[Tom-R-Main]

An agent can now read JavaScript-populated page text without a desktop session. footwork browser lightpanda read URL launches a pinned external prebuilt engine, checks its private Unix peer against the actual child, and captures bounded DOM text over one connection before cleanup.

The result names its engine, withholds form/editable values and unqualified link bases, exposes no action refs, and separates capture, cleanup and completion. No click, typing, raw-script or raw-CDP route is available. Website loading can have network effects; layout, visibility, atomic capture and business completion remain unqualified. Existing Chromium guard and durable authority stay unchanged. The light dependency metadata split remains a separate package.

Tools, doctor, schema, guide and artifact expose the route and its limits. The README was revised using the user's selected local Colophon profile, with all 149 original protected spans and the two intended additions preserved. Private guidance and profile sources are excluded.

Validation on runtime 2b5325e / wheel 373a43fcccb905ae70fef7f63cb53a478ee40537134384fcc4d534787f43edbb:

• 2,786 source regression tests passed; 95 skipped, 66 browser/live checks deselected. Lint and notebook checks passed.
• 308 exact installed checks passed on macOS, hosted Linux and the authorized VM chat, without CUA or browser-use. Documented HTTPS reads passed on both platforms.
• The VM measured 4 CPUs / 16 GiB. The parent matched all 15 test hashes; the VM byte-verified 211 installed product files. Its clean wrapper used no extra signals and ended with no owned descendants after 204 test-only zombie reaps. These are not product/global cleanup guarantees or a 2-CPU/8-GB measurement.
• The separate inherited-model review found no actionable defects in the initial source. Automated review then found two projection gaps; red controls reproduce them and the corrections pass. The actual engine reports no child metadata for the srcdoc/data fixture, so that metadata branch remains modeled while real DOM omission passes.
• Earlier path, frame-count and diagnostic-monitor failures remain recorded. The monitor failure's lost wait/cleanup custody is not erased by the fresh-consumer repeat. Predictions precede the substantive tests in separate commits.

Report, requirement ledger, reference-source hashes and compact VM receipts. The full VM records and history remain linked at immutable evidence commit 2438c65. Final-head CI and merge are tracked on this PR; the evidence-only head retains the same wheel bytes.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-31T20:35:18.998397Z	d9ca03e	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d9ca03ee60

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".