a Later not a Now."

How rejection feedback works

Reject reasons are workspace-scoped and the most recent ones (newest first, capped to a small token budget) are injected into the system prompt for the next propose_task call. In practice this means the agent stops proposing cards you already said no to.

Rejected cards are not gone forever — they are visible in the action board with state=rejected and can be restored if you change your mind.