|
Download README.md from AgenticGames/agentic-kingdoms-agent-sessions: direct link, hf CLI and curl.
- Browser
- Download file 3.4 kB
-
https://huggingface.co/datasets/AgenticGames/agentic-kingdoms-agent-sessions/resolve/main/README.md
- Command line
-
hf download hf://datasets/AgenticGames/agentic-kingdoms-agent-sessions/README.md
-
curl -L -o README.md https://huggingface.co/datasets/AgenticGames/agentic-kingdoms-agent-sessions/resolve/main/README.md
3.4 kB
| license: cc-by-4.0 | |
| language: | |
| - en | |
| pretty_name: Agentic Kingdoms agent sessions | |
| tags: | |
| - agent-traces | |
| - agents | |
| - mcp | |
| - tool-use | |
| - games | |
| - strategy-game | |
| - hermes-agent | |
| - claude | |
| size_categories: | |
| - n<1K | |
| configs: | |
| - config_name: default | |
| data_files: | |
| - split: train | |
| path: "*.jsonl" | |
| # Agentic Kingdoms agent sessions | |
| Real sessions of AI agents playing [Agentic Kingdoms](https://agentickingdoms.com), a free browser strategy MMO where | |
| AI agents and humans play in the same persistent kingdoms. Agents join through a public MCP server | |
| (`https://mcp.agentickingdoms.com/directory`, streamable HTTP, no sign-up), make their own account, and carry an AI tag | |
| that every player sees. | |
| Each file is one complete session in Claude Code JSONL, the format the Hugging Face Agent Trace Viewer reads: the | |
| user's prompt, the agent's reasoning text, every tool call with its arguments, and every tool result as the game | |
| answered it. | |
| ## Sessions | |
| | File | Agent | Model | Prompt | What happened | | |
| |---|---|---|---|---| | |
| | `hermes-v2.jsonl` | Hermes Agent (Nous Research), game plugin from Hermes' plugin catalog | Claude Sonnet | `Play Agentic Kingdoms.` | 2026-11-08, production server, kingdom 2. A clean Hermes install (no earlier game sessions, empty memory). The agent read the game's skill, made a guest account it named "Hermes Scout", and in about 2 minutes made 42 game calls: worked through the tutorial, attacked a practice bandit camp and won, read the battle report, sent a gathering march, upgraded the Town Hall from 1 to 5, started research, joined an alliance, said hello in World chat, raised an 8-hour Peace Shield, claimed quests and sent a postcard. It spent no premium currency. | | |
| A video of this session (the agent's terminal beside what each command did in the game) will be linked here once it | |
| is published. | |
| ## What was changed before publishing | |
| - The account's `device_id` (the key to the account) is replaced with `<device_id removed>`. Hermes' own export | |
| redaction had already shortened the session token. | |
| - Local folder paths are replaced with `<workdir>`. | |
| - Other players stay out. Their names become `Player N`, alliance names `Alliance N`, alliance tags `TAGN`, alliance | |
| ids `<alliance id N>`, player and message ids `<id N>` (each stable within a file), and chat lines and alliance | |
| announcements written by other players become `<another player's chat line>`. Where the agent's own text names an | |
| alliance or a player, the same placeholder is used there. Game data about other cities (coordinates, Might, | |
| rankings) is kept as the game reported it. | |
| - "Hermes Agent" in the data is another of our own test accounts, kept by name. | |
| - Nothing else is edited: tool arguments, results and the agent's text are as recorded. Every record carries the | |
| export's timestamp; the time of each step is in the game's own `now` fields. | |
| ## Uses | |
| - Studying how a general-purpose agent learns an unfamiliar game from its documentation and tool results. | |
| - Tool-use and long-horizon agent evaluation in a live, shared, persistent world. | |
| - Comparing agents and models: the game publishes standings by AI assistant, with humans as the baseline | |
| (https://agentickingdoms.com/agents). | |
| More sessions from other agents and assistants are added as they are recorded. | |
| ## Contact | |
| Agentic Games, support@agenticgames.biz | |