ADBT/AI What's New: Nov 9, 2026

[Matthew_Sabol]

Here’s the next update in our series on Amazon Devices Builder Tools for AI (ADBT for AI). This one covers releases 1.0.17 to 1.0.20, so you can see what’s changed at a glance and dig into the parts that matter for your development.

One partner held a 30-minute meeting to plan moving their Fire OS WebView app to Vega. Out of curiosity, someone started the ADBT migration process.

… before the meeting ended, their agent already had a working Vega app with several integrations.

  What’s new at a glimpse

• An accessibility audit for your existing VegaScript app, with each issue tied to its fix, available for Vega apps. (1.0.17*)
• Ask your agent “what’s new in ADBT?” and it answers from ADBT’s own release notes, available for Vega and Fire OS apps. (1.0.17*)
• Scalable UI Audit, now fully available: it catches sizes that break on 720p, 900p, or 1440p screens, now including WebView/HTML5 apps, available for Fire OS apps. (1.0.17*)
• First-class support for native (C++) Vega apps: Time To Fully Drawn, native-aware debugging, dropped-frame analysis, Vega SDK 2.0 API migration, and a staged Fire OS to Vega migration, available for Vega and Fire OS apps. (1.0.17*)
• Your agent notices when your installed ADBT files are out of date and offers to update them, available for Vega and Fire OS apps. (1.0.18*)
• Help picking between a Vega Web App and React Native for Vega, based on your team’s size and skills, available for Vega apps. (1.0.19*)
• React Native 0.83 guidance updated for general availability, available for Vega apps. (1.0.19*)
• Find native (C++) memory leaks with Heaptrack and see which thread uses your CPU with Flashlight, available for Vega apps. (1.0.20*)

* Listed versions are where features became available.

  New here? Quick recap on ADBT for AI

Amazon Devices Builder Tools for AI (ADBT for AI) is a suite of AI-powered tools for Fire TV development both on Vega and Fire OS: the Model Context Protocol (MCP) server plus a set of agent skills. It makes coding agents such as Kiro, Claude, and Codex aware of Fire TV-specific knowledge and procedures, so they stop guessing at the parts of Vega and Fire OS development that don’t exist in a general model’s training data.

This matters because device development is device-specific, multi-step, and environment-heavy. ADBT supplies the Amazon Devices knowledge layer as purpose-built tools and skills grounded in current, versioned Amazon documentation, so your agent follows an Amazon procedure instead of reconstructing one from a training snapshot.

Where applicable, ADBT runs steps as scripts instead of instructions; this ensures deterministic, accurate results while actively reducing token spend.

Start using ADBT now: https://developer.amazon.com/docs/adbt/get-started

  A closer look at what’s new

• Setup now saves into your project: init-context asks for your project’s root directory and saves the project guidance and .adbt-config.json there, instead of wherever you started the command. Your home directory, a folder that contains it, and the filesystem root can’t be the project root; projects inside your home directory work as usual. With --force, start it from your project’s root directory.
  ⚠️ Important: If an earlier version saved the guidance in your home directory (for example ~/AGENTS.md or ~/CLAUDE.md), ADBT no longer updates it there. Remove its Amazon Devices Platform Context section, then run init-context from your project.

• Audit your VegaScript app for accessibility: Beyond the accessibility knowledge base, your agent can now audit an existing VegaScript app and tell you whether it is accessible and ready, with each issue tied to the guidance that explains the fix.

  

    Ask

  “Review my VegaScript app for accessibility issues.”

  “Audit my app against the accessibility guidelines.”

• Ask your agent what’s new: ADBT’s own release notes are now part of what your agent can read, so you can ask it directly instead of opening the npm package page.

  

    Ask

  “What’s new in Amazon Devices Builder Tools?”

  “What changed in the latest ADBT release?”

• Scalable UI Audit, now fully available: For Fire OS apps, this skill reviews your UI for sizing that does not scale cleanly across 720p, 900p, and 1440p screens, flags the responsible code, and explains the fix. New in 1.0.19, each finding shows its impact, the screen sizes it affects, whether the code is live or behind a named flag, and how confident the finding is, with the most severe first. WebView/HTML5 and Adobe AIR apps are now audited too, and you can tell the audit which screens are not shipping so their findings are listed separately.

