AI Mayor v0.3.0: Cities Can Now Be Read, Fixed, Checked, and Saved

I released cities-skylines1-agent-skill v0.3.0.

This release strengthens the loop for letting an AI agent help with Cities: Skylines 1 city-building: read the current city, make small fixes, check the result, and save.

Instead of asking an AI to look at the screen and click around blindly, the agent can read in-game state through local APIs, inspect roads, zones, taxes, and problem icons, then execute small targeted operations.

Demo video:

https://x.com/hAru_mAki_ch/status/2054546518473158711

What Can This Release Do?

The AI can act more like a city assistant inside Cities: Skylines 1.

For example, it can help with tasks like these:

• Read population, demand, and problem icons
• Inspect roads, water pipes, power lines, and facilities
• Check whether roads are really connected or awkwardly overlapping
• Add roads and infrastructure when needed
• Place, move, or remove buildings
• Paint zones
• Change tax rates and simulation speed
• Save the game

The key point is that the AI does not have to rely only on what the screen looks like. It can read game data first, then act.

What It Looks Like

The image below shows an example city worked on by the AI. Roads, residential areas, service buildings, and connections to outside roads have been created.

The upper-left area of the game screen also shows API operation history. You can see actions such as reading city problems, building roads, and saving, which makes it easier to follow what the AI is doing in the background.

How Does It Work?

The basic structure has three parts:

1. Install the MOD for Cities: Skylines 1
2. The MOD starts a local API
3. Codex or another AI agent calls that API

The API is the interface that lets the AI ask the game to perform operations.

For example, a human would select the road tool with a mouse and draw a road. The AI can instead send a request like this:

Build a Basic Road from this coordinate to that coordinate.

Then the MOD calls the corresponding Cities: Skylines 1 game-side operation and creates the road.

Why Use APIs Instead of Only Screenshots?

Showing the game screen to an AI can work for some operations.

But in Cities: Skylines 1, many important details are hard to judge from visuals alone:

• Roads may look like they intersect, but not be connected internally
• Water pipes and power lines can be hard to verify from the current camera angle
• Problem icons are small and can be hard to read depending on zoom and angle
• A save file may not be easy to verify from the screen alone

That is why this release focuses on letting the AI read internal game data directly.

For example, /state/road-anomalies can inspect overlapping roads, dead ends, and places where roads visually cross without a node. /state/saves can confirm whether save files actually exist.

Main Changes in v0.3.0

v0.3.0 adds more observation points for growing and maintaining a city.

• Read demand bars: /state/demand
• Read citizen messages: /state/chirps
• Read zone area: /state/zones
• Read tax rates: /state/economy
• Check outside road connections: /state/external-connections
• Inspect zone gaps and mixed zoning: /state/zone-anomalies

It also adds more operation-side APIs.

• Change tax rates: /commands/set-tax-rate
• Turn buildings on or off: /commands/set-building-active
• Disable blocked assets: /commands/disable-blocked-assets
• Repair zones around existing growable buildings: /commands/repair-zones-to-growables
• Repair patchy zone clusters: /commands/repair-zone-clusters

In short, it is now easier to build a loop where the AI reads the city, makes a small fix, and checks again.

A Good Starting Workflow

Rather than asking the AI to do everything at once, I recommend starting with a small request like this:

Check this city's problems through the API.
If there are road or infrastructure issues, fix them one by one.
After making changes, save the game and report what you did.

The agent usually follows a loop like this:

1. Read the city summary with /state/summary
2. Read problem icons with /state/problems
3. Inspect road anomalies with /state/road-anomalies
4. If needed, remove bad roads or buildings with /commands/bulldoze
5. Rebuild with /commands/build-network or /commands/place-building
6. Advance time a little with /commands/set-simulation-speed
7. Check the city again through the API
8. Save with /commands/save

The important part is not to fix everything in one big step. Make a small change, check it, save it, and repeat.

Who Is This For?

This project is for people who:

• Like Cities: Skylines 1
• Want to experiment with AI agents operating a game
• Are interested in automation based on internal game data, not only screen recognition
• Want to try MOD development and local API integration
• Want to let an AI think, execute, verify, and save in a controlled loop

It is still experimental, so I do not recommend using it first on an important production save. Start with a test save or a new map.

Summary

v0.3.0 moves cities-skylines1-agent-skill closer to an operational loop where an AI can read city state, make grounded fixes, and verify the result, rather than just experimentally controlling Cities: Skylines 1.

For AI-assisted city operation, the fundamentals matter more than flashy automation:

• Read the city state correctly
• Split work into small operations
• Check again after each operation
• Save so you can recover

v0.3.0 adds more APIs and workflow support for running that basic loop.

Demo video:

https://x.com/hAru_mAki_ch/status/2054546518473158711

Related links:

v0.3.0 release notes

https://sunwood-ai-labs.github.io/cities-skylines1-agent-skill/ja/releases/v0.3.0/

API reference

https://sunwood-ai-labs.github.io/cities-skylines1-agent-skill/ja/api

Agent operation guide

https://sunwood-ai-labs.github.io/cities-skylines1-agent-skill/ja/guide/usage

GitHub repository

https://github.com/Sunwood-ai-labs/cities-skylines1-agent-skill