Advanced guides¶
Already connected? Choose a guide for your next task. For a first installation, start with Install and first run.
Agent workflows¶
| Goal | Guide |
|---|---|
| Teach an agent to discover tools and resolve players | Install the portable Toolkit Skill |
| Understand live player context and chat-time bundles | Player identity and context |
| Trigger an agent from game events | Unattended Hermes: webhook + Skill |
| Understand Hermes integration and the legacy loop | Hermes integration status |
The signed event sender and Skill are available. Signed end-to-end Hermes
delivery still depends on mc-agent-bridge#7.
Treat the unattended guide as integration work, rather than the default first-run
path. The optional legacy mc-agent-loop is described in the Hermes guide.
Worlds and experiments¶
| Goal | Guide |
|---|---|
| Fork and restore a world | World-fork protocol |
| Check a restored fork | Fork verification |
| Provision an isolated server | Headless lab servers |
| Build observation code for a lab | Building a mod |
Integration and evidence¶
Verified example: Minecart ROM¶
Issue #13 and its child issues
14–#21 are complete. All three stages were run, independently reviewed and¶
merged on 2026-09-26:
| Stage | Accepted result | Evidence and reproduction |
|---|---|---|
| Toolchain readiness | Live gate 8/8 PASS | Acceptance runbook |
| Autonomous agent cold start | Agent-written logger, 5 operated/captured carts, 5/5 audit PASS | Original cold-start evidence |
| Frozen real-game regression | 3 fresh instances, 2 programs, 1 cold download, 8 negatives | Regression guide · Committed evidence |
The cold start records the original autonomous model run. The regression reuses its successful recipe without a model and is evidence of repeatability. The runbook separates offline verification, the recorded demo and fresh live-game runs. Optional Hermes webhook compatibility remains a separate follow-up.
Technical runbooks¶
These are developer and experiment runbooks; they are not required to install the Toolkit or connect an agent.
- Auditable cold-start runs
- Read-only minecart audit mod
- Stage-one integration gate
- Stage-one integration run
- Minecart ROM acceptance runbook
- Minecart ROM regression
Design and contributing¶
- Architecture and terminology — current component boundaries.
- RFC 0001 — historical interface and daemon design.
- RFC 0002 — deferred programmable tool calls.
- Documentation development — preview and check both languages.