# Android Tv Apps Helper > Use when a user wants to connect an Android TV over ADB, install or inspect TV APKs, simplify the launcher, or diagnose app launch, picture, sound, or remote-control problems. - Skill: `swainwong/android-tv-apps-helper` (Agent Skill, multi-file: 9 files) - Install (CLI): `npx skillmds@latest add swainwong/android-tv-apps-helper` - Raw SKILL.md: https://api.skillmd.com/api/skills/swainwong/android-tv-apps-helper/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: SwainWong (https://skillmd.com/u/swainwong) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/swainwong/android-tv-apps-helper --- # Android TV Apps Helper The harness owns questions, transitions, validation and approvals; supply verified observations. ## Start every invocation Read [references/platforms.md](references/platforms.md), [references/interaction-contract.md](references/interaction-contract.md), and [references/workflow.md](references/workflow.md). Require a local-computer context with local shell and LAN access. Cloud-only execution stops before ADB and shows one bounded switch/retry/exit question. Follow platforms.md capability detection. In Codex, prefer available `request_user_input_async` using `--native-tool request_user_input_async`. Initialize: ```sh python3 ../../scripts/tv-helper init-session /session.json --surface auto --host-platform --execution-context local_computer ``` Run entry immediately for Release checks, 24-hour update snooze and passive precheck. Validate replacement identity/version/SHA-256 before updating. Precheck locates ADB and runs `adb version` and `adb devices -l`; it cannot scan or connect. ```sh python3 ../../scripts/tv-helper workflow-entry /session.json --installed-version 0.3.4 ``` For `native_required`, display `context_markdown`, then call `tool_name` with exact `tool_input`. Keep context outside the short card. Every decision requires native choices; never ask for numbered-text answers. Record observed tool failures: For `response_delivery=async_user_message`, preserve question/presentation IDs and yield until the actual user reply. Tool acknowledgement or preselection is not an answer. ```sh python3 ../../scripts/tv-helper record-surface-failure /session.json --question-id --presentation-id --tool-name --reason --detail ``` Retry the returned native presentation once. For `native_blocked`, display its table and pause; resume only after host recovery using `resume-native`. Never use a text menu. Preserve options, explicit consent and safe exit. ## Continue a question Submit the exact visible answer through: ```sh python3 ../../scripts/tv-helper workflow-native-answer /session.json --question-id --presentation-id --value ``` Use `--values-json` for actual native multi-select. Pagination/toggles stay in the current question. Input starts with a native fill-information choice; only `accepts_free_input=true` permits the supplemental input field after the options. On exit code 2, redisplay context and the returned native card, not a text menu. When the result contains `action_required`, execute only that approved action. Record real stdout/stderr, exit code, inspected identity/hash, or read-only check in an evidence JSON object, then submit it before displaying any completion state: ```sh python3 ../../scripts/tv-helper workflow-action-result /session.json --action-id --status --evidence /action-evidence.json ``` Never move past a pending action by writing the session or merely saying it succeeded. For `PASSIVE-DISCOVERY-ACTION`, run `workflow-discover `: it executes and records fresh results together. Never reuse the entry precheck or manually reconstruct discovery evidence. Precheck is read-only. Never download or install ADB, Platform-Tools, an APK, or dependencies without `action_required` after an explicit accepted answer. `configure_adb`, recommendations, and model inference are not approval. For `PREPARE-INSTALL-PLAN-ACTION`, run `prepare-install-plan`; return its plan as evidence. Never hand-author one. ```sh python3 ../../scripts/tv-helper prepare-install-plan /session.json --out /install-plan.json ``` ## Operations Read [references/adb-operations.md](references/adb-operations.md) before device commands and [references/apk-policy.md](references/apk-policy.md) before downloads. Use `adb -s `. Downloads, installs, Home and wallpaper require separate confirmations. Never root, flash, reset, clear data, disable the factory launcher, or use unknown mirrors. Use `../../catalog/apps.json`. All apps stay selectable; source-review status never blocks downloading. For missing URLs, resolve after confirmation per apk-policy.md. Use fixed task labels and the application table. Use native multi-select or native toggle pages; confirm names and versions before downloading. Download is not install approval. Dangbei uses only publisher-linked sources. After reading target identity, run `prepare-device-context` with the packaged guide and compatibility data. After verifying Emotn UI, ask separately about Home and wallpaper. Custom wallpaper needs a valid image and apply confirmation; always show identity, independent risks, and possible vendor reset. ## Evidence and finish Every interactive and final reply is table-first. Status symbols come from harness evidence: ✅ completed, ❌ attempted but incomplete, ⚠️ attention or user verification, ⏳ pending, ⏭️ skipped. Keep file validation, installation, launch, runtime signal, and on-site picture/sound/remote acceptance separate. All exits after ADB use route through `FINISH-SAFETY-Q1`. Match the device guide and ask the user to close ADB/network/wireless debugging and the developer-options master switch. Disconnection alone is not proof. If ADB still responds after the user says it is closed, keep the same question and show the conflict. After reaching a terminal state, render and save the harness final report: ```sh python3 ../../scripts/tv-helper final-report /session.json --out /final-report.md ```