# Checkpoints and debriefs

> Answer By Dawn's agent when a mission step stops for you, from approving a plan to signing off a pull request, and read the debrief a finished mission leaves.

This page helps you answer By Dawn's agent when its work stops for you, and read what a mission did once it ends. By
the end you'll know each kind of question the agent can put to you, what each answer does, and how to decide
whether a ticket is done.

A **checkpoint** is a mission step waiting on a person: a plan to approve, mockups to pick from, a pull request to
sign off. The agent (the **worker** in the app) stops at it and waits for your answer, so nothing past it happens
until you give one. Which steps pause for you is chosen when the ticket starts (see [Start a
ticket](https://docs.bydawn.com/start-a-ticket/#choose-the-steps)).

## Where checkpoints show up

On the ticket, the mission strip above the agent's session turns to **Needs you**, and the checkpoint opens as a
card just above the message box, next to the run it's about. On a phone it opens as a sheet, and **Answer the
checkpoint** brings it back if you close it.

You don't have to be watching the ticket. Each checkpoint also lands in your inbox under **Needs you**, with a
button that names the job, such as **Review plan**, **Pick**, **Review PR** or **Answer**. It opens the ticket at
the card.

To judge a plan you often want to read what the agent did first. Fold the card down to read the run behind it, then
open it again with **Review**. Folding doesn't answer anything, so the agent keeps waiting.

## Approve the plan

The card shows **The plan on this ticket** in full, and gives you three answers:

- **Approve plan** agrees the plan. What happens next depends on how the ticket was started, and the line under the
  button says which: "Ends the planning run; you start the build." for **Plan first**, "Starts the build to this
  plan." for **Plan + build**, and "Unblocks the build." for **Build now**.
- **Request changes** opens a note, "What should change?". **Send back** sends it, and the agent revises the plan
  and asks again.
- **Reject** asks "Why reject it?". It stops the work on the ticket, and the agent won't build it or ask again.

Press `⌘+Enter` (or `Ctrl+Enter`) to send an open note.

Until the build starts you can change your mind. **Reopen plan approval**, in the ticket's **Mission** section, puts
the question back, and the build then follows the plan you agree next.

## Pick a mockup

**Show me mockups** asks the agent to draw a few designs before it builds a screen. The card shows them side by
side ("3 mockups — pick the one to build."). Choose **Pick this one** on the one you want, or press its number key,
then **Build this one**. **View larger** opens a mockup full size, and a note is optional.

If none of them is right, choose **None of these** and say "What should the new mockups change?". The agent draws
new mockups to your note.

When the agent can't get the built screen to match the mockup you picked, **Match the mockup** shows the two side by
side with the differences that are left:

- **Accept as is** passes the step with those differences.
- **Keep going** asks the agent for another round of fixes.
- **Send back** gives the agent your note to work from, so a note is required.

## Sign off the work

**Your sign-off** comes after the agent has finished. By Dawn raises it, not the agent, and it lists the pull requests
and what By Dawn will do with them, such as "SAW approves and merges".

- **Ship it** lets By Dawn carry out the steps after your sign-off, for example approving and merging the pull request,
  then moving the card to Done. If the mission doesn't include merging, the card says merging is up to you.
- **Send back** gives the agent your note. By Dawn asks you again once the agent has pushed new commits.

A note on its own, with no option picked, sends the work back.

Other steps that ask for your approval read "The worker is waiting on your sign-off" and work the same way:
**Approve**, with an optional **Add a note**, or one of the two ways to say no.

## Approve a split into tickets

When a planning run proposes breaking the work into tickets, the card shows the **New epic**, its **Issues** and
which issue is **Blocked by** which. **Create the tree** creates the epic and puts every issue under it. **Let me
reshape it** sends it back with your note.

## Answer a question

The agent can also stop to ask you something no step asked it to. The card says "The worker raised this — no step
asked it to", with the options it offers. Choose one and **Answer**, which unblocks the agent. Where the agent
allows it, you can type a steer instead ("Or steer it — …").

## When a check fails

If a check keeps failing and the agent can't clear it, the card reads "A check failed and the worker can't clear
it", with the last lines of output, and the agent waits until you decide:

- **Retry now** runs the check again, for example after you've fixed something outside By Dawn.
- **I can't run this here** is for a check that can't pass in this setup. Say why in one line, then **Record as
  Reported & continue**. The step is marked **Reported**, not Verified, and logged under your name.

## Who answers a checkpoint

Most checkpoints wait for a person. Some steps let whoever starts the ticket choose **Who approves**: **A person**,
**SAW reviews, then a person**, or **SAW reviews and approves**. With **SAW reviews, then a person**, By Dawn's own
reviewer reads the work against the step first and a person gives the answer. With **SAW reviews and approves**, the
reviewer answers on its own. Its answers are always marked as **SAW's reviewer**, in the debrief too, so they're
never mistaken for a person's.

## The debrief

When a mission ends, the ticket replaces the live strip with a **debrief**: the mission as a record of what
happened. It shows:

- the mission's name and how it ended, such as "Mission 1 · Plan" and **Completed** or **Abandoned**, with how long
  it ran;
- a count of its steps, such as "12 steps · 7 done · 2 by SAW · 2 decided · 1 skipped";
- whether the plan was approved ("snapshot on the ticket — the build works to this"), sent back or rejected;
- the decisions people answered, by name;
- for an abandoned mission, the required steps that were **Unmet when it ended**.

**Read the receipts** opens each step's proof, at the level it earned: Verified, AI-reviewed or Reported (see [How
By Dawn shows what was checked](https://docs.bydawn.com/missions/#how-by-dawn-shows-what-was-checked)). Earlier missions on the same ticket stay
readable from **Earlier missions**, as they stood when they ended.

## Is this ticket done?

Under the debrief, By Dawn asks **Is this ticket done?** It doesn't guess:

- **Done** moves the card to your board's last column.
- **Needs more work** opens the picker for the next mission, so you can start it.

After an approved plan, **Needs more work** is already chosen and a build is drafted from the plan, because that's
almost always what comes next. After a build, nothing is chosen. After an abandoned mission, nothing is drafted
either: By Dawn asks you to "Pick what the next mission is for."

## FAQ

**What happens if nobody answers?**

The agent keeps waiting, and the ticket stays held at that step. Nothing past it runs, so a checkpoint never times
out into a yes.

**What's the difference between sending back and rejecting a plan?**

**Send back** is a revision: the agent changes the plan and asks you again. **Reject** ends the work on the ticket,
and the agent won't ask again.

**Can I answer from chat?**

Yes, if your chat account is linked with the **Answers** scope. See [Connect your chat account](https://docs.bydawn.com/chat-accounts/#scopes).

## Where to go next

- [Missions and templates](https://docs.bydawn.com/missions/): where the steps that pause for you come from.
- [Start a ticket](https://docs.bydawn.com/start-a-ticket/): choosing which steps pause for you before the work begins.

Source: https://docs.bydawn.com/mission-checkpoints/