Feat: UXCore dark theme

.claude/commands/qa-canonical.md

@@ -0,0 +1,42 @@
+You are running a QA pass on keepsimple.io.
+
+Profile: **canonical** — the desktop half of a full pass. Source of truth for desktop findings.
+
+Scope: all sections × all locales × desktop.
+Time budget: 90-120 min, 30-min hard cap per section.
+Output: reports/YYYY-MM-DD.md (and rendered .html)
+Diff target: previous canonical run.
+
+Note: canonical alone is incomplete. Pair with mobile-followup for a full-matrix pass.
+
+SETUP
+
+1. Read .claude/keepsimple-qa/SKILL.md in full. This is your playbook for method.
+2. Read qa-config.yml for routes, locales, viewports, and any per-section `auth_required:` lists.
+3. Read known-issues.md — do not refile anything listed there.
+4. Read .claude/keepsimple-qa/PROFILES.md and confirm the canonical profile spec matches what's above. Treat the profile as authoritative for scope; treat the skill as authoritative for method.
+5. Confirm Playwright MCP is connected before proceeding. If not, stop and report what's missing.
+
+EXECUTION
+
+Run the pass per the skill. Honor the profile's scope exactly. Routes listed under `auth_required:` in their section are recorded as `skipped — auth required (configured)` and never visited. If you must cut scope mid-run, follow the skill's "Run is incomplete" rule — explicit, loud, in the report header.
+
+OUTPUT
+
+Produce the report at reports/YYYY-MM-DD.md following the skill's report structure:
+
+- Metadata header
+- Coverage report table
+- Findings summary
+- Diff vs prior canonical
+- Routes covered
+- Routes skipped with reason
+- Fixed-pass checklist
+- JSON finding blocks
+- Out-of-scope observations
+
+Then render HTML: `node .claude/keepsimple-qa/render-report.js reports/YYYY-MM-DD.md`.
+
+Do NOT auto-update qa-runs/findings-register.md or any other tracking file. The user updates those manually after reviewing the report.
+
+Return a 5-8 line summary in chat when done: profile, route coverage %, finding counts by severity, persistent vs new breakdown, run status.

.claude/commands/qa-deploy-check.md

@@ -0,0 +1,53 @@
+You are running a **post-deploy regression check** on keepsimple.io. This is the most token-efficient profile — it only audits routes that could have broken (changed since last run) or were already broken (open findings from the prior run).
+
+Profile: **deploy-check** — surgical regression pass.
+
+Scope:
+
+- **Always include:** every route with an open finding in the most recent canonical / smoke / mobile-followup / full-matrix report.
+- **Always include:** every route whose fingerprint differs from the prior batch (i.e. `changed` or `new` per `batch-fingerprint`).
+- **Skip:** routes that are both `unchanged` AND have zero open prior findings.
+
+Time budget: 5 min per route, 30 min hard cap total. Designed to run after every deploy.
+Output: reports/YYYY-MM-DD-deploy-check.md (and rendered .html)
+Diff target: most recent canonical (or full-matrix) run.
+
+SETUP
+
+1. Read .claude/keepsimple-qa/SKILL.md in full.
+2. Read qa-config.yml for routes, locales, viewports, and any per-section `auth_required:` lists.
+3. Read known-issues.md — do not refile anything listed there.
+4. Read .claude/keepsimple-qa/PROFILES.md and confirm the deploy-check profile spec matches what's above.
+5. Confirm Playwright MCP is connected before proceeding. If not, stop and report what's missing.
+
+DECIDE THE SCOPE
+
+1. Run `node .claude/keepsimple-qa/helper.mjs batch-fingerprint keepsimple --save`. This reads every route from qa-config.yml, fetches each, computes a fingerprint, and compares against the prior batch stored in `qa-runs/state/fingerprints.json`. Output is a JSON array with per-route verdicts: `unchanged | changed | new | errored`.
+2. Read the most recent prior report under `reports/`. Extract every finding whose status is open (`persistent`, `persistent (unchanged route)`, `new`, `confirmed`, `not seen this run`). Note their routes.
+3. Build the **deploy-check route set** = (routes with `verdict: changed | new` from step 1) ∪ (routes with open prior findings from step 2).
+4. Subtract any routes listed under `auth_required:` in their section — those are configured-skip.
+5. If the resulting set is empty, write a 5-line "All clear — nothing changed and no open findings" report and exit. This is the happy path post-deploy.
+
+EXECUTION
+
+For every route in the deploy-check route set, run the per-route checklist from the skill (snapshot, console+network, click primary elements, fill any forms without submit, scroll, resize) plus the must-work fixed-pass items for any section whose hub route is in the set. Cap at 5 min per route; if you hit the cap mid-route, mark it `partial` and move on.
+
+For routes NOT in the deploy-check set (unchanged + no open findings), include them in the coverage table with status `unchanged-skip` and `Source: <prior report date>`.
+
+OUTPUT
+
+Produce the report at reports/YYYY-MM-DD-deploy-check.md following the skill's report structure:
+
+- Metadata header (build IDs, change summary from batch-fingerprint)
+- **Deploy verdict** at the top: `pass | regressions found | inconclusive` based on whether any new high/critical findings appeared or any prior open findings persisted on changed routes
+- Coverage table (every route from qa-config.yml, with per-route status: deep-audited / unchanged-skip / configured-skip)
+- Findings summary
+- Diff vs prior canonical (status of each prior finding: refuted / persistent / changed)
+- JSON finding blocks
+- Out-of-scope observations
+
+Then render HTML: `node .claude/keepsimple-qa/render-report.js reports/YYYY-MM-DD-deploy-check.md`.
+
+Do NOT auto-update qa-runs/findings-register.md or any other tracking file. The user updates those manually after reviewing the report.
+
+Return a 5-7 line summary in chat: deploy verdict, route counts (re-audited / skipped-unchanged / skipped-auth-required), prior-findings status (X resolved / Y persistent / Z new), run status.

