feat(cli): shared page session

[nucliweb]

Problem

When the decision tree fires a follow-up snippet (e.g. LCP-Subparts when LCP > 2.5s), the runner closes and relaunches the browser, re-navigating the page. This wastes time and risks different results across measurements.

Current flow

launch browser → navigate → evaluate [LCP, CLS] → close browser
  → launch browser → navigate → evaluate [LCP-Subparts] → close browser

Proposed flow

launch browser
  → navigate + waitMs
  → evaluate initial steps [LCP, CLS]
  → apply nextSteps(results) → follow-up items
  → evaluate follow-ups [LCP-Subparts] in same page (no re-navigation)
  → close browser
  → return { url, navMs, results: [...initial, ...followUps], pageErrors }

Implementation

runner.js current signature:

export async function runSnippets({ url, items, waitMs, headless, viewport })

New signature:

export async function runMeasurement({ url, workflow, rules, waitMs, headless, viewport })

Where rules is the array from decision-tree.js. The runner applies nextSteps() internally after the first round, within the same browser session.

Files

• cli/src/runner.js — main refactor
• cli/src/bin.js — update call site

Acceptance criteria

• LCP + LCP-Subparts complete with a single navigation
• Follow-up snippets run against the same page state as initial snippets
• Existing tests pass