修复: 走查发现(空 id 跨轮全局唯一 + generate_image SSRF/OOM + 技能日志 + fallback 测试)

- 空 tool_call id fallback 改全局 AtomicU64 计数器(跨轮跨 assistant 唯一,治空气泡+
  L1 误熔断+三元组错位,实证 af2fab4e);fallback 测试硬编码改 starts_with(解耦计数器)
- generate_image provider POST 补 SSRF 防护(复用 http.rs validate_url/resolve_and_check_host,
  治 base_url 用户可配打内网)+ b64_json 解码前估算长度防 OOM
- SkillResolver 加 tracing 日志(定位技能发送后未注入:name 匹配/read 失败)

src-tauri/src/commands/ai/augmentation/resolvers.rs

@@ -306,10 +306,21 @@ impl MentionResolver for SkillResolver {
         // 取剥 frontmatter 正文(注入用);None = 缓存未命中或文件读失败
         let body = read_skill_content_stripped(name.clone())
             .await
-            .ok_or_else(|| ResolveError::NotFound {
-                kind: "skill".to_string(),
-                ref_id: name.clone(),
+            .ok_or_else(|| {
+                tracing::warn!(
+                    skill = %name,
+                    "[SkillResolver] 未命中缓存/读失败 → aug 空,skill 未注入(用户感知:发送后未真使用)"
+                );
+                ResolveError::NotFound {
+                    kind: "skill".to_string(),
+                    ref_id: name.clone(),
+                }
             })?;
+        tracing::info!(
+            skill = %name,
+            body_len = body.len(),
+            "[SkillResolver] resolve 成功,正文将注入 system prompt"
+        );
         // source 从缓存取 SkillInfo.source;失败(缓存不一致)默认 "skill"
         let source = crate::commands::ai::skills::skills_cached()
             .await