Workshops: Agentic Coding Tools in Depth
One Repo, Many Agents: AGENTS.md, CLAUDE.md & Cursor Rules Workshop
When one team uses Claude Code, another Cursor and the cloud agent reads AGENTS.md, instructions drift apart fast and every agent behaves differently in the same repository. In this half-day workshop you build a single source of truth for agent instructions, connect each tool to it and adopt repository conventions that make every agent more reliable.
Who is this workshop for?
- Tech leads and maintainers of repositories used with several coding agents
- Platform and developer experience teams setting standards across many repositories
- Teams with a mixed tool landscape of Claude Code, Cursor, GitHub Copilot or OpenAI Codex
- Monorepo owners who need different instructions per package
- Architects who want conventions followed by agents and people alike
What you will take away
- Know which instruction files Claude Code, OpenAI Codex, Cursor, GitHub Copilot and other agents read, and how they are scoped.
- Set up AGENTS.md as the shared core and connect CLAUDE.md, Cursor rules and Copilot instructions to it without duplicating content.
- Structure instructions for monorepos with nested files and path-specific rules.
- Make your repository agent-friendly with one-command setup, unambiguous test and lint commands and a clear structure.
- Add a CI check that detects drift between instruction files.
Agenda
Module 1
Who reads what?
- AGENTS.md, CLAUDE.md, Cursor project rules, Copilot instruction files and other tool-specific files
- Scope and precedence: global, repository, directory, file pattern
- Exercise: inventory of instruction files in your own repositories
Module 2
One source of truth
- A shared core in AGENTS.md, thin tool-specific layers on top
- Imports, references and symlinks: what works where
- What belongs in the core and what only in one tool's file
Module 3
Conventions that help every agent
- Setup, build, test and lint as single, documented commands
- Folder structure, naming, architecture notes and decision records
- Ignore files and off-limits areas for agents
Module 4
Staying consistent
- Ownership and review of instruction changes
- A CI check for drift and outdated commands
- Measuring the effect: fewer corrections in review
Why copied instructions fail
Copied instructions diverge within weeks. One file still names the old test command, another forbids a library the team has long since adopted. Agents follow what they read, so inconsistent files produce inconsistent code, and reviewers end up correcting the same mistakes in different tools.
A layered structure fixes this: shared rules live in one place, each tool gets only what is truly specific to it, and the repository itself carries the conventions through scripts, linters and tests.
Less prose, more automation
The best instruction is one an agent cannot skip. Wherever possible we turn written rules into executable checks. Formatters, linters, type checks and tests give every agent and every developer the same immediate feedback.
- A rule that can be linted belongs in the linter
- A command that must be run belongs in a script
- Prose remains for intent, context and trade-offs
What you take home
A reviewed AGENTS.md core for one of your repositories, connected tool files, a template for further repositories and a drift check for your CI. Guided by neooutput AI Code Engineers, remote or on-site, in English or German.
Frequently asked questions
Do all coding agents support AGENTS.md?
Many do, including OpenAI Codex, Cursor and GitHub Copilot, and support keeps growing. Claude Code uses CLAUDE.md, which can import other files, so a CLAUDE.md that references AGENTS.md is a common pattern. We check the current behaviour of your tools during the workshop.
How long should an instruction file be?
As short as possible and as long as necessary. Every line takes up context in every session, so we focus on what agents cannot infer from the code: commands, conventions, boundaries and pitfalls.
We already have a CONTRIBUTING guide. Isn't that enough?
It is a good basis, but it is written for humans and often lacks exact commands and boundaries. We reuse what fits and reference it rather than duplicating it.
Workshop
Get in touch
Should all your coding agents follow the same rules in the same repository?
Tell us roughly how many people will join and pick dates that work – we will get back to you within 24 hours with a proposal for content, format and schedule.
Prefer to talk first?
15 minutes, free and without obligation – with our AI Trainers and AI Code Engineers.
Book a call for free (opens in new tab) +49 761 519 751 50 (24/7)Request a training / workshop
Workshop: One Repo, Many Agents: AGENTS.md, CLAUDE.md & Cursor Rules Workshop · Half day (approx. 4 hours)
More workshops
- Workshops: Agentic Coding Tools in DepthMCP Servers in Daily Development: Hands-On Training for Dev Teams
- Workshops: Agentic Coding Tools in DepthRolling Out Claude Code in the Enterprise: Admin Workshop
- Workshops: Agentic Coding Tools in DepthBrowser Agents with Playwright MCP: UI Testing & Debugging Workshop
- Workshops: Agentic Coding Tools in DepthClaude Code Hooks, Skills & Plugins: Advanced Team Workshop
- Workshops: Agentic Coding Tools in DepthClaude Code Subagents & Git Worktrees: Parallel Agents Workshop
- Agentic Coding ToolsSecurity Risks of Coding Agents: Prompt Injection, Secrets, Permissions
