How Vstorm Put AGENTS.md Into Every AI Agent-Ready Project It Ships

* Antoni Kozelski CEO & Founder Antoni Kozelski is the CEO and Founder of Vstorm. ---------------------------------------- Vstorm, an Applied Agentic AI Engineering Consultancy and AAIF Silver Member, embedded AGENTS.md auto-generation into its open source agentic AI project template. Every FastAPI + Next.js project the tool scaffolds now ships a tailored AGENTS.md and CLAUDE.md from day one: giving AI coding agents correct build commands, architectural conventions, and off-limits boundaries without any manual authoring. With 830,000+ PyPI downloads, the template has become one of the wider distribution channels for the AGENTS.md standard in the Python ecosystem. WHEN AI AGENTS READ YOUR CODE BUT MISS YOUR CONVENTIONS AI coding agents can now traverse entire repositories: reading, refactoring, and generating new features from plain language. The gap that remains is project knowledge. An agent understands syntax well and project conventions not at all. Without explicit guidance, agents guess. They run pip install instead of uv sync. They hand-edit auto-generated migration files. They apply architectural patterns the team stopped using months ago. On a single project, this costs hours of review. Across dozens of generated projects, it becomes a structural problem. For Vstorm’s OSS team, the friction was specific: the full-stack-ai-agent-template, a CLI that generates production-ready FastAPI + Next.js applications with 75+ configuration options, was producing projects that AI coding agents would routinely mishandle. There was no standardized briefing layer, so every agent working on a generated project had to guess from first principles. The fragmentation across tools compounded the problem. GitHub Copilot reads .github/copilot-instructions.md. Cursor uses .cursor/rules/. Claude Code reads CLAUDE.md. * “Open a typical project that’s been through a few months of AI-assisted development. You’ll find some combination of CLAUDE.md, .cursorrules, and copilot-instructions.md, AGENTS.md, and maybe a Gemini.md for good measure. Almost the same content in each one. Slowly drifting apart.” — Augment Code, “How to Build Your AGENTS.md”, March 2026 Teams using more than one tool faced maintaining two or three overlapping files, or maintaining none of them. Simon Willison, creator of Django and Datasette, described his own answer to this in September 2025: * “Rather than leaning on MCP, I like to create an AGENTS.md (or equivalent) file with details of packages I think they may need to use.” — Simon Willison, “Designing agentic loops”, September 2025 The instinct is the same one Vstorm acted on at scale: give agents a single, predictable place to read project instructions before they act. * “Your README explains your project to humans. Your .gitignore tells Git what to skip. But who tells Copilot not to touch your generated migration files? Without AGENTS.md, nobody does.” — Kacper Włodarczyk, Software Engineer, Vstorm [source] When the AAIF adopted AGENTS.md as the tool-agnostic open standard in December 2025, it gave the ecosystem a single convergence point. The question for Vstorm was how to make adoption automatic, not aspirational. BAKING AGENTS.MD INTO EVERY PROJECT FROM DAY ONE The solution was direct: generate the context file as part of the project scaffold itself. Vstorm added Jinja2-conditional logic to produce both AGENTS.md and CLAUDE.md alongside every generated project. The files are not generic boilerplate. Each one is tailored to the exact configuration the developer selected: * Choose PostgreSQL → Alembic migration commands appear in the Commands section * Enable JWT authentication → SECRET_KEY is listed in environment variables * Select PydanticAI as the agent framework → the project structure section shows the agents/ directory * Pick Docker → the compose commands are included The six sections that make up each generated file follow a structure designed to reduce the most common agent errors: project overview, build and test commands, directory structure, key conventions, files not to modify, and pointers to deeper documentation. AGENTS.md and CLAUDE.md serve distinct purposes. AGENTS.md is the tool-agnostic layer, covering any agent that reads the standard. CLAUDE.md covers Claude Code-specific depth: memory patterns, tool preferences, and project-specific interaction context. Both ship together so teams using multiple tools get complete coverage from the same scaffold. The same convention-awareness extends into Vstorm’s pydantic-deepagents framework (794 GitHub stars), which reads AGENTS.md context automatically at runtime and supports MCP natively, meaning agents executing in production inherit the same project knowledge that coding agents use during development. 830,000+ DOWNLOADS LATER The full-stack-ai-agent-template has reached 1,730+ GitHub stars and 830,000+ PyPI downloads. Every project generated by the tool ships AGENTS.md by default, no manual step, no documentation debt. Results across the template’s adoption: * Every generated project delivers a tailored AGENTS.md matched to its exact stack — database, auth method, AI framework, and deployment configuration * Teams using multiple coding agents (Copilot, Codex, Claude Code) get consistent coverage from one file rather than three * The pydantic-deepagents runtime extends AGENTS.md convention-awareness from development into production agent execution * MCP support in the framework means generated projects can connect any MCP server as a tool capability from the first line of code An AGENTS.md generated from configuration choices stays accurate as long as the template does. A manually maintained file starts drifting on day two. For an open source agentic AI ecosystem to deliver on the promise of consistent agent behaviour across projects, the standard needs to be as automatic as a .gitignore. WHAT COMES NEXT Vstorm is extending AGENTS.md generation across all repositories in the OSS ecosystem, not only the full-stack template. The team is also exploring how pydantic-deepagents can consume AGENTS.md at runtime to sharpen subagent specialisation in multi-agent systems, where convention consistency across parallel agents is an active engineering problem. TRY IT AND JOIN THE COMMUNITY Explore the full-stack-ai-agent-template and generate your first AGENTS.md-ready project: 🔗 github.com/vstorm-co/full-stack-fastapi-nextjs-llm-template Want to discuss AGENTS.md adoption, share your implementation, or connect with engineers working on open source agentic AI standards? 🔗 Join the AAIF Discord community ABOUT VSTORM Vstorm is an Applied Agentic AI Engineering Consultancy supporting mid-market organisations achieve agentic transformation. The team has delivered 30+ production agentic AI implementations and runs an open source — oss.vstorm.co — maintaining 13 repositories used by developers worldwide.