[EVMCOD]

What and why

Screenshots for the website and Product Hunt were taken from a fictional demo vault (a small design studio: projects, meetings, daily notes, reading notes) on a fresh install with no key. Four things a first visitor would see were wrong:

Seen	Fix
The ◐ toggle did nothing on the dashboard: body.dark was set, the page stayed white. app.css declares its aliases (--bg: var(--cl-bg)…) on :root, where they resolve with the light palette, and body.dark only changes --cl-* on body. Checked on main: body background stays rgb(255,255,255) after the toggle.	The aliases are declared on :root, body. After the toggle: rgb(15,17,21).
Without a key, "What did we decide about pricing?" quoted "Slow Sunday… pricing" and "Decided in [[…]]", not the decision in Pricing experiment (under ## Decision, the line says "Keep both plans and make yearly the default"). "decide" did not match "Decision", and headings were ignored.	Words of one family match (a rough stem cut to four letters: decide/decided/decision → deci, pricing/price → pric). A line inherits the question words of its heading (full for the first line under it, half after) and is shown as "Decision (2026-10-21) — Keep both plans…". The note's title adds a quarter point per word but never makes a line a candidate alone. Notes are ordered by their best line.
The note view's Key points / Summarize / Action items / Expand were hard-coded data-lang="es": every user got Spanish.	They follow the browser language (es-* → Spanish, otherwise English), the pill says which; the API route and note_ai.DEFAULT_LANG default to English. Translate buttons unchanged.
Search results on the notes home showed a few words ("pricing.").	They use content_snip, the words around the match.

How it was tested

• The four questions run through brain.ask with the mock provider on the demo vault: pricing → the decision line first; runway → "Cash: 48,000 €… about 11 months of runway"; Sam → the sync-bug lines from three notes.
• Screenshots retaken (light and dark) from a server on this branch; the dark toggle measured before/after on this branch and on main.
• New tests: word families, heading context and its display, best note first (test_brain_no_key.py); note actions follow the browser and search uses content_snip (test_notes_home_static.py). test_csp_static unchanged.
• pytest -m "not live" (offline, headless Chrome): 2126 passed, 22 skipped, 1 failed — the fastembed test that needs the model on disk and fails the same way on main.

Checklist

• Verified locally.
• Tests added or updated.
• No migrations.
• CHANGELOG.md entry under Unreleased.
• No personal data and no secrets.

🤖 Generated with Claude Code

[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 2026-10-30T10:56:11.869404Z	cb39f82	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.