Performance budget prompts turn a general request to make a website faster into a measurable agreement. The team needs to know which routes matter, what conditions represent its users and what happens when a limit is exceeded. A single score can obscure a large script increase or a slow interaction. These prompts help choose practical budgets, explain growth and design a check that remains useful after the next release.
Establish a baseline before choosing limits
Provide representative routes, current measurements, device and network conditions, and the task people perform on each page. Include whether the data comes from a controlled lab run or actual visits. Describe the product constraint, such as keeping a key interaction responsive. Avoid asking the agent to invent a universal size or time threshold without that context.
web.dev's performance budget introduction explains budgets as limits on measures that affect performance. Use it to understand the concept. The prompts below propose an original workflow for selecting limits from your evidence; they do not present sample numbers as guaranteed ranking or conversion improvements.
Choose limits tied to page responsibilities
Group routes by what they do. An article and an authenticated editor may need different checks, but both should have an understandable reason for their limits. Ask the agent to identify a small set of useful measurements instead of tracking every available score. Include ownership and a process for reviewing an intentional exception.
Propose performance budgets for {{route_groups}}.
Use {{baseline_measurements}}, {{user_conditions}} and {{product_goals}}.
Choose a small set of metrics with collection method, proposed limit,
rationale and owner. Separate lab regression checks from field monitoring.
Mark unsupported thresholds as hypotheses. Include an exception process
that records the tradeoff rather than silently raising the budget.
Explain a bundle increase before optimizing it
A larger bundle may come from a new dependency, an import boundary or a changed build configuration. Ask for an attributable comparison between the same routes and build mode. Tree-shaking advice without a concrete imported path is often too vague to act on. Preserve functionality and accessibility while considering a smaller dependency or deferred loading.
Investigate bundle growth between {{base_build}} and {{current_build}}.
Focus on {{routes}} and use {{bundle_evidence}}.
Attribute the increase to modules or assets where possible. Compare
removal, smaller imports and deferred loading with their user impact.
Propose one focused change and repeat the same build measurement.
Report transferred size assumptions and remaining uncertainty.
Design a regression check people can trust
A noisy check soon becomes an ignored check. State how many runs you collect, which conditions are controlled and how variance is handled. Ask what happens when the test environment is unhealthy. Distinguish a hard resource-size regression from a timing signal that needs interpretation. The output should help reviewers reproduce a failure.
Design a CI performance check for {{budget_definition}}.
Environment: {{runner_and_test_conditions}}. Existing tools: {{tools}}.
Specify route setup, measurement, variance handling and failure output.
Show how a reviewer reproduces a failure and records an approved exception.
Separate infrastructure failures from budget breaches. Verify using one
controlled regression before enabling the check for all changes.
Revisit limits using comparable evidence
After shipping an optimization, compare the same route and conditions with the baseline. For real-user outcomes, allow enough relevant observations and state the window being compared. Do not attribute a change in business results to one optimization without evidence that supports that conclusion. Keep the budget a decision tool rather than a decorative dashboard.
Save the approved budget prompt in Kipt with variables for route group, baseline and measurement method. For diagnosing a specific slow operation, continue with the performance optimization prompts.