Skip to Main Content

Agentic tools automate code generation, but the bottleneck has moved. It's no longer how to build. It's what to build, and whether it's right. Agents don't check if a request conflicts with business goals, existing specs, or product policies. They build fast, correctly, and wrong. Specdown addresses this with three principles. Traceability: business goals, product goals, features, acceptance criteria, tests, and code are bidirectionally linked. Change one layer, the system knows what else breaks. Consistency: if a new spec contradicts existing specs or policies, it's caught before any code is written. Executable specs: one file is the spec, the test, and the report. Run it and you get green or red. Spec and test are the same artifact, so they can never drift apart. We built a Codex pipeline on top of this. User feedback comes in. Codex profiles who the user is, analyzes root cause against existing specs, generates a new spec verified for consistency against business goals and policies, builds code in parallel worktrees until specs pass, reviews the PR against the spec, and merges on approval. The spec is the interface between every Codex task. What gets planned becomes the build target, the review criterion, and the deploy condition.

Team

JayhojaeyoungHyunu

Placement

Participant

More from Global Codex Hackathon: [Singapore]

AI Tender Assistant is a web app that helps teams decide whether to bid on a tender and then generate a first-pass response package faster. It combines a…

Chieftns.ai project preview

CodexCalls is a real-time human-in-the-loop system that enables AI coding agents to safely make high-impact decisions by calling a human when needed. When an…

V2.AI project preview

End to end workflow of skills and a custom human annotation app for expert review and skill iteration.

Project Daedalus - PwC project preview

Launchpad is a Codex-powered starter platform for rapidly turning a client brief into a working prototype without leaving the Codex window. It combines…

Time Under Tension project preview

AutoCTO analyses GitHub organisations using a multi-agent Codex pipeline. It reviews repos for code quality, architecture, and tech debt, then surfaces…

loom.com/…

Project Worldview is an AI-powered command center that transforms fragmented project and system signals into a unified, real-time operational…

cap.so/…