tui: parallelize skill downloads for faster loading

Refactored skill discovery to download skills in parallel instead of sequentially,

reducing load times when multiple skills need to be fetched from remote URLs.

Also updated AGENTS.md with guidance on using dev branch for diffs.

packages/opencode/src/skill/skill.ts

@@ -153,9 +153,9 @@ export namespace Skill {
     }
 
     // Download and load skills from URLs
-    for (const skillUrl of config.skills?.urls ?? []) {
-      const downloadedDirs = await Discovery.pull(skillUrl)
-      for (const dir of downloadedDirs) {
+    for (const url of config.skills?.urls ?? []) {
+      const list = await Discovery.pull(url)
+      for (const dir of list) {
         dirs.add(dir)
         for await (const match of SKILL_GLOB.scan({
           cwd: dir,