Guide · Planning
Feature planning checklist: what to ask before the sprint.
Twelve areas and the questions that most often turn up mid-build when nobody asked them in planning. Run it against your plan, turn each answer into a requirement or an open question with an owner.
In short. A feature planning checklist is a list of questions you answer before a feature enters a sprint. It covers the areas a plan usually skips: permissions, data migrations, error states, email, billing, security, accessibility and rollout. Each answer becomes a requirement, a “Not now” item, or an open question with an owner.
How to use this checklist
Go through the areas that apply to your feature and skip the rest. A feature that sends no email can skip that section; one that touches data, permissions and money cannot skip much.
For each question, write one of three things in the plan. A requirement, if the answer changes what gets built. A “Not now” item, with a reason, if you are choosing to leave it out. An open question, with an owner and a date to decide by, if you don’t know yet.
What you must not do is leave a question unanswered and unwritten. That is how a four-day task becomes a nine-day one: the question comes up in week two, and by then the design already assumes the wrong answer.
The questions below come from the check library inside SprintProof, which has a set of them for each of the twelve areas. This page shows the ones that matter most for most features. Write the feature down first (the PRD template works, and the PRD generator fills it in as you type), then walk the list against it.
Auth and permissions
Permission mistakes are the most expensive to undo, because users have already relied on the wrong behaviour.
- Does every read and write check that the caller may touch this specific record, not only that they are logged in?
- Is there a written table of which role may do what, and does the UI hide and the API refuse the same things?
- Can someone grant a role higher than their own, and what stops the last owner from leaving or being removed?
- Can a user see and end their other sessions, and does removal from a workspace invalidate existing tokens?
APIs
Anything another program calls needs rules that a screen can get away without.
- Is every field validated for type, length, range and allowed values on the server, with errors the client can show next to the field?
- What happens if the same create request arrives twice?
- Is the list paginated with a stable sort, and what is the maximum page size?
- What is the rate limit per user and per IP, and what does the client see when it is hit?
Data and migrations
Schema changes outlive the sprint. A migration that works on a test database can still lock a production table.
- Will this migration lock the table in production, and how long will it take on the real row count?
- How do existing rows get their new values, and what do old clients write while the backfill runs?
- Is delete reversible, what happens to the things that point at the record, and when is it really gone?
- Does every query filter by tenant, and is there a safety net if a developer forgets?
Frontend states
A screen is more than its happy path. Each of these is a requirement someone has to build.
- What does each list show when it is empty, loading or failed?
- What does the button do while the request is in flight, and what if the user clicks it twice?
- Where do validation errors appear, is typed work kept on error, and what happens when the user navigates away?
- What does this screen look like at 375 px wide, and what happens to the table?
- What happens with a 120-character name, a single unbroken word or an empty value?
- If the user refreshes, presses back or copies the URL, do they land in the same view?
Security
Ask these even when the feature “is only an internal screen”. Internal screens get exposed.
- Is user content escaped on output, and if HTML or markdown is allowed, is it sanitized with an allow-list?
- Is every dynamic value a bound parameter, and are sort columns and file names checked against an allow-list?
- Can a user make the server request internal addresses, and are redirects and DNS rebinding handled?
- Which secrets are stored, how are they encrypted, and who can read them in the database and logs?
- How many wrong guesses are allowed, and does the error message reveal whether an account exists?
Email and notifications
Anything that emails an address a user typed can be used to send spam from your domain.
For a worked example with an estimate, see the email invitations checklist.
- Is sending rate-limited?
- Does the link expire, work once, and only for the intended person?
- How does the sender find out an email bounced?
- Is the email sent from a queue with retries, so a provider outage does not fail the user’s request or lose the message?
- Can people turn this email off, and should they be able to?
- Should this also show up in the app, in case the email is missed?
Billing and plans
If the feature is limited by plan, seats or credits, the limit has to be enforced where it can’t be skipped.
- Where is the plan rule checked, and does the backend enforce it, not just the UI?
- What happens when an action goes over the plan’s seat limit: block, warn the admin or add a paid seat?
- Is your database updated only from verified provider webhooks, and is it safe when events arrive late, twice or out of order?
- What happens on the first failed payment: grace period, retries, emails and what access remains?
- Can two simultaneous actions both spend the last credit, and does a failed action refund what it took?
Observability
You will not read the code when it breaks at night. Decide now what you will see instead.
- When this fails at 2 a.m., who is told, and what do they see?
- How would you know the job stopped running, is stuck or is falling behind?
- Can you follow one user request through every log line and job it caused?
- Which event or number will tell you in two weeks whether this feature worked?
- Could any log line contain a password, token, email body or personal data?
Accessibility
Cheaper to plan than to retrofit. Most of these are decisions about markup and behaviour, not extra features.
- Can every action be reached and used with the keyboard alone, in a sensible order?
- Where does focus go when a dialog opens, can the user leave with Escape, and where does focus return?
- Does every field have a visible label, a programmatic error message and correct autocomplete?
- Does text meet 4.5:1 contrast, and is any meaning carried by color alone?
- Does the motion respect the reduced-motion setting?
Localization
Skip this section if the product is English-only and will stay so. Otherwise these are far cheaper before the strings are written.
- Is every string in a message catalog, or are sentences glued together in code?
- Are plurals handled with ICU rules rather than an “s” appended in code?
- Are dates, numbers and currency formatted for the viewer’s locale?
- Which time zone applies to each displayed date, and is it shown when it could be ambiguous?
- Do the fields assume a first and last name, a five-digit postcode or a US phone number?
Performance
Test at the size of your largest customer, not at the size of your seed data.
- How many queries does the list endpoint run for 50 items and for 500?
- Which index serves this query, and has it been run against a table the size of your largest customer?
- What is the largest input you have tried, and does the process stream it or hold it all in memory?
Rollout and flags
Shipping is part of the feature. A plan with no rollback is a plan that hopes.
- Can you turn this off in production in under a minute without a deploy?
- Do old and new code both work against the schema during the deploy, and what does a rollback do?
- What do current customers experience on day one, and did anyone tell them?
- Who gets it first, what do you watch, and what makes you pause?
- Who removes the flag, and by when?
- For the process every release follows, use the release management template.
After the questions: estimate and sign off
Once the answers are in the plan, the work has changed shape: some questions added parts, some moved things to “Not now”. Estimate after the checklist, not before, and break the answers down into parts as described in how to break down a feature. Give each part a best, likely and worst case from the people who will build it, then read the dates at 50, 80 and 90% confidence rather than one number. The free estimation tool does the arithmetic from your three numbers per part, and the software estimation methods guide covers how. Once the plan is settled, the sprint planning template turns it into the sprint.
Then get the owners of the areas the feature touches to read it. Billing, security and support rarely say no, but they often say “not like that”, and you want to hear that before the build.
Doing this in SprintProof
SprintProof runs this checklist for you. After the AI interview and the first breakdown, its “Did you think of…?” gap review matches your plan against the check library and asks only the questions that apply, each with why it matters and a suggested part to add. Nothing is added until you click.
You can add your own checks too, so the questions your team keeps learning the hard way are asked on the next plan. See pricing for the Team plan.
Plan the feature, not just the document.
Describe the feature and SprintProof interviews you, drafts the breakdown, asks the questions reviewers ask, and gives an estimate as a range with 50, 80 and 90% dates.