Google on November 18, 2026 added an entry to the Google Analytics release notes stating that app developers who use the Google Analytics for Firebase SDK "can now use Firebase agent skills to empower your coding agents to make changes to your behavioral and conversion tracking for your iOS and Android apps." The capability behind that sentence is narrower than it reads. According to the same entry, "Initial integration and screen view logging are supported, with support for additional development tasks coming soon."

The entry links to a Firebase blog post published eight days earlier, on November 10, by Firebase product managers Don Kahn and Charlotte Liang. That post introduced "agent skills for Google Analytics for Firebase". Agent skills are packaged instructions that AI coding tools such as Antigravity, Claude Code, Codex and Cursor can load on demand; these ones install the Analytics SDK in a mobile app and add screen view tracking to apps built with declarative interface frameworks. Firebase's agent skills documentation, last updated on November 18, lists the addition as a single skill, firebase-analytics, one of 13 in the package.

In Short

Google has given AI coding assistants a set of instructions that lets them add Google Analytics to iPhone and Android apps and set up tracking of which screens people view. This matters to marketers because app screen and engagement data feeds Google Analytics reports and, through linked accounts, Google Ads, and more of that tracking code will now be written by software agents rather than typed by developers. For now the skill covers only the first installation and screen view logging; Google says further tasks are "coming soon" but gives no date, and conversion tracking, which its release note mentions, is not yet among them.

One skill, two jobs

The Firebase post frames the release around two problems. The skills, according to Kahn and Liang, "help you with two critical challenges for analytics: getting started and adding screenview instrumentation." Screen views are the app counterpart of page views on a website, and the post explains why they matter in plain terms: "Logging information about screenviews helps you understand user engagement or user behavior per screen of your mobile app."

In practice, a skill is a folder of instructions, optionally with scripts and reference files, organised around a SKILL.md file that the agent reads only when a request calls for it. Firebase's repository on GitHub describes its own skills as packaged instructions and scripts that follow the Agent Skills format. The approach differs from a chatbot inside a product interface: the skill shapes the code an agent writes in a developer's project, rather than answering questions about data already collected.

The post speaks of skills in the plural. Firebase's documentation, however, lists one Analytics entry. Its table describes firebase-analytics as a skill that assists with getting started and with measuring screen views for Google Analytics in iOS and Android apps. The other 12 skills cover the rest of the Firebase platform.

Skill name in Firebase documentation Firebase product area
firebase-ai-logic-basics Firebase AI Logic
firebase-analytics Google Analytics for Firebase (new)
firebase-app-hosting-basics App Hosting
firebase-auth-basics Authentication
firebase-basics General Firebase setup
firebase-crashlytics Crashlytics
firebase-local-env-setup Local environment setup
firebase-data-connect-basics Data Connect
firebase-firestore-standard Cloud Firestore, Standard edition
firebase-firestore-enterprise-native-mode Cloud Firestore, Enterprise edition in native mode
firebase-hosting-basics Hosting
firestore-security-rules-auditor Firestore security rules
firebase-remote-config-basics Remote Config

Source: Firebase agent skills documentation, last updated November 18, 2026. Product area labels are PPC Land's reading of the skill names.

What the release note says, and what the blog post says

The two Google documents describe the same release in different terms. The Analytics release note, which appears on the "What's new in Google Analytics" help page, uses the abbreviation GA4F for the Firebase SDK and talks about "behavioral and conversion tracking". The word conversion does not appear in the Firebase post at all.

Google Analytics release note Firebase blog post
Date November 18, 2026 November 10, 2026
Byline None Don Kahn and Charlotte Liang, Product Managers
Stated purpose "make changes to your behavioral and conversion tracking" "getting started and adding screenview instrumentation"
What works now "Initial integration and screen view logging" Setup with "a single prompt" and screen view logging for declarative UI apps
Platforms iOS and Android iOS and Android; Jetpack Compose, SwiftUI, Flutter and React Native named as reasons for manual logging
What comes next "support for additional development tasks coming soon" "As we build support for more development workflows, we'd love to hear your feedback"
Feedback channel Not stated GitHub discussion and Firebase UserVoice

The November 18 entry is the twentieth dated release on the 2026 version of the help page, and the third in November. It follows a diagnostic for user-provided data issues on November 11 and the beta of Data Manager in Google Analytics on November 6.

How setup works, according to Google

The Firebase post describes the setup skill as a flow that starts from one sentence typed into a coding agent. According to the post, it runs in six steps.

