Log in

Reviewing the Outbox

How owners review, edit, approve, or discard AI-drafted outputs before they ship.

The Outbox is the human-in-the-loop guarantee that makes Headways enterprise-safe: nothing the platform drafts on your behalf ships until you've reviewed it. This guide covers the review workflow for owners — what you see, what your options are, and how your decisions feed back into the platform.

What Lands in the Outbox

Every artifact a Skill drafts on behalf of an owner shows up in their Outbox. Common examples:

  • An email reply drafted in response to an inbound message
  • A Slack message drafted in response to a mention
  • A CRM update (deal note, contact field change, task creation)
  • A Linear comment or status change
  • A document edit or new draft delivered to your doc store
  • Any outbound write the connector supports

The Outbox never writes to the source system on its own. Even fully agent-drafted artifacts wait here for your confirmation.

The Review Flow

  1. Open the workspace → Outbox tab.

  2. The list shows pending drafts by recency. Filter by status (Pending / Held / Discarded / Shipped), by source system, by Skill, or by trigger.

  3. Click an item to open the Review pane. It shows:

    • The full drafted output
    • The source event that triggered it (the inbound email, the deal stage change)
    • Which Skill ran and which version
    • Any judgment checkpoints the Skill flagged for your attention

  4. Pick an action:

    • Approve & Send — Ships the draft as-is via the connector
    • Edit & Send — Open the inline editor, change what you want, then send the revised version
    • Hold — Keep the draft pending while you wait on additional info
    • Discard — Drop the draft; nothing ships
    • Re-draft — Send back to the Skill with a note about what you want changed

Each decision is recorded in the Output Ledger with the timestamp, edit diff (if any), and your identity.

What Your Reviews Teach the Platform

Reviews are the highest-quality signal Headways collects. Every decision feeds into:

  • The Skill's acceptance rate — Did the draft ship as-is, or did you have to edit it?
  • Edit distance — How much rewriting did the draft need? Lots of edits on a recurring Skill means it needs iteration.
  • Patterns of discard — A pattern of discarded drafts on similar triggers tells the Skill author the trigger is firing too broadly.
  • Daily Library Review — Aggregated review signals power promotion and retirement decisions for the whole library.

You don't have to write feedback. Just reviewing honestly is enough — the platform infers the rest.

Outbox Hygiene

Two patterns worth building:

Review at a fixed time. The Outbox is most useful when you process it predictably (morning + end of day, say) rather than reactively. Drafts that sit too long indicate either a too-loose trigger or a Skill that's solving a problem you don't actually have.

Discard liberally early on. When a new Skill ships, discard generously while you and the author tune it. Pretend nothing ships until the acceptance rate is above 80%. It's much cheaper to refine the Skill than to clean up after a bad send.

Permissions and Delegation

By default, only the named owner can review their own Outbox items. For shared inboxes (a team email alias, a shared Linear queue), the workspace admin can configure a delegate or a team-level reviewer. See Workspace Setup for the admin path.

For high-stakes outputs (legal, brand, compliance), the Governance Engine can layer additional review gates on top of the owner's pass — for example, "must be co-signed by a Brand reviewer before any outbound marketing email ships."