.claude/commands/qa-full-matrix.md

@@ -0,0 +1,40 @@
+You are running a QA pass on keepsimple.io.
+
+Profile: **full-matrix** — single pass covering everything: every section, every locale, every viewport.
+
+Scope: all sections × all locales × desktop+mobile.
+Time budget: 3-4 hours. May need to be split across multiple Claude Code sessions; if so, sequence them and produce sub-reports per session, then a consolidated final report.
+Output: reports/YYYY-MM-DD-full-matrix.md (or reports/YYYY-MM-DD-full-matrix-part-N.md for split runs) (and rendered .html)
+Diff target: previous full-matrix run, or most recent canonical+mobile-followup pair.
+
+SETUP
+
+1. Read .claude/keepsimple-qa/SKILL.md in full. This is your playbook for method.
+2. Read qa-config.yml for routes, locales, viewports, and any per-section `auth_required:` lists.
+3. Read known-issues.md — do not refile anything listed there.
+4. Read .claude/keepsimple-qa/PROFILES.md and confirm the full-matrix profile spec matches what's above. Treat the profile as authoritative for scope; treat the skill as authoritative for method.
+5. Confirm Playwright MCP is connected before proceeding. If not, stop and report what's missing.
+
+EXECUTION
+
+Run the pass per the skill. Honor the profile's scope exactly. Routes listed under `auth_required:` in their section are recorded as `skipped — auth required (configured)` and never visited. If the run needs to be split across sessions due to time/context limits, produce part-N reports and call out clearly that a final consolidated report is needed. If you must cut scope mid-run, follow the skill's "Run is incomplete" rule — explicit, loud, in the report header.
+
+OUTPUT
+
+Produce the report at reports/YYYY-MM-DD-full-matrix.md (or part-N variant) following the skill's report structure:
+
+- Metadata header
+- Coverage report table
+- Findings summary
+- Diff vs prior full-matrix or canonical+mobile-followup pair
+- Routes covered
+- Routes skipped with reason
+- Fixed-pass checklist
+- JSON finding blocks
+- Out-of-scope observations
+
+Then render HTML: `node .claude/keepsimple-qa/render-report.js reports/YYYY-MM-DD-full-matrix.md` (or the part-N variant).
+
+Do NOT auto-update qa-runs/findings-register.md or any other tracking file. The user updates those manually after reviewing the report.
+
+Return a 5-8 line summary in chat when done: profile, route coverage %, finding counts by severity, persistent vs new breakdown, run status.

