aaravkashyap12/advise-project-approach

This is an add-on skill for AI coding agents like Claude Code and OpenAI Codex. Once installed, you can ask your agent to research the best way to build a project before you start writing code. Instead of getting a generic answer, the agent follows a structured process: gather your constraints, find real comparable open-source projects, study what those projects do and why, compare the tradeoffs, then give you a recommendation along with the conditions under which that recommendation would become the wrong one. The skill works at three stages. Before you start, it helps you pick the right technology stack by studying similar finished projects. Midway through a project, it can inspect your own code repository and identify what actually needs fixing, ordered by impact rather than trend. After you finish, it can review your project against mature examples and call out gaps to address before shipping. The README emphasizes that claims need to be grounded in real evidence. The skill is designed to refuse to invent star counts or commit dates, and to say clearly when it only had a description to work with rather than actual files to inspect. If you give it a large codebase, it maps the structure first and samples by relevance rather than reading everything. Installation requires Node.js and runs a single command to fetch the skill from GitHub. You can also download the packaged file manually and upload it through your agent's skill settings, or copy it into a local skills folder if your agent supports that. The repository includes example outputs showing pre-build, mid-build, and post-build scenarios, as well as comparisons between answers generated with and without the skill active. The project is in Python and released under an open license. Version 0.2 added a clearer decision methodology and stricter rules about what counts as acceptable evidence in a recommendation.