AGENTS.md — keep your AI coding agent within the tool rules

Hi everyone,

More and more tools these days are built with the help of AI coding agents (Claude Code, Cursor, Copilot, Codex, and so on). That's a good thing overall — it lowers the barrier to building useful stuff. The catch is that these agents don't know OGame's rules or how the game actually works, so they'll happily implement things that are flatly forbidden: automation, scheduling, attack alarms and Discord pings, direct probing from custom target lists, hiding ads, background polling, and so on. A fair number of recent submissions have failed review for exactly these reasons.

To help with that, I've put together a resource you can drop straight into your project:

https://github.com/Rivenscryr/origin-tooldev-agents

What it is

It's an AGENTS.md file — a standard instructions file that most modern AI coding tools read automatically. This one encodes our forbidden-features and toleration rules as hard constraints, so your agent knows the boundaries while it's helping you build, instead of finding out at review time.

What it does

• Makes the agent refuse forbidden features — automation & macros, scheduling, auto-refresh & polling, automatically registered alarms/notifications/webhooks, direct probing, imitating Dark Matter features, blocking monetization, paywalls, silent data scraping — and explain why in plain language rather than quietly building them.
• Flags the gray-area features that need a ToolDev admin's approval before you build them.
• Enforces sane API and background-call behavior (the community proxy, page-load-only calls, the accountInfo and cp pitfalls, and what "activity" actually is and why it matters).
• Includes a "does my tool even need toleration?" guide, a pre-submission self-audit checklist, and a quick request → verdict table.

How to use it

1. Copy AGENTS.md into the root of your project's repository.
2. Most agentic tools pick it up automatically. If yours expects a different filename, copy or symlink it (CLAUDE.md, .cursorrules, .github/copilot-instructions.md), or just tell your agent at the start of a session: "Read AGENTS.md and follow it strictly."
3. Build. When you ask for something forbidden, the agent should push back and explain — don't override it. If you think you have a genuine exception, ask a ToolDev admin first.

A couple of important notes

• This is a recommendation, not a requirement, and it is not a guarantee of toleration. Tools that need review still get reviewed on their own merits. The file just makes it far less likely you'll build something that gets rejected on sight, and it makes reviews quicker for everyone.
• The forum threads remain the source of truth for the rules. The file only encodes them.
• AI agents still make mistakes. Read what yours produces, and when in doubt, ask.
• AGENTS.md is guidance for your agent, not a substitute for understanding your own code. You remain responsible for everything your tool does, whether you wrote it by hand or not.

Feedback wanted

This is a first version and I'd genuinely like your input — especially from experienced tool devs. If a rule is stated inaccurately, an example is unclear, or something important is missing, please open an issue or PR on GitHub, or just reply here. The rules have a lot of nuance, and more eyes will make this better for everyone.

One thing to keep in mind if you'd like to contribute: the file only encodes the rules, it doesn't set them. Anything that changes what's actually allowed has to come from an official decision, not a personal preference — so if in doubt, open an issue and let's discuss first.

Thanks, and happy building.

Quote from OGameDB

Hello RiV-

I have taken some time to read the agent file. Can i ask you to clarify below rule.

"Let me build and organize custom target lists (view only)"

✅ Allowed (§1.5.1)

Can you clarify is this is player based only or planet based, asking as up to now, i do only see player based listings ?

Thanks

Display More

Hi, and thanks for reading through it so carefully.

One important bit of framing first: the AGENTS.md is a set of guardrails for your coding agent, not the rules themselves. Dropped into a project, it stops the agent from building things that are clearly forbidden and nudges it to ask before anything borderline — but the actual authority is the Forbidden features guidelines and the ToolDev team's decisions on specific cases. So that quick-reference line isn't meant to be read as a precise boundary; it's a compressed summary of the fuller §1.5.1.

On the substance: player-vs-planet isn't really the axis that line is about. What it actually turns on is display vs. direct probing. Building and organizing custom target lists is fine either way — by player, or by planet/moon/coordinate, it doesn't matter. What's not allowed is bolting a direct probe onto those tool-provided targets: a "probe now" icon per coordinate, an auto-pulled inactives list you can probe straight from, or probing from within a custom list. Probing a new target has to go through the game's own flow — click the coordinate, land in the galaxy view, use the game's probe icon — or from a spy report already sitting in your inbox. That's the line; the granularity of the list isn't.

So planet-based listings aren't disallowed by that line — you're free to display them. If you've got a specific feature in mind, though, it's always worth running the exact design past us before you build it, precisely because the file is guardrails and not the last word. Happy to take a look.