refactor: composer surface polish, header layout cleanup, and model selector relocation

[batmn-dev]

Summary

• Align the composer surface with ChatGPT-style shadow tokens and refine prompt suggestion styles
• Move the model selector from the header into the multi-chat input area and simplify layout
• Add cursor-pointer to the shared button base so all buttons get it by default

Composer & Styling

• Apply shadow-composer and composer-bg CSS tokens to PromptInput instead of generic border/shadow-xs
• Darken --secondary to oklch(0.30) for better contrast in dark mode
• Bump header height from 52px to 56px
• Switch prompt suggestions from outline/ghost to secondary/ghost with hover:bg-secondary
• Remove manual cursor-pointer from suggestions (now inherited from button base)
• Remove the "Writing" suggestion category

Header & Layout

• Convert header from sticky to fixed positioning
• Switch app shell from h-svh to h-dvh
• Simplify the header three-column layout structure
• Remove ModelSelectorHeader component and header-level model selector
• Update scroll-area calc to account for fixed header height

Multi-Model State

• Delete MultiModelSelectionProvider context — localize multi-model selection state into MultiChat component
• Add ModelSelector directly inside MultiChatInput with new onSelectedModelIdsChange prop
• Adjust multi-conversation top padding for fixed header clearance (pt-4 -> pt-20)

Cleanup

• Remove obsolete .agents/ plan and skill files superseded on main

Test Plan

• Verify composer input has correct shadow/background in light and dark mode
• Verify all buttons show pointer cursor on hover
• Verify prompt suggestions use correct variant and hover styles
• Verify header stays fixed at top during scroll
• Verify model selector works in multi-chat input area
• Verify single-model and multi-model modes still function correctly
• Run bun run lint and bun run typecheck with no regressions

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
not-a-wrapper	Ready	Preview, Comment	Feb 21, 2026 2:23am

[cubic-dev-ai]

No issues found across 7 files

[greptile-apps]

Greptile Summary

This PR refactors the composer surface styling to align with ChatGPT's design tokens and consolidates button cursor behavior. The changes improve visual consistency in both light and dark modes while simplifying the component styling hierarchy.

Key Changes

• Composer styling: Migrated from generic border/shadow-xs to ChatGPT-aligned shadow-composer utility and --composer-bg CSS tokens in PromptInput
• Button base enhancement: Added cursor-pointer to the shared button variants, removing the need for manual cursor styling in individual components
• Dark mode contrast: Darkened --secondary from oklch(0.32) to oklch(0.30) for better readability
• Prompt suggestions: Switched from outline/ghost to secondary/ghost variants with hover:bg-secondary for consistent hover states
• Configuration cleanup: Removed "Writing" suggestion category from SUGGESTIONS array
• Documentation updates: Expanded ChatGPT design token reference with detailed shadow strategy analysis; added PR creation guidelines and timeline estimate restrictions to AGENTS.md and CLAUDE.md

All changes are cosmetic refactoring with no functional impact. The CSS tokens are properly defined in globals.css with mode-aware values.

Confidence Score: 5/5

• This PR is safe to merge — all changes are cosmetic styling refactoring with no functional modifications
• The refactoring is well-executed with proper CSS token definitions, consistent application across components, and no breaking changes. All modified files show clean, focused changes that align with the stated objectives.
• No files require special attention

Important Files Changed

Filename	Overview
components/ui/button.tsx	Added cursor-pointer to button base variants for consistent pointer cursor across all buttons
components/ui/prompt-input.tsx	Switched from generic border/shadow to ChatGPT-aligned shadow-composer and --composer-bg CSS tokens
components/ui/prompt-suggestion.tsx	Changed variant from outline/ghost to secondary/ghost, updated hover from bg-accent to bg-secondary, removed manual cursor-pointer
app/components/chat-input/chat-input.tsx	Removed redundant bg-popover, shadow-xs, and backdrop-blur-xl classes from PromptInput (now handled by component defaults)
app/globals.css	Darkened --secondary from oklch(0.32) to oklch(0.30) in dark mode for better contrast
lib/config.ts	Removed "Writing" suggestion category and cleaned up unused Note01Icon import

_{Last reviewed commit: 9cd9a98}

[cubic-dev-ai]

1 issue found across 2 files (changes from recent commits).

Prompt for AI agents (all issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="AGENTS.md">

<violation number="1" location="AGENTS.md:132">
P3: Fix the typo in the new guideline so the documentation reads correctly (“appropriate”).</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

[cubic-dev-ai]

P3: Fix the typo in the new guideline so the documentation reads correctly (“appropriate”).

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At AGENTS.md, line 132:

<comment>Fix the typo in the new guideline so the documentation reads correctly (“appropriate”).</comment>

<file context>
@@ -127,6 +127,10 @@ When the user asks to "create a prompt" (or similar), return the prompt directly
 
+## No Timeline Estimates
+
+**Never include time estimates, durations, or effort assessments** in plans, summaries, or implementation outputs. This includes phrases like "~30 minutes", "2-3 hours", "Phase 1 (Day 1)", "Quick win", or any similar timeline/effort language. AI-generated timeline estimates are unreliable and misleading. Only include difficulty complexity risk estimates when appropiate.
+
 ## Critical Patterns
</file context>

Suggested change

	**Never include time estimates, durations, or effort assessments** in plans, summaries, or implementation outputs. This includes phrases like "~30 minutes", "2-3 hours", "Phase 1 (Day 1)", "Quick win", or any similar timeline/effort language. AI-generated timeline estimates are unreliable and misleading. Only include difficulty complexity risk estimates when appropiate.
	**Never include time estimates, durations, or effort assessments** in plans, summaries, or implementation outputs. This includes phrases like "~30 minutes", "2-3 hours", "Phase 1 (Day 1)", "Quick win", or any similar timeline/effort language. AI-generated timeline estimates are unreliable and misleading. Only include difficulty complexity risk estimates when appropriate.