feat(landing): redesign landing page and update pipeline

[sungmanch]

Summary

• Landing page redesign: Remove 17 deprecated components (AnalysisShowcase, SectionNav, etc.) and add 6 new components (BrandLogo, HeroFlowchart, LandingHeader, PricingSection, ProductPreviewSection, TeamSection) with updated layout and styling
• Analyzer pipeline updates: Refine all worker prompts and stage prompts, merge duplicate filter functions in content-writer, remove dead code, and update data schemas for coding-style, learning-behavior, session-outcome, and thinking-quality
• Dashboard restructuring: Simplify BenchmarksContent wrapper patterns, extract className helpers, hoist static data to module level, and clean up redundant default exports across report components

Changes

• refactor(landing): remove deprecated landing page components (17 files deleted)
• feat(landing): add new landing page components (12 files added)
• refactor(landing): update existing components and views for redesign (17 files modified)
• refactor(analyzer): update pipeline prompts and data schemas (22 files modified)
• refactor(dashboard): restructure app routes and report components (22 files modified)
• chore: update tests, documentation, scripts, and package config (9 files modified)

Test Plan

• TypeScript compilation passes (tsc --noEmit)
• All 1157 tests pass
• Manual verification of landing page layout
• Manual verification of dashboard/report views

---

Generated with Claude Code using /ship-it

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
no-more-ai-slop	Ready	Preview, Comment	Feb 26, 2026 5:15pm

[sungmanch]

Code Review

Found 2 issues (filtered from 17 candidates, threshold: score ≥ 80/100):

---

1. 🔴 Prompt–Schema Enum Mismatch → Zod Validation Failure (Score: 100/100)

session-outcome-prompts.ts instructs the LLM to use review_feedback as a goal category, but session-outcome-data.ts GoalCategoryEnum only accepts code_review. When the LLM follows the prompt and outputs "primaryGoal": "review_feedback", Zod validation will fail and crash the SessionOutcome worker.

This is the same class of bug documented in CLAUDE.md: "SessionOutcomeWorker prompt lacked MINIMUM 150 characters guide for recommendation, causing Zod validation failure"

• Prompt: https://github.com/sungmanch/no-more-AI-slop/blob/d3745a93a5c4e9ae86a4eaf52ae510d70a982640/src/lib/analyzer/workers/prompts/session-outcome-prompts.ts#L56
• Schema: https://github.com/sungmanch/no-more-AI-slop/blob/d3745a93a5c4e9ae86a4eaf52ae510d70a982640/src/lib/models/session-outcome-data.ts#L57

Fix: Either rename code_review → review_feedback in GoalCategoryEnum, or revert the prompt change.

---

2. 🟡 Incomplete Rebrand in User-Facing Survey (Score: 100/100)

SurveyBottomSheet.tsx STEPS array has 4 questions — step 1 was updated to "BetterPrompt" but steps 0, 2, and 3 still say "NoMoreAISlop". Users see inconsistent branding within the same survey flow.

• File: https://github.com/sungmanch/no-more-AI-slop/blob/d3745a93a5c4e9ae86a4eaf52ae510d70a982640/src/components/report/SurveyBottomSheet.tsx#L29-L32

Fix: Update all 4 STEPS to use "BetterPrompt" consistently.

---

Generated with Claude Code using /ship-it

_{If useful, react with a thumbs-up. Otherwise, thumbs-down.}

[sungmanch]

Code Review — Quick Fix Pipeline (6 new commits)

Found 4 issues (score >= 80/100):

---

1. Wrong Gemini model in desktop analyzer (Score: 95)

Desktop quick-fix-analyzer.ts calls gemini-2.0-flash but CLAUDE.md specifies gemini-3-flash-preview for all pipeline stages. This causes output quality divergence between desktop and server-side analysis paths.

https://github.com/sungmanch/no-more-AI-slop/blob/2bb5b9c7888175c05d9fafeb4a16c3c770f33a6d/packages/desktop/src/main/quick-fix-analyzer.ts#L204

---

2. Path decoding breaks hyphenated project names (Score: 90)

dirName.replace(/-/g, '/') converts ALL hyphens to slashes. A project at /home/user/my-project (encoded as -home-user-my-project) becomes /home/user/my/project. The codebase has decodeProjectPath in src/lib/parser/jsonl-reader.ts — the desktop module should reuse it or replicate its logic correctly.

https://github.com/sungmanch/no-more-AI-slop/blob/2bb5b9c7888175c05d9fafeb4a16c3c770f33a6d/packages/desktop/src/main/ipc-handlers.ts#L218-L221

---

3. BrowserWindow destroyed during async analysis → main process crash (Score: 85)

mainWindow is captured at line 262, but analyzeQuickFix runs for ~30 seconds. If the window is closed mid-analysis, mainWindow.webContents.send() in the onProgress callback throws Object has been destroyed. This exception escapes the outer try/catch (it fires inside the callback) and can crash the main process as an unhandled rejection.

https://github.com/sungmanch/no-more-AI-slop/blob/2bb5b9c7888175c05d9fafeb4a16c3c770f33a6d/packages/desktop/src/main/ipc-handlers.ts#L262-L299

---

4. Empty catch block violates No Fallback Policy (Score: 80)

Session file parsing loop has a completely empty catch {} block. If 4 of 5 files fail to parse, the user gets "1 session analyzed" without knowing data was lost. At minimum, add console.warn() to match the pattern used by all other IPC handlers in this file.

https://github.com/sungmanch/no-more-AI-slop/blob/2bb5b9c7888175c05d9fafeb4a16c3c770f33a6d/packages/desktop/src/main/ipc-handlers.ts#L282-L284

---

Notable (below threshold): apiKey = '' on QuickFixPage.tsx:111 (marked TODO — feature non-functional until settings UI exists), overallHealthScore || 50 fallback in desktop analyzer (No Fallback Policy), type drift across 3 independent QuickFixResult definitions.

Generated with Claude Code

_{If useful, react with a thumbs-up. Otherwise, thumbs-down.}