One SKILL.md package that initializes any project to a universal documentation standard — works across IDEs, AI coding apps and model backends. When a new project is created, the agent will:
One SKILL.md package that initializes any project to a universal documentation standard — works across IDEs, AI coding apps and model backends. When a new project is created, the agent will:
main + a baseline .gitignore.PRD.md, TRD.md and EXECUTED.md (append-only execution log) at the project root.docs/ in four audience tracks — ai/, technical/, layman/, non-technical/ — with an index README.AGENTS.md (+ CLAUDE.md / GEMINI.md pointers) so every future agent session keeps docs in sync automatically.README.md per folder and prepends a header comment to every source file — purpose, Docs: link, author, ISO-8601 timestamp.npx skills add mridun-gupta/project-init.skills # GitHub Copilot gh skill add mridun-gupta/project-init.skills # Claude Code (repo self-hosts a marketplace) /plugin marketplace add mridun-gupta/project-init.skills # Gemini CLI gemini skills install https://github.com/mridun-gupta/project-init.skills
Vendor-neutral manual spot: copy skills/project-init/ into ~/.agents/skills/ (reaches Devin, Cursor, Codex, Copilot, Gemini CLI, Zed, Cline…) or <repo>/.agents/skills/ per-project. Full per-tool matrix: INSTALL.md.
upgrade later.It always asks first — project name, one-line purpose, mode, author (defaults to git config user.name). Nothing is fabricated; unknowns become TODO markers.
my-project/
├── PRD.md TRD.md EXECUTED.md # documents of record
├── AGENTS.md # standing conventions
├── CLAUDE.md GEMINI.md # pointers → AGENTS.md
├── README.md .gitignore LICENSE
└── docs/
├── README.md # KB index
└── ai/ technical/ layman/ non-technical/ # 4 seeded tracks
Claude Code · Codex CLI / ChatGPT desktop · Cursor · VS Code Copilot · Devin · Google Antigravity · Gemini CLI · Windsurf · Kiro · Amp · OpenCode · Zed · Cline · Roo — anything reading the open Agent Skills format. For tools without skill support (ChatGPT web, Grok, local models): paste skills/project-init/SKILL.md as a system prompt, or drop templates/AGENTS.md into the project.
Known limits: auto-invocation on "new project" phrasing depends on the host agent (manual /project-init always works); files that can't hold comments (JSON, lockfiles, binaries) get no header; full on very large repos asks you to scope.
Docs mirror — canonical source is the repo's README/INSTALL.