new-project-init

Your project has been running for months, documentation is scattered, and switching to a new AI tool means starting from zero with no context on what was done, what the conventions are, or what comes next. new-project-init tackles exactly that. Instead of writing code for you, it runs a question-driven workflow: it first learns your tech stack, team habits, and current progress, then generates a tailored set of specification files (platform entry doc, memory three-piece set, docs, and module spec five-piece set) so that any AI entering the project knows what to read first, how to work, and how to leave a record. Its core scenario is remediating existing documentation: it organizes docs and workflows without touching your code or refactoring in-flight modules.

The capabilities are organized in three progressive layers. The first is graded questioning: critical decisions are asked one at a time with recommendations and trade-offs, batch questions can be accepted with one-click defaults, and any round you want to skip is simply skipped. The second is a closed-loop workflow: plan, develop, independent review, test, and retrospective are sequential gates where each file must be confirmed before the next one begins, preventing granularity-jump rework. The third is memory and discipline: the three-piece memory set is read on every entry, activity logs are written on trigger, and each convention is defined in exactly one place with everywhere else referencing it, structurally eliminating multi-source drift. Thirty-one templates cover Java/Web, C++ backend, and embedded (STM32/ESP32) directions, specialized by your answers.

new-project-init is designed for three kinds of people: teams or individuals whose project docs have gotten messy and who want to establish order without rewriting code; developers joining an existing project mid-stream and needing to quickly build a complete documentation system; and tech leads starting a new project who want to lock down the documentation structure, AI collaboration workflow, and memory discipline from day one. It is cross-platform: the rules, templates, and outputs are platform-agnostic and work with DeepSeek Harness, Claude Code, Reasonix, Codex, and any other skill-capable agent. Simply tell your AI to use new-project-init to improve documentation and let it guide you through the questions.