Snapshot, plan-approval, and settings UX overhaul

[DevMando]

What this delivers

A round of UX work across three areas users touch constantly — saving conversation context (snapshots), approving proposed plans, and configuring an agent (settings) — plus a real bug fix in the settings form. The goal was to make these flows less blocking, easier to understand, and harder to get stuck in. All changes are in the desktop UI only; the shared agent engine is untouched.

Why it matters

• Fewer dead ends & clearer choices. Two flows (snapshot offer, plan approval) previously used blocking pop-ups that covered the very thing you needed to read. They're now non-blocking cards that keep the content visible.
• A stuck-state bug is fixed. Clearing a numeric setting could leave it permanently blank and unusable — a genuine "I can't recover without restarting" trap.
• Settings are simpler to navigate (fewer tabs, clearer grouping) and now have an escape hatch (reset to defaults).

What's changing (user-facing)

Saving a conversation (snapshots)

• When you switch a tab's model, a slim "Snapshot available" notification now slides in from the top instead of a bar pinned above your input. You can ignore it and keep working; it waits.
• Clicking Create expands it in place to name the snapshot and pick which model summarizes it, with a clear "Creating…" progress state (long summaries no longer look frozen).
• Snapshot cards are clearer: icon, timestamp, message count, and the name you gave it.

Approving a proposed plan

• The plan stays fully readable in the conversation. The approve/decline choice now appears as a bar at the bottom instead of a pop-up that hid the plan.
• Three clearly labeled, color-coded actions: Execute plan (run it step by step), One-shot it (skip the plan, do it in one pass — with a hover explanation), Cancel request (stop). Wording was reworked after testing showed "Reject" read like "Cancel."

Settings

• Reduced from 6 tabs to 4 (Model, Behavior, Integrations), with Appearance moved to its own left-rail button because it's app-wide, not per-agent.
• All MCP configuration now lives on the MCP page (the on/off toggle moved there), so it's in one place.
• Each tab has a Reset to factory defaults button (leaves your connection details and API key alone).
• Bug fix: clearing a numeric field (e.g. Max response tokens) left it stuck empty with broken up/down buttons; it now restores the last valid value, and all numeric fields have safe minimums.

Polish

• New "Phosphor Fwog" color theme.
• Plain-language pass on descriptions/tooltips (clearer explanation of what MCP and snapshots are; removed CLI-only jargon).
• Consistent status-pill sizing and rail tooltips.

Scope, risk, and testing

• Scope: desktop UI/UX only. No changes to the underlying agent/model engine.
• Build: compiles clean (0 warnings, 0 errors).
• Not yet exercised end-to-end in the running app — the interaction/animation changes (top snapshot card, bottom plan bar, reset button, numeric-field recovery) should get a manual click-through in QA before release.
• History: builds on the recently-merged snapshot work (PR Rework context snapshots into opt-in, LLM-summarized handoffs + UI polish #3); this PR's diff contains only the new UX changes described above.

🤖 Generated with Claude Code