Web UI QA: manual testing with screenshots/video of browsing flow

[timon0305]

Problem

The web UI is a zero-build vanilla JS / Jinja2 frontend (static/js/app.js, static/js/download.js, templates/) that renders AI conversation history with markdown, code blocks, tool call metadata, and thinking blocks. While XSS sanitization is verified by static source tests (test_xss_sanitization.py), the actual rendering behavior — correct markdown display, proper conversation ordering, functional search, workspace navigation, export download buttons — has no automated or documented manual verification. The project claims five-platform support (Windows native, WSL2, macOS, Linux, SSH remote) but CI runs only on ubuntu-latest. Visual regressions and platform-specific rendering issues are invisible.

Severity

Medium · 5 points

Acceptance criteria

• Documented manual test pass with screenshots or screen recording for each major view: home (workspace list), workspace detail (conversation list), conversation view (bubble rendering), search results, config page
• Markdown rendering verified: code blocks with syntax highlighting, inline code, headers, lists, links
• Tool call display verified: tool name, input, output, status all visible in conversation bubbles
• Thinking blocks verified: collapsible sections render with duration
• Export functionality verified: Markdown, JSON, CSV, PDF download buttons each produce a valid file
• Search verified: full-text search returns results, highlights matching text, navigates to correct conversation
• Tested in Chrome and Firefox at minimum
• Visual bugs filed as follow-up issues
• Evidence (screenshots or video link) attached to PR or linked issue
• PR approved by at least 1 reviewer

[timon0305]

Heads-up for whoever reviews PR #34: two visual overflow bugs were found and fixed in the same PR alongside the QA pass, both narrow CSS:

• .main-content { min-width: 0 } — fixes the conversation column overflowing the viewport when a bubble has unbreakable children (long code blocks, tool-call outputs).
• .sidebar-item-time truncation — mirrors the chat-title row's overflow handling so a long model badge or localised timestamp can't break out of the 280px sidebar column.

Both documented in §8 of tests/web-ui-qa-checklist.md. Issue #36 (which I originally filed for the sidebar one as a follow-up) is closed as redundant — the fix went here.