- title.txt: 'OpenCode Mobile' -> 'OpenCode Mobile: AI Coding' (26/30 chars,
adds the 'AI coding' search keyword to the strongest Play ranking field)
- short_description: front-load 'AI coding agent for Android' (71/80)
- full_description: dead privacy URL opencode.vibebrowser.app (000) ->
live dzianisv.github.io/opencode-mobile/privacy/ (200) — was a Play
rejection risk
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Visual proof above the fold-ish: 3 UI previews with captions + descriptive
alt text. Responsive flex strip, lazy-loaded. Conversion lever for first-time
visitors — landing had QR codes but no app visuals.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
3 long-tail pages targeting 'opencode vs chatgpt/copilot', 'vs termux',
'claude code on android'. Added to sitemap, footer nav, in-content links
from landing + guide + troubleshooting. All have canonical + install CTAs.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Targets "opencode mobile features", "AI coding android", and
"opencode vs ChatGPT for coding" queries not covered by existing pages.
Includes SoftwareApplication + BreadcrumbList structured data.
Adds /features/ to sitemap and internal footer nav on all 5 pages.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The upstream opencode project asks 'opencode-*' named projects to clarify they are
not built by or affiliated with the opencode team. Adds that note near the top.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Authentic troubleshooting content mapping the app's six connection-failure
diagnostics (malformed-url, no-internet, tls, health-failed, timeout,
server-unreachable) to concrete fixes. Targets real problem queries
('opencode mobile can't connect', 'opencode tailscale setup', 'network request
failed'). FAQPage + BreadcrumbList JSON-LD for rich results. Linked from landing,
guide, and download footers; added to sitemap.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
build.yml only built the APK — the 65 unit tests and typecheck never ran in CI,
so regressions in the covered logic (headers/SSE/diagnostics/settings/etc.) could
land silently. Add a fast 'test' job (Node 24, native TS test-running) so every
push and PR enforces typecheck + npm test before merge.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Extract TOOL_STATUS + statusFromPart from stores/events.ts into a pure
status-labels.ts (type-only Part import, erased at runtime; events.ts delegates).
8 tests pin the live 'what is the agent doing' labels: reasoning/text/known-tool
mappings, shared labels (search/edit groups), unknown-tool degrade to
'Running <tool>...', and the no-tool/unknown-type fallthroughs. typecheck clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Extract parseUrl + classify from diagnostics.ts into a pure diagnostics-classify.ts
(diagnostics.ts imports + re-exports the types; behavior unchanged) so the
connection-failure decision tree — the user-facing 'why did connecting fail'
guidance — is unit-testable without react-native/expo. 15 tests: URL parsing
(port defaults, IPv4-vs-hostname, path/query stripping, malformed rejection) and
every classify branch (malformed, ok, tls, no-internet, health-failed, timeout,
server-unreachable incl. the hostname-only MagicDNS hint). typecheck clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Extract two pure helpers so they're testable without zustand/expo:
- settings-merge.ts: clampPageSize + forward-compatible mergeStoredSettings (the
upgrade path where stored data predates a new notification category must yield
the default, not undefined). stores/settings.ts now delegates.
- notify-format.ts: sanitizeBody (strip C0/DEL control chars, trim, cap at 200)
used for server-supplied notification text. stores/events.ts now imports it.
Behavior unchanged; typecheck clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Extract the SSE chunk-buffering from sdk.ts into a pure SSEParser (sdk.ts delegates;
behavior unchanged) and pin the framing rules that are easy to break when an event
splits across network reads: partial trailing lines held until completed, frames
reassembled across 2-3 reads, [DONE] sentinel and empty/non-data lines filtered,
each frame emitted exactly once. typecheck clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Extract the auth + directory-header encoding from sdk.ts into a pure, dep-free
headers.ts (sdk.ts now delegates — behavior unchanged) so the connection-critical
logic is unit-testable without expo/fetch. Add 8 header tests (ASCII passthrough,
non-ASCII/CJK percent-encoding stays header-safe, Basic auth, empty-dir handling)
and 8 logbuffer tests (serialization incl. circular-ref fallback, 200-entry ring
cap, copy semantics, formatting). typecheck clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Landing/guide/download descriptions were 257/246/217 chars and truncated in
Google results. Tightened to ~150 while keeping target keywords.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Targets 'download opencode android', 'opencode apk', 'opencode f-droid'.
Serves the live install channels (self-hosted F-Droid repo + signed APK)
with SoftwareApplication + BreadcrumbList structured data. Linked from
landing hero, landing footer, and guide footer.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Easier to maintain in this TS repo: typed gate definitions, native fetch/JSON
(drops the fragile jq/python3/grep branches), runs via `node check.ts` with no build/deps.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Set expo.autolinking.buildFromSource: [".*"] so F-Droid mainline's
scanner doesn't see the prebuilt .aar + -sources.jar files that Expo
ships inside every expo-* npm package's local-maven-repo/ directory.
When this flag is set, expo-modules-autolinking routes each module
through a Gradle source build instead of consuming the prebuilt AAR.
Verified: `./gradlew assembleRelease` succeeds (11 min, 92 MB APK)
with this flag and no prebuilt AARs consumed.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Addresses linsui review: summary/description pulled from fastlane; honest, policy-compliant copy.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Privacy URL was inconsistent across owner docs (agentlabs.cc/opencode/privacy
in 3 spots vs opencode.vibebrowser.app/privacy in the majority). Reconciled
all to the canonical opencode.vibebrowser.app/privacy (Search-Console-verified
domain; what privacy-policy.html self-references). Also fixed stale v0.4.2 AAB
reference in PLAY-APP-CONTENT-ANSWERS.md.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
onCreateInDirectory navigated with dir.trim() (raw user input) while the
sibling create paths use the server-authoritative session.directory. When
the opencode server normalizes the path (trailing slash/symlink/relative),
the nav param diverged from the session's real scope, causing the new
session/[id] screen to scope to the wrong client -> 'Failed to load
session'. Mirror the sibling paths' session.directory spread.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
androguard 4.1.4 raises 'NoOverwriteDict object has no attribute append' in
parse_v2_v3_signature when fdroidserver extracts the signer cert — this broke the
self-hosted F-Droid publish from v0.4.2 on. 4.1.3 (which shipped v0.3.2–v0.4.1)
parses our re-signed v1+v2-only APK cleanly; verified locally with
fdroidserver.common.get_first_signer_certificate.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The self-hosted F-Droid repo (https://dzianisv.github.io/opencode-mobile/fdroid/repo)
has been stuck at v0.4.1 because publish-fdroid crashed in androguard parsing the
CI APK's v2+v3 signature block pair ('NoOverwriteDict' object has no attribute
'append'). Force v1+v2-only signing: gradle flags for local builds, plus a
deterministic apksigner re-sign step in the workflow (expo prebuild regenerates
build.gradle, so the workflow step is the real guarantee). Bump to v0.4.3 /
versionCode 5 so a fresh tag re-runs the publish with the verified bug fixes
(#10 scope fixes, send-error fix) included.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
sendMessage captured currentSession at call time; if the user switched sessions
while a prompt was in flight and it failed, the catch handler surfaced the error
on and refetched the NOW-current session. Clear the sending flag for the session
we actually sent to, but only set error / refreshMessages when it's still on
screen. Found via runtime bug audit.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
