Vibe coding is fast, but production handoff is fragile. Conservatory is a review-first AI Agent built with OpenAI & Codex ChatGPT-6 Astra that audits web apps for Supabase RLS leaks, WCAG & AEO, and generates safe GitHub Draft PRs—keeping humans in 100% control.

What became possible in your product with Astra that was not practical before?

Before ChatGPT-6 Astra and advanced OpenAI models, code analysis tools were either rigid static scanners that flooded developers with false positives, or basic LLM chat assistants that hallucinated code and broke production builds during fixes. With Astra's multi-step reasoning and Codex capabilities, Conservatory can perform deep AST analysis across entire AI-generated web apps (built with tools like Lovable, Base44, Bolt, v0, and Cursor). It intelligently cross-references frontend client bundles with backend database rules (such as detecting missing Supabase Row-Level Security policies), translates complex security and WCAG audit evidence into precise code diffs, and prepares human-reviewable GitHub Draft Pull Requests—eliminating the "Fix-and-Break" loop without requiring risky autonomous merges.

Hello Product Hunt 👋 Luciano here, founder of Symphony. A few months ago, we launched Prelude on Product Hunt — an open-source terminal CLI to reveal why websites fail AI search readiness. The response from developers was amazing, but people kept asking us the exact same follow-up question: "Prelude showed me what's broken in my web, schema, and AI readiness... but how do I fix it safely without breaking production?" At the same time, the boom of Vibe Coding (Lovable, Base44, Bolt.new, v0, Cursor) made building the first draft of an app addictively fast. But moving those AI-built drafts into production introduced a whole new set of headaches: • Critical Security Leaks: Missing Supabase Row-Level Security (RLS) policies and exposed keys. • Accessibility & AI Invisibility: Over 95% of AI-generated layouts fail WCAG standards, making them invisible to AI search engines (AEO/GEO). • The Fix-and-Break Loop: Asking AI agents in chat to fix bugs often results in regressions or broken production deployments. So we built Conservatory v1.1 for the GPT-6 Astra Challenge. 🚀 Built with OpenAI & Codex, Conservatory is a neutral, review-first AI Agent designed to turn report evidence into safe, reviewable code: 1. Audit & Evidence: Scans public URLs or repos for cybersecurity leaks, WCAG accessibility, and AEO/GEO signals, producing an immutable Evidence Report (PDF/Markdown). 2. AI Agent Remediation: Explains report evidence, prioritizes launch-blocking issues, and prepares safe fix previews. 3. GitHub Draft PRs: Opens reviewable Draft Pull Requests on GitHub. We enforce a strict human-in-the-loop rule: no autonomous merges, no unreviewed deploys. You stay in 100% control. 🎁 GPT-6 Astra Challenge Special: You can run a free instant audit right now at https://conservatory.app. Use the promo code ASTRA2026 inside the app to unlock 3 full Evidence Reports and GitHub Draft PR remediations for free! I'd love your honest feedback today: • If you're using AI builders, what is your biggest headache when moving from preview to production? • How does your team currently review AI-generated code before client handoff? • Does the review-first (Draft PR) workflow give you the confidence you need? I’ll be here all day answering everything 🙌 Thanks!

What became possible in your product with Astra that was not practical before?

Conservatory by Symphony

Review-first AI Agent for Code Security & GitHub PRs

Review-first AI Agent for Code Security & GitHub PRs