## What

<!-- 1-2 sentences. What does this PR change? -->

## Why

<!-- Link the issue, design doc, or chat thread. Without a "why" the reviewer
has to guess whether the change is wanted at all. -->

Refs: #<!-- issue or N/A -->

## How to test

```bash
# Commands the reviewer can run to reproduce / verify:
docker run --rm -v "$PWD":/work -v "$HOME/.m2":/root/.m2 -w /work \
  maven:3.9-eclipse-temurin-25 mvn -B test
```

<!-- Manual test steps (if any): load a save, join a server, trigger a
death in the void, etc. State what you observed. -->

<!-- Server console output if relevant (translatable components render
in English there — that's normal). -->

## Risk

<!-- Which surface is touched? chat-only / item-loss / chunk-load / permissions.
Is a rollback plan needed? Does it need a config.yml default? -->

- [ ] chat message flow (immutable after send)
- [ ] pt-BR singular-form rule (`5.966 blocos de Pedra`, not `5.966 Pedras`)
- [ ] Bedrock equivalent (`comando:` typed fallback, `texto:` ASCII label)
- [ ] stats path (`<world>/players/stats`, not `<world>/stats`)
- [ ] reaction late-window (`reacao-validade-minutos`)
- [ ] Bukkit main thread (no async chunk loads in event handlers)
- [ ] new `comando:` added to `plugin.yml`
- [ ] new permission declared in `plugin.yml`

## Checklist

- [ ] `mvn -B test` passes locally (JDK 25)
- [ ] new behaviour has a test (`src/test/java/...`)
- [ ] no hand edits to generated sources (none in this repo)
- [ ] PR title is conventional-commit style
- [ ] branch is up to date with `main`

## AI review

Label this PR **`AI-REVIEW`** to trigger `pragent`. Add **`AI-USAGE`** for a
token/cost report on the review. The bot reads `.pr-review.json` from `main`
+ a static `architecture.md` from Nexus raw-hosted — so house rules (chat
immutability, singular-form, Bedrock fallback, JDK 25) are baked in.