.claude/commands/qa-init.md

@@ -0,0 +1,136 @@
+---
+description: Configure (or re-configure) the QA agent for keepsimple.io. Walks through site URL, locales, routes, viewports, and writes qa-config.yml.
+---
+
+You are the QA agent's setup wizard. Your job is to interactively configure the QA pass by writing or updating `qa-config.yml` at the repo root.
+
+Do NOT start any QA pass during this wizard. Your only output is the config file (and a friendly recap at the end).
+
+## Pre-check
+
+If `qa-config.yml` already exists, ask: "qa-config.yml already exists. Back it up to `qa-config.yml.bak` and rewrite, edit in place, or cancel?" Do NOT silently overwrite.
+
+If the user picks "edit in place," surface the existing values as defaults at each step so they can keep what's there with a one-word reply.
+
+## Steps
+
+Walk the user through these questions a few at a time. Don't dump the whole list. Ask, wait for the answer, ask the next.
+
+### 1. Site URL
+
+Ask: "Production site URL? (default: https://keepsimple.io)"
+
+Then ask: "Staging URL too? (or 'skip')"
+
+### 2. Locales
+
+Ask: "Locales to test? (default: en, ru, hy)"
+
+If non-default, ask: "How are locales URL-prefixed?"
+
+- **A) default-unprefixed** — en routes are `/uxcore`, ru routes are `/ru/uxcore` _(current default for keepsimple)_
+- **B) all-prefixed** — every route has a locale segment
+- **C) no-prefix** — single-locale, no segments
+
+### 3. Route discovery
+
+Ask: "How should we discover routes?"
+
+- **A) Pull from `/sitemap.xml`**
+- **B) Crawl from the homepage** (one click deep from `/`)
+- **C) Keep the routes in the existing qa-config.yml** _(default if you're re-running)_
+- **D) I'll paste the routes manually**
+
+If A/B: fetch via Playwright MCP, group discovered URLs by inferred section (path's first segment), present them back, ask: "Confirm these sections + routes? (yes / I want to edit)"
+
+If D: prompt for routes one section at a time:
+
+- "Section name (e.g. 'uxcore', 'site'):"
+- "Routes for that section, one per line, leading slash, blank line to finish:"
+- "Add another section? (yes/no)"
+
+### 4. Per-section options
+
+For each section, ask:
+
+- "Mark this section as `default: true` for smoke runs? (only one section gets this)"
+- "Any `primary_interaction:` worth declaring? (e.g. 'open the bias modal on a card', or 'skip')"
+- "Any routes in this section that require auth? (comma-separated paths, or 'none')"
+
+### 5. Viewports
+
+Ask: "Viewports? (default: desktop 1920×1080, mobile 390×844)"
+
+### 6. Timezone
+
+Ask: "Timezone for report timestamps? (default: Asia/Yerevan)"
+
+### 7. Design system reference
+
+Ask: "Path to design-system reference doc? (default: uxcore-visual-design-system.md, or 'skip')"
+
+### 8. Confirm and write
+
+Show the user the proposed `qa-config.yml` content as a code block. Ask: "Write this to qa-config.yml? (yes/edit)"
+
+Use this template (fill in the captured values):
+
+```yaml
+# QA agent configuration — read by .claude/skills/keepsimple-qa/SKILL.md
+# Generated by /qa-init on YYYY-MM-DD.
+
+environment:
+  production: <site_url>
+  staging: <staging_url or empty>
+
+locales:
+  - <locale_1>
+  # ...
+
+locale_url_strategy: <strategy>
+
+viewports:
+  desktop: { width: 1920, height: 1080 }
+  mobile: { width: 390, height: 844 }
+
+report_timezone: <tz>
+
+design_system: <path or omit>
+
+sections:
+  <section_name>:
+    # default: true     # uncomment on exactly one section
+    routes:
+      - /<route_a>
+      - /<route_b>
+    # auth_required:
+    #   - /<route_b>
+    # primary_interaction: "..."
+  # ... repeat for each section
+```
+
+### 9. Recap
+
+Print:
+
+```
+✓ qa-config.yml written ({N} sections, {M} locales, {V} viewports)
+✓ Default smoke section: <section>
+
+Next steps:
+  /qa-smoke              — quick render check
+  /qa-locale-smoke       — locale routing only
+  /qa-canonical          — desktop pass across all locales
+  /qa-mobile-followup    — mobile pass to pair with canonical
+  /qa-full-matrix        — everything in one report
+  /qa-deploy-check       — surgical post-deploy regression
+  /qa-retest F#,F#       — verify specific finding IDs after a fix
+```
+
+## Constraints
+
+- Do NOT begin a QA pass during this wizard. The user must invoke a profile separately.
+- Do NOT auto-discover routes by visiting EVERY page on the site — keep discovery to sitemap or a single homepage crawl.
+- Do NOT silently overwrite `qa-config.yml`. If one exists, ask first.
+- Do NOT install Playwright MCP for the user. If it's not connected, just tell them how to install it (`claude mcp add playwright npx -- @playwright/mcp@latest`) and stop.
+- Keep the conversation tight. Two or three questions per turn maximum.

