release: bump versionCode 5->6 for v0.4.4 + add changelog 6.txt

v0.4.3 already shipped as versionCode 5; a duplicate code would be rejected by Play and ignored by F-Droid. Bump to 6 unblocks the v0.4.4 release. QA gate passed (units + on-device E2E + visual render check screenshots in docs/qa/render-check/).

docs/marketing/x-thread.md

@@ -0,0 +1,83 @@
+# X (Twitter) thread — copy-paste ready
+
+> Post as a native thread (write tweet 1, then "add another tweet" for 2–8). Don't paste links in tweet 1 — X suppresses reach on link-first tweets. Put the main link in the last tweet and/or the first reply.
+> Best windows: 9–11am ET or 1–3pm ET on weekdays.
+> After posting, quote-tweet your own thread later that day linking the HN/PH post for a second impression.
+> `{{PLAY_URL}}` = `https://play.google.com/store/apps/details?id=cc.agentlabs.opencode` once approved.
+
+---
+
+**Tweet 1 (hook)**
+```
+I built an Android app that connects to your self-hosted AI coding agent.
+
+Token-by-token streaming. Inline diff viewer. Tool-call approval — from your phone.
+
+OpenCode Mobile. MIT, free, no backend of mine.
+
+🧵
+```
+
+**Tweet 2 (the problem)**
+```
+The problem: opencode runs a long coding session on my server. I leave my desk.
+
+SSH + a terminal loses the session UX. Generic AI chat apps have no concept of file diffs or tool-call approval.
+
+I wanted a real client, not a workaround.
+```
+
+**Tweet 3 (how it works)**
+```
+How it works:
+
+opencode exposes an HTTP + SSE API. The app speaks it — streaming output, session state, diffs, tool calls.
+
+Your code stays on your server. Your provider keys stay on your server. The app is a thin client. No proxy, no middleman.
+```
+
+**Tweet 4 (features)**
+```
+What you get:
+
+→ Watch the agent generate code token by token
+→ See every file diff before it's written
+→ Approve or reject shell commands + file writes
+→ Manage multiple sessions
+→ Connect via Tailscale, Cloudflare Tunnel, ngrok, or local network
+```
+
+**Tweet 5 (local-model angle)**
+```
+Works with any model opencode supports: Claude, GPT, Gemini — or local LLMs via any OpenAI-compatible endpoint (ollama, LM Studio, llama.cpp).
+
+Your keys. Your hardware. Your inference budget.
+```
+
+**Tweet 6 (OSS / trust)**
+```
+MIT licensed. No subscription. No ads.
+
+Sentry crash reporting is opt-in and OFF by default. Connection secrets live in the Android Keystore.
+
+Source: github.com/dzianisv/opencode-mobile
+```
+
+**Tweet 7 (CTA)**
+```
+On Android now:
+
+Install (F-Droid + direct APK): dzianisv.github.io/opencode-mobile/download
+Setup guide: dzianisv.github.io/opencode-mobile/guide
+
+Google Play going live shortly: {{PLAY_URL}}
+
+If you run opencode, give it a try. Issues + PRs welcome.
+```
+
+**Tweet 8 (optional, post as reply for social proof)**
+```
+Also on Hacker News today if you want the technical deep-dive (SSE polyfill, WebView-free diff renderer, coordinating two async streams for approval):
+
+[paste HN link here after you post Show HN]
+```