fix(wip): WIP - partial unify WeChat/H5 skill prompts (build agent changes)

- Reuse H5's buildAutoChatSkillPrefix in buildWechatAgentPrompt signature
- Inject autoSkillPrefix into voice/text message prompts
- Add origin tagging in GET /sessions for h5/wechat channel labels

Still TODO (will complete in next cycle):
- Full resolveGrantedSkills → buildWechatAgentPrompt flow
- ensureWechatAgentSession idle rotation logic
- setSessionOrigin + touchWechatAgentRoute calls
- Backend schema/migrations (h5_user_sessions origin column)
- Frontend type + UI component updates

Co-Authored-By: Claude <noreply@anthropic.com>

tkmind-proxy.mjs

@@ -1164,6 +1164,16 @@ export function createTkmindProxy({
 
           const sessions = [...sessionsById.values()].sort(sortSessionsByRecent);
           await enrichSessionHistory(sessions);
+          if (typeof userAuth.getSessionOrigins === 'function' && sessions.length > 0) {
+            try {
+              const origins = await userAuth.getSessionOrigins(sessions.map((s) => s.id));
+              for (const session of sessions) {
+                session.origin = origins.get(session.id) ?? 'h5';
+              }
+            } catch {
+              // leave sessions untagged on lookup failure
+            }
+          }
           const summaries = sessions.map(projectSessionSummary).filter((item) => matchesSessionQuery(item, query));
           const paged = summaries.slice(offset, offset + limit);
           res.json({