Type: url
Original: https://developers.cyberagent.co.jp/blog/archives/62010/
Status: generated
Summary
A CyberAgent engineer walks through a five-phase AI-driven development workflow built around OpenAI Codex, using a sample feature-request voting board as the worked example. Each phase — vision, project initialization, requirements and domain modeling, architecture and stack selection, environment setup, and implementation — is presented as a concrete prompt together with the output the author expects back. The workflow’s core devices are AGENTS.md as the agent’s long-term memory, project-rules.md for naming and placement conventions, and a per-story context.json recording implementation state, so that prompts can state requirements rather than output paths.