AI SDLC Scaffold, repo template for AI-assisted software development
I built an open-source repo template that brings structure to AI-assisted software development, starting from the pre-coding phases: objectives, user stories, requirements, architecture decisions. It's designed around Claude Code but the ideas are tool-agnostic. I've been a computer science researcher and full-stack software engineer for 25 years, working mainly in startups. I've been using this approach on my personal projects for a while, then, when I decided to package it up as scaffold for more easy reuse, I figured it might be useful to others too. I published it under Apache 2.0, fork…
In plain words
AI SDLC Scaffold is an open-source repository template that organizes AI-assisted software development by establishing structure before coding begins. It guides users through defining objectives, user stories, requirements, and architecture decisions, working with AI coding tools like Claude Code. The template is tool-agnostic and designed for developers who want to provide AI agents with clear context about project goals and constraints, rather than jumping directly to implementation. It is available under Apache 2.0 license.
written from the facts on this page · September 2026
From the sources
In the maker’s words, at launch
I built an open-source repo template that brings structure to AI-assisted software development, starting from the pre-coding phases: objectives, user stories, requirements, architecture decisions. It's designed around Claude Code but the ideas are tool-agnostic. I've been a computer science researcher and full-stack software engineer for 25 years, working mainly in startups. I've been using this approach on my personal projects for a while, then, when I decided to package it up as scaffold for more easy reuse, I figured it might be useful to others too. I published it under Apache 2.0, fork it and make it yours. You can easily try it out: follow the instructions in the README to start using it. The problem it solves: AI coding agents are great at writing code, but they work much better when they have clear context about what to build and why. Most projects jump straight to implementation. This scaffold provides a structured workflow for the pre-coding phases, and organizes the output so that agents can navigate it efficiently across sessions. How it works: Everything lives in the repo alongside source code. The AI guidance is split into three layers, each optimized for context-window usage: 1. Instruction files (CLAUDE.md, CLAUDE.<phase>.md): always loaded, kept small. They are organized hierarchically, describe repo structure, maintain artifact indexes, and define cross-phase rules like traceability invariants. 2. Skills (.claude/skills/SDLC-*): loaded on demand. Step-by-step procedures for each SDLC activity: eliciting requirements, gap analysis, drafting architecture, decomposing into components, planning tasks, implementation. 3. Project artifacts: structured markdown files that accumulate as work progresses: stakeholders, goals, user stories, requirements, assumptions, constraints, decisions, architecture, data model, API design, task tracking. Accessed selectively through indexes. This separation matters because instruction files stay in the context window permanently and must be lean, skills can be detailed since they're loaded only when invoked, and artifacts scale with the project but are navigated via indexed tables rather than read in full. Key design choices: Context-window efficiency: artifact collections use markdown index tables (one-line description and trigger conditions) so the agent can locate what it needs without reading everything. Decision capture: decisions made during AI reasoning and human feedback are persisted as a structured artifact, to make them reviewable, traceable, and consistently applied across sessions. Waterfall-ish flow: sequential phases with defined outputs. Tedious for human teams, but AI agents don't mind the overhead, and the explicit structure prevents the unconstrained "just start vibecoding" failure mode. How I use it: Short, focused sessions. Each session invokes one skill, produces its output, and ends. The knowledge organization means the next session picks up without losing context. I've found that free-form prompting between skills is usually a sign the workflow is missing a piece. Current limitations: I haven't found a good way to integrate Figma MCP for importing existing UI/UX designs into the workflow. Suggestions welcome. Feedback, criticism, and contributions are very welcome!
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 27d ago · cactuscompute.com
Launched alongside, March 2026
the whole month →
Switch from ChatGPT to Claude with import memory feature
AI · Mar 2026 · claude.com
