docs(libgbm): reading guide — the doc records three rounds, last one wins

The plan now supersedes itself twice (section 2's C2 by section 8, section 8's
B3 by section 12.1), and a reader going top-down would act on retired advice.
A short guide up front says which sections are authoritative and which are
kept only as a record of reasoning that was overturned — worth keeping,
because what overturned each round was a measurement, and those measurements
are the durable part.

Also states the honest pattern: the only thing that kept changing was the
mcpp-side shape, and that is exactly the part I did not read the source or
run an experiment on before proposing.

Author: Sunrisepeak

.agents/docs/2026-08-30-gbm-cross-repo-closed-loop-plan.md

@@ -2,6 +2,22 @@
 
 Date: 2026-08-30 · 起因:`compat.libgbm`(mcpp-index PR #281)· 状态:待 review
 
+## 怎么读这份文档
+
+它记录了**三轮修正**,后面的推翻前面的。**结论以最后一轮为准**,前面保留是为了记录
+被否掉的推理——因为每一次被否,否掉它的都是一次实测,而那些实测本身是资产。
+
+| 想知道 | 看哪节 | 注意 |
+|---|---|---|
+| **最终结论与交付** | §12(实现结果)、§13(状态)、§14(要不要做 mesa 包) | 权威 |
+| 为什么 `compat.libgbm` 该独立存在 | §3、§10.1 | 理由换过三次,结论没变 |
+| 任务拆分与依赖 | §11 | |
+| mcpp 侧该怎么修 | **§12.1** | ⚠ §2 的 C2(B1/B2)与 §8 的 B3 **都已作废** |
+| 实测证据 | §1、§12.3、§12.4 | |
+
+**唯一反复改的是 mcpp 侧的形态**(B1 → B3 → 「装到全局 scope 就够了」),因为那是唯一
+一处我没有一开始就去读源码 / 做实验的部分。理由被换掉三次而结论不变的那几条,才是稳的。
+
 ## TL;DR
 
 让 `gbm_create_device()` 在 mcpp 工程里能用,**不需要新机制**。所需的机制三层全都已经存在