.claude/commands/qa-locale-smoke.md

@@ -0,0 +1,40 @@
+You are running a QA pass on keepsimple.io.
+
+Profile: **locale-smoke** — confirm locale-prefix routing works across all configured locales.
+
+Scope: one section × all locales × desktop. Default section is the one marked `default: true` in `qa-config.yml` (or the first section if none is marked).
+Time budget: 25 min hard cap.
+Output: reports/YYYY-MM-DD-locale-smoke.md (and rendered .html)
+Diff target: most recent canonical run.
+
+SETUP
+
+1. Read .claude/keepsimple-qa/SKILL.md in full. This is your playbook for method.
+2. Read qa-config.yml for routes, locales, viewports, `locale_url_strategy:`, and any per-section `auth_required:` lists.
+3. Read known-issues.md — do not refile anything listed there.
+4. Read .claude/keepsimple-qa/PROFILES.md and confirm the locale-smoke profile spec matches what's above. Treat the profile as authoritative for scope; treat the skill as authoritative for method.
+5. Confirm Playwright MCP is connected before proceeding. If not, stop and report what's missing.
+
+EXECUTION
+
+Run the pass per the skill. Honor the profile's scope exactly. Routes listed under `auth_required:` in their section are recorded as `skipped — auth required (configured)` and never visited. If you must cut scope mid-run, follow the skill's "Run is incomplete" rule — explicit, loud, in the report header.
+
+OUTPUT
+
+Produce the report at reports/YYYY-MM-DD-locale-smoke.md following the skill's report structure:
+
+- Metadata header
+- Coverage report table
+- Findings summary
+- Diff vs prior canonical
+- Routes covered
+- Routes skipped with reason
+- Fixed-pass checklist
+- JSON finding blocks
+- Out-of-scope observations
+
+Then render HTML: `node .claude/keepsimple-qa/render-report.js reports/YYYY-MM-DD-locale-smoke.md`.
+
+Do NOT auto-update qa-runs/findings-register.md or any other tracking file. The user updates those manually after reviewing the report.
+
+Return a 5-8 line summary in chat when done: profile, route coverage %, finding counts by severity, persistent vs new breakdown, run status.

.claude/commands/qa-mobile-followup.md

@@ -0,0 +1,42 @@
+You are running a QA pass on keepsimple.io.
+
+Profile: **mobile-followup** — the mobile half of a full pass. Pairs with canonical to close the full matrix.
+
+Scope: all sections × all locales × mobile.
+Time budget: 90 min, 30-min hard cap per section.
+Output: reports/YYYY-MM-DD-mobile.md (and rendered .html)
+Diff target: paired canonical run from the same day or build.
+
+Note: mobile-followup alone is incomplete. Pair with canonical for a full-matrix pass.
+
+SETUP
+
+1. Read .claude/keepsimple-qa/SKILL.md in full. This is your playbook for method.
+2. Read qa-config.yml for routes, locales, viewports, and any per-section `auth_required:` lists.
+3. Read known-issues.md — do not refile anything listed there.
+4. Read .claude/keepsimple-qa/PROFILES.md and confirm the mobile-followup profile spec matches what's above. Treat the profile as authoritative for scope; treat the skill as authoritative for method.
+5. Confirm Playwright MCP is connected before proceeding. If not, stop and report what's missing.
+
+EXECUTION
+
+Run the pass per the skill. Honor the profile's scope exactly. Routes listed under `auth_required:` in their section are recorded as `skipped — auth required (configured)` and never visited. If you must cut scope mid-run, follow the skill's "Run is incomplete" rule — explicit, loud, in the report header.
+
+OUTPUT
+
+Produce the report at reports/YYYY-MM-DD-mobile.md following the skill's report structure:
+
+- Metadata header
+- Coverage report table
+- Findings summary
+- Diff vs paired canonical run
+- Routes covered
+- Routes skipped with reason
+- Fixed-pass checklist
+- JSON finding blocks
+- Out-of-scope observations
+
+Then render HTML: `node .claude/keepsimple-qa/render-report.js reports/YYYY-MM-DD-mobile.md`.
+
+Do NOT auto-update qa-runs/findings-register.md or any other tracking file. The user updates those manually after reviewing the report.
+
+Return a 5-8 line summary in chat when done: profile, route coverage %, finding counts by severity, persistent vs new breakdown, run status.