docs: 纳入监督/授权文档与现行 opencode 配置

overnight-quality-brief/overnight-status/overnight-completion-brief/skill-sourcing-authorization 四份监督与授权文档入库；opencode.json 记录现行 go-a/glm-5.3-flash 默认通道；.gitignore 增补 opencode.json.before-luna 备份与 scripts/m3_state.json.corrupt_* 损坏状态文件。

1 files changed, 11 insertions, 0 deletions

diff --git a/docs/skill-sourcing-authorization.md b/docs/skill-sourcing-authorization.md
new file mode 100644
index 0000000..4b9ee8b
--- /dev/null
+++ b/docs/skill-sourcing-authorization.md
@@ -0,0 +1,11 @@
+# External skill sourcing authorization
+
+User explicitly authorizes finding suitable skills on GitHub and giving them to OpenCode to improve Living Village development. This supplements ongoing full-game completion authorization; do not turn skill collection into another blocker.
+
+Candidate found via gh search and README inspected: https://github.com/alca259/monogame-skills (MIT reported by GitHub API). Potentially relevant: monogame-game-loop, monogame-input, monogame-2d, monogame-ui-layout, monogame-ui-interaction, monogame-platform. Only README/license metadata inspected so far; contents NOT audited or installed.
+
+Prefer existing installed debugging/test/verification skills to duplicates. Use gh CLI for GitHub. Audit exact skill text and sibling templates/scripts, license and pinned commit before using or installing. No blind npx installer, no executing downloaded scripts before review, no credential requests or unrelated changes. Verify OpenCode's documented discovery roots and collisions before project-local installation; preserve complete required package, validate tests and discovery. Missing trustworthy tests means reference-only pending validation, not claimed installed.
+
+The candidate focuses on C# and zero-allocation/OOP or ECS patterns. Our architecture remains F# + MonoGame with deterministic pure simulation and immutable World. Borrow applicable rendering/input/UI guidance; do not rewrite into C#, replace simulation with ECS, build an unnecessary custom UI framework, or compromise deterministic semantics based on generic skill advice. Verify every API against installed versions and compile/run tests.
+
+Record selected skill source+commit, audit result, installed path/discovery or reference-only state, and which concrete pending feature it helped. Avoid downloading an entire catalogue merely to increase skill count.