- Home
- Blog
- PM Artifacts
- Acceptance criteria examples a coding agent can check
Acceptance criteria examples a coding agent can check
Vague criteria let a coding agent decide what done means. Examples rewritten as checks an agent can run and you can confirm in minutes.
On this page
Good acceptance criteria examples share one trait: each line describes something you can observe and mark pass or fail. “Rescheduling should be easy” can’t be checked. “After the client confirms Thursday at 14:00, the booking shows Thursday 14:00 and the old slot is free” can, by you or by a coding agent.
- A criterion is checkable when it names a starting state, an action and a visible result.
- Cover empty states, errors, permissions and limits, not only the happy path.
- Write a number only when you can test it against known data.
- Ask the agent to report how it checked each criterion, with the output.
Name a starting state, an action and a result
A coding agent stops when the work looks finished. If your criteria are vague, finished means whatever the agent decides. If each criterion names where things start, what the person does and what they see afterwards, the agent can test its own work and you can confirm it quickly.
Three signs give a weak criterion away. Two reasonable people could disagree about whether it passed. It leans on words like easy, fast, intuitive or properly. Or it never says what the screen or the data looks like after the action.
Clients can reschedule appointments easily.
Given a booking tomorrow at 10:00, when the client picks Thursday at 14:00 and confirms, the booking shows Thursday 14:00 and the 10:00 slot is open to other clients.
The second version tells the agent what to build and how to prove it works. It also tells you exactly what to click when you review.
Acceptance criteria examples, rewritten
For example, a founder building a booking tool for dog groomers is adding rescheduling. Clients book a monthly groom, mostly from a phone. Here is how each vague line from a first draft becomes a check.
| Vague | Checkable |
|---|---|
| Show available times | The picker lists only free slots in the next 14 days, in the shop’s time zone |
| Handle the empty case | With no free slots in 14 days, the picker says “No free times in the next two weeks” and offers the waitlist |
| Handle conflicts | If someone takes the slot first, the client sees “That time was just booked” and the original booking stays as it was |
| Only the right people | A client can move only their own bookings; a request for another client’s booking returns 404 |
| Respect the shop’s policy | Rescheduling closes 24 hours before the start time; the button is disabled and says why |
| Notify the groomer | The groomer gets one email per change, naming the old and new times |
| Should be fast | With 2,000 bookings in the seed data, free slots appear in under 2 seconds on the test machine |
Notice what each rewrite adds: a condition, a visible result, and often the exact words on screen. Exact copy is worth writing. It removes a whole category of back-and-forth about wording.
- Easy to reschedule
- Handle errors properly
- Should be fast
- Only the right people
- New slot shows after confirm
- Taken slot shows a message
- Loads in under 2 seconds
- Other clients' bookings return 404
Use Given, When, Then for behaviour and a checklist for rules
Two formats cover nearly everything. Given, When, Then suits behaviour: a starting state, an action, a result. A plain list suits rules that hold everywhere, such as who can see what or which fields are required. Coding agents read both without trouble, so pick whichever makes the line clearest.
Story: Client reschedules a booking
For: dog owners who book a monthly groom, mostly on a phone
AC1 Given a booking tomorrow at 10:00
When the client picks Thursday 14:00 and confirms
Then the booking shows Thursday 14:00
And the 10:00 slot is open to other clients
AC2 Given the new slot was taken a moment ago
When the client confirms
Then they see "That time was just booked"
And the original booking is unchanged
Rules
AC3 Rescheduling closes 24 hours before the start time
AC4 A client can move only their own bookings
AC5 The groomer gets one email per change
Keep each Then to one result you can see. Extra “And” lines are fine. A Then that needs a paragraph is really two criteria, so split it.
Write a number only when you can test it
“Fast” invites the agent to pick its own bar. A number fixes that, but only when the test is concrete: how much data, on what machine, measured how. “Under 2 seconds” with no seed data is still a guess dressed as a target.
Note. If you have no baseline yet, write the criterion as a measurement to take: “Log how long the slot picker takes with 2,000 bookings.” Set the target once you know the starting point.
Cover the states a quick build skips
Agents build the main path well because it’s what you described. The gaps sit in the states nobody wrote down. Run this list against every story before you hand it over.
- What the screen shows when there’s no data yet.
- What happens when the action fails or the connection drops.
- Who can’t do this, and what they see if they try.
- What happens at a limit, such as a time window, a file size or a count.
- What changes elsewhere: emails, totals, other people’s views.
- What must stay the same after the change.
The last line matters most once you have users. “Rescheduling keeps the original price, even if prices changed since booking” protects people who already rely on the product. Without it, an agent may recalculate the price because that looks tidier.
Ask the agent to prove each criterion
Checkable criteria are half the job. The other half is asking for evidence instead of a summary. “All criteria met” tells you little. A test name and its output for AC2 tells you a lot.
- Number the criteria. The agent’s report and your review then use the same labels, AC1 to AC5.
- Ask for one check per criterion. A test, a script, or a browser check with what the agent saw on screen.
- Read the report against the list. Treat any criterion without evidence as unchecked, whatever the summary says.
- Spot-check two yourself. Pick the error case and the permission rule, because a passing main path says nothing about either.
Criteria sit inside a larger spec that also carries who the feature is for, why it matters and what’s out of scope. The Build Pack format guide shows a full spec with criteria in place, and the document templates include story and PRD layouts you can copy.
See criteria in a spec an agent builds from. A sample Build Pack for a made-up product, with every section’s source.
See a sample Build PackDoing this in MindBacklog
Mind drafts the stories with acceptance criteria and the Build Pack, each naming its source. The Build Pack carries why it matters, who it’s for, the metric, and the must-haves with acceptance criteria, and it says what’s already broken and what already exists. Copy it for Claude Code or Cursor, or let your agent pull it over MCP on the Growth and Scale plans. How the Build Pack works.
Questions
What are good acceptance criteria examples?
Good ones name a starting state, an action and a visible result. “Given a booking tomorrow at 10:00, when the client picks Thursday 14:00, the booking shows Thursday 14:00” is a good example. “Rescheduling is easy” is not, because nobody can mark it pass or fail.
What’s the difference between acceptance criteria and a definition of done?
Acceptance criteria belong to one story and describe what that story must do. A definition of done applies to every piece of work, such as tests passing and nothing existing breaking. A story is finished when it meets both.
How many acceptance criteria should a story have?
Enough to cover the main path, the empty state, the main error and any permission rule. That’s usually three to six. If a story needs more than eight, it’s often two stories.
