Type: url
Original: https://developers.cyberagent.co.jp/blog/archives/60229/
Status: generated
Summary
A CyberAgent Developers Blog post from the company’s 2025 Advent Calendar, written by a generative-AI and backend tech lead running a proof of concept for a spec-driven development platform on an existing large application. It sets out CyberAgent’s four-level AI maturity model for product development (L1 code completion to L4 PRD to production), defines spec-driven development as treating specifications as structured context from which AI generates code, and names three context problems — context rot, context confusion and context poisoning — that context engineering is meant to address. Most of the post is a set of design practices for custom slash commands: one responsibility and one output file per command, minimal context, predefined templates, golden-master testing, tool-independent command files, avoiding sub-agents, and keeping instructions simple.