[nucliweb]

Summary

Phase 1 of the budgets: a budget for each metric the CLI measures, and no budget that passes without being checked. The budgets in the reports and the JSON come in the next PR.

• --budget-fcp, --budget-ttfb and --budget-inp, next to --budget-lcp and --budget-cls.
• The interaction workflow measures INP, in a step INP that needs a click or a key press, like interactions and input-latency. No workflow ran the INP snippet before, so an INP budget only worked with --snippet INP.
• A budget never passes unchecked:
  • A budget of a metric that the run does not measure, such as --budget-lcp with --workflow audit, exits with 2 before the page loads, naming the workflow or snippet that measures it.
  • A budget whose metric was not measured, such as INP without interactions or an LCP that failed, exits with 1 and says why.
  • A value that is not a number of 0 or more exits with 2. --budget-lcp abc was NaN, and passed.
• stderr names the failed budgets also with --json, whose report goes to stdout.
• cli/src/budgets.js holds the logic: which snippets measure each metric, the parsing, and the verdict of each budget against the value, or the median with --runs. bin.js uses it for a page and for each combination of a matrix.
• The GitHub Action takes budget-fcp, budget-ttfb and budget-inp. Its budget-lcp and budget-cls default to auto: 2500 and 0.1 with core-web-vitals, and no budget with the other workflows, which do not measure LCP or CLS. An action with workflow: audit and the defaults keeps passing instead of exiting with 2; an empty value still disables a budget.

cli/README.md, content/CLI.mdx (options, the inputs of the Action, the steps of the interaction workflow and a paragraph on what each budget needs) and the CLI changelog describe it.

Test plan

• cli/tests/unit/budgets.test.js (local): the five options, values that are not numbers, the snippets that measure each metric checked against their sources, the interaction workflow with INP, the metrics a run does not measure, equal to the budget passes, not measured with the reason of its snippet, the step before a follow-up, the median of the runs, the words of stderr.
• cli/tests/e2e/budgets.test.js (local, new): INP over budget with a slow click in the interaction workflow exits with 1; without a click, not measured and 1; --budget-lcp with --workflow audit and --budget-fcp fast exit with 2 without a request to the page; FCP and TTFB pass with a loose budget and fail with an impossible one, named on stderr with --json. They failed before the change.
• github-action.test.js (local): the new inputs reach the CLI, and auto gives 2500 and 0.1 with core-web-vitals and nothing with audit.
• interaction-workflow.test.js and mcp-workflow.test.js (local): INP among the steps that need input, skipped without a click and measured with one.
• Lint, check:emoji, CLI unit 339 of 339, CLI e2e 582 of 582, test:site 44 of 44, test:ci 22 of 22 (local).
• By hand with the CLI (local, a page with a button whose click blocks for 300 ms):
  • --workflow interaction --interact-script click.json --budget-inp 200: exit 1, Budget violation: INP 304ms exceeds budget 200ms.
  • The same without the script: exit 1, INP was not measured (budget 200ms): The interactions have no click or type step; add one with --interact-script.
  • --workflow audit --budget-lcp 2500: exit 2, --budget-lcp needs a run that measures LCP, and the audit workflow does not. Use --workflow core-web-vitals or --snippet LCP.
  • --runs 3 --budget-lcp 2500 --budget-cls 0.1 on the default workflow: exit 0.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
webperf-snippets	Ready	Preview	Oct 25, 2026 11:41pm UTC