Run game commands

Use this Skill when the player explicitly asks OpenAllay to execute a Minecraft command, or when the task must discover the exact syntax of an installed mod's command before executing it.

The commands object exists only because the player enabled the experimental command capability for this request:

• commands.list() returns the complete Brigadier tree visible to the current player. It contains vanilla, server, loader, and mod-registered nodes.
• commands.describe(path) returns one exact literal/argument path from that detached tree.
• commands.run(command) submits the exact command as the requesting player, waits off the render thread for the associated client-visible feedback window, and returns the messages that Minecraft produced. A single optional leading / is removed; the remaining string is unchanged.

For a command-only run_javascript call, set the Tool input roots to ["commands"]. The binding is named commands directly; there is no mc.commands.

If the player supplied an exact command, or the required vanilla command and syntax are already unambiguous, this main document is enough: call commands.run directly. Do not list the whole command tree and do not load the reference merely to confirm a known command.

Load references/commands.md before writing the program only when the syntax is unknown, the command contains a modern item/text component, or the task needs several ordered commands. For unknown syntax, filter commands.list().nodes inside the same program by the relevant literal/mod prefix, use commands.describe(path) on the exact candidate, and return only the focused candidates; never return the unfiltered catalog.

Always inspect and return the commands.run result. state: "feedback" means Minecraft emitted one or more messages in the command feedback window; state: "no_feedback" means the command produced no observable message before that window closed. Do not describe a command as successful merely because it was submitted.

Minecraft remains authoritative for parsing and permissions. Submissions are not transactional: if a later statement fails or the Agent is cancelled, commands already submitted stay submitted and are never rolled back.

Treat parser errors and permission errors in messages as rejection, not success. Correct a parser error at most once after inspecting its actual message and the matching described path; never make blind variants. If an exact player-visible path cannot be found, report that it is unavailable instead of trying similarly named mutations.