  

    Ask

  “Audit my Fire OS app for Scalable UI readiness.”

  “Will my app’s layout scale to a 1440p screen?”

  “Audit my WebView app for Scalable UI issues, and skip the screens we are not shipping.”

• Native (C++) Vega apps get dedicated support: If your Vega app is native rather than React Native, ADBT now works with it directly. A guide shows how to emit the fully_drawn marker so the KPI Visualizer can measure cold start instead of reporting Time To Fully Drawn as null. Debugging routes to native tools instead of React Native DevTools, and dropped frames are found in your Perfetto trace and pointed at the native code responsible. If your build fails because a Vega SDK 2.0 API was removed, your agent finds every affected call and walks you through the change. And the Fire OS to Vega migration for native apps now lets you build one feature at a time (for example Browse first, then Playback and Settings).

  

    Ask

  “Why is my app’s TTFD showing as null?”

  “My Vega SDK 2.0 build is failing after an API was removed, help me fix it.”

  “Migrate my Fire OS native app to Vega, starting with just the Browse screen.”

• Find native memory leaks and see per-thread CPU: Two new guides help you measure a running Vega app. The Heaptrack guide, for native (C++) apps, shows how to find native heap memory leaks, and names what Heaptrack can’t see, such as graphics and video buffers and the JavaScript heap. Memory still held when the app stops reads as leaked, so a reported amount must grow with runtime before you call it a leak. The Flashlight guide covers the open-source performance monitor: per-thread CPU, total app CPU, and memory over time, plus when to use it instead of Activity Monitor, Memory Monitor, or Vtrace.

  

    Ask

  “Help me find native memory leaks in my Vega app with Heaptrack.”

  “Which thread in my Vega app is using the most CPU?”

  “When should I use Flashlight instead of Activity Monitor?”

• Keep your ADBT setup current: init-context --update refreshes only what you already installed, such as your steering document, skills, and the adbt-dev agent file, with no prompts and nothing new added. Your agent also checks once per session whether your installed files are out of date and offers to run it for you.

  

    Ask

  “Update my ADBT setup to the latest version.”

  “Check if my ADBT installation needs updating.”

• Vega Web App or React Native for Vega?: You can wrap your website as a Vega Web App, or build in React Native for Vega. Your agent now helps your team pick between the two, based on your team’s size and skills.

  

    Ask

  “Should we build our Vega app as a Vega Web App or in React Native for Vega?”

• React Native 0.83, now generally available: The RN 0.83 guidance reflects general availability. If you joined the preview, cleanup steps for the preview npm registry and SDK configuration come before the standard upgrade. The upgrade now matches the current Vega app template (@amazon-devices/react-native-kepler 4.0.x, Node.js 22.14.0 or later, React Native Community CLI 20.0.0), and if your manifest.toml has no [os.version] table, your agent asks for your lowest supported OS version instead of guessing.

  

    Ask

  “Upgrade my Vega app from React Native 0.72 to 0.83.”

  “I was in the React Native 0.83 preview, help me move to the general release.”

• Move off legacy In-App Purchasing: If your Fire OS app still uses the legacy In-App Purchasing v2.0 library or an older Amazon Appstore SDK 3.x, your agent now migrates it onto the current Appstore SDK 3.x.

  

    Ask

  “Help me migrate my app off the legacy In-App Purchasing library.”

• Improved Robustness: Media playback guidance now names the device you need for DRM testing (a physical Fire TV Stick for Widevine and PlayReady L1/L2; the Vega Virtual Device supports L3 only, on Mac M-series hosts). The Matter Casting workflow can resume from a requested step, and if the ADBT MCP server is disconnected during a Vega port, your agent stops and tells you how to reconnect. ANR crash analysis now also handles crash reports with no ANR reason or an unknown one. Plus clearer tool descriptions and error messages, one place for React Native package versions, better detection of unsupported dependencies when upgrading library projects, and a corrected read_document example in the Fire OS get-started guide.

---

Full release history RELEASE_NOTES.md, in the npm package.