Developer Tools●Mid
Almost Vibe Coded 3D C++ OpenGL Engine
Agentic-coded 3D engine; impressive if the AI actually nailed the architecture.
WizardryBold Bet
Umomgg
203mo ago
Back to browse
Show HN submissions tripled and are now mostly vibe-coded
by hubraumhugo·Apr 22, 2026·333 points·235 comments
AI Analysis
●MidDark Horse
Detects AI design slop with CSS checks, not LLM vision.
Strengths
- •Deterministic DOM/CSS checks instead of subjective LLM scoring
- •Identified 15 specific patterns like Inter font, colored left borders
- •Manual QA verified only 5-10% false positive rate on 500 sites
Weaknesses
- •It's observational research, not a tool you can actually use
- •Scoring code not open source yet, limits replication
Category
Target Audience
HN community members, designers tracking AI trends
Similar Projects
Developer Tools●●●Banger
Vibe-coded your repo? Drift finds the architecture problems
Catches AI code debt deterministically with AST analysis when linters only check style.
Solve My ProblemBig BrainDark Horse
sauremilk
102mo ago
Education●Mid
A logic puzzle game I built for my 5-year-old daughter(mostly AI coded)
Cute UI, but puzzle mechanics and progression are unclear from the landing page.
Eye CandyShip It
oser
203mo ago
AI/ML●●●Banger
Vibe code your agents without vibe coding your agent
Closes the loop: agents read eval traces to fix their own regressions.
Big BrainSolve My Problem
jeffreyip
601mo ago
Developer Tools●●Solid
Spekkio: Reverse-engineer specs from vibe-coded apps
Spec extraction from vibe-coded apps via reverse engineering—ambitious, but early and single-integration.
Big BrainBold Bet
tessier0ashpool
503mo ago
Open Source●●Solid
I analyzed 157K HN posts and built skills with guardrails against BS
Three practical Claude skills — /hn-score, /hn-spin, and /hn-optimize — bundle headline angles, timing advice, and an 87-point scorecard driven by a 157K-post dataset and a reverse-engineered HN ranking formula. The repo's real cleverness is the HN-specific heuristics (penalty multipliers, flamewar detection, domain performance stats) turned into actionable prompts, not merely another headline generator. It's an MVP that rewards anyone who actually runs Claude and wants data-backed posting playbooks, though it needs more launch-day case studies to prove the claims.
Niche GemBig Brain
buss_jan
203mo ago