AI game master
Rewards a quest once.
Step 1 of 8
The agent you build
Quest q-1Your agentEconomygrant · moves gold
Your agent reads a finished quest and rewards the player from the quest table in a small game economy. It looks at the completion's rewards before it rewards, and holds large rewards for a game master.
Read the whole guide as text
- The agent you build. Your agent reads a finished quest and rewards the player from the quest table in a small game economy. It looks at the completion's rewards before it rewards, and holds large rewards for a game master.
- Your assistant builds it. Your own coding assistant builds the agent; Geometer does not build agents. With Geometer added as a tool, the assistant checks it right after.
- Watch it run. The agent as written reads the quest, looks at its rewards, then rewards once.
- Break it. The mistake: the agent no longer looks at the completion's rewards before it rewards. Geometer's result: Broke, rule one-reward-per-completion.
- The one-line fix. Put the original line back. That is the agent as written.
- Replay it, keep the check. The agent as written: no problems found within these limits (rules checked: one-reward-per-completion). Keep the check, and CI runs it on every change.
- Show others what was checked. Issue a report of what was tried and what held. It is opt-in and self-issued, and anyone can check it is unchanged with the open verifier.
- Take it. The agent is the folder game-master in the public agents repository.
Check your own agent like this
Read more
Recorded runs of the agent as written and with its one-line mistake; the result line is from the agent’s README. What a real check covers is in the alpha limits.