  1. The developer asks the agent: "Help me set up my app to use Google Analytics for Firebase."
  2. The agent "will connect your app to Firebase (if it's not already)".
  3. The agent will "set up Google Analytics in your Firebase project."
  4. The agent can then "add all the required dependencies and initialization code into your actual app."
  5. Where special setup is needed, the post gives the example "like creating a Google Analytics account", and says the agent "will walk you through the details" rather than completing the step itself.
  6. The skill accounts for platform rules that the post calls "the finicky bits", such as "remembering to include the -ObjC linker flag for your iOS app, or ensuring you're declaring the SDK correctly in your Gradle file for your Android app." It also covers "project-specific context (such as your app's directory structure or how you manage multiple app build variants), which could cause unexpected failures during build time."

The post does not say which tool carries out the project-level steps in the Firebase console. The agent skills documentation recommends installing, alongside the skills, Firebase's server for MCP, the open standard through which an AI application can discover and call tools on an external server.

Firebase's platform guides set out the steps the skill has to handle. Both pages were last updated on November 19, 2026.

Platform Requirement in Firebase documentation Detail given
iOS Dependency management Swift Package Manager, from the firebase-ios-sdk repository on GitHub
iOS Linker setting The -ObjC flag goes in the Other Linker Flags section of the target's build settings
Android Version control Firebase Android BoM, com.google.firebase:firebase-bom:35.2.0
Android Analytics library com.google.firebase:firebase-analytics, with no version number when the BoM is used
Android Without the BoM com.google.firebase:firebase-analytics:24.0.0

Neither guide states a minimum iOS, Xcode or Android API level.

Why screen views became manual work again

The Firebase SDK logs screen views automatically by default. The Firebase post says the SDK "supports automatic logging of screenview events, lightening your development load, with full flexibility to opt out whenever needed." It then explains why that is no longer enough for many apps: "with the shift toward declarative UI via modern frameworks (like Jetpack Compose or SwiftUI and cross-platform frameworks like Flutter or React Native), it's increasingly necessary for you to manually log screenview events."

Firebase's screen view documentation shows the mechanism behind the problem. Automatic logging identifies a screen by the class name of the UIViewController or Activity in focus, and on Apple platforms it depends on method swizzling, a technique that intercepts system calls at runtime. Declarative frameworks commonly draw many screens inside a single activity or controller, which leaves automatic logging with little to identify. According to the documentation, SwiftUI apps must set screen names manually through the FirebaseAnalyticsSwift module, or log screen views themselves. The documentation also gives games as an example of apps that do not use a separate controller or activity for each screen.

Manual logging uses the screen_view event with the parameters firebase_screen and firebase_screen_class. The documentation places the call in onAppear or viewDidAppear on Apple platforms and in onResume on Android. Events logged on a screen are tagged with firebase_screen_class and a generated firebase_screen_id.

Platform Switch for automatic screen reporting Manual screen view event
iOS, Swift FirebaseAutomaticScreenReportingEnabled set to NO in Info.plist AnalyticsEventScreenView
iOS, Objective-C Same Info.plist key kFIREventScreenView
Android, Kotlin and Java meta-data google_analytics_automatic_screen_reporting_enabled set to false inside the application tag of AndroidManifest.xml FirebaseAnalytics.Event.SCREEN_VIEW
Flutter (Dart) and web Not described logEvent with "screen_view" and the firebase_screen and firebase_screen_class parameters

Source: "Measure screenviews", Google Analytics for Firebase documentation, last updated November 19, 2026.

The second skill prompt in the Firebase post is aimed at this work. "Your AI coding agent can now do the work for you to instrument your apps that use modern declarative UIs so that you're accurately logging screenviews," Kahn and Liang write, suggesting the prompt "Help me instrument my app to log screenviews using Google Analytics for Firebase."

Installing the skills

According to the Firebase post, the skills can be installed "into most AI coding agents (like Antigravity, Claude Code, Codex, or Cursor agents)." Developers who already had the package installed were told to "make sure to update your agent skills to pull in all the new skills." For Claude Code and Codex, the package arrives as one of the agent plugins that bundle skills and connectors into a single install.

Agent Route in the Firebase blog post Commands and paths in the documentation Update method in the documentation
Antigravity Antigravity Marketplace Customizations, Marketplace tab, + on the Firebase plugin card; in the Antigravity CLI, /plugin install firebase@agent-marketplace; or npx skills add firebase/agent-skills --agent=antigravity Marketplace plugins are updated automatically by Antigravity; npx route uses npx skills update --all
Claude Code claude plugin marketplace add firebase/agent-skills, then claude plugin install firebase@firebase Same claude plugin marketplace update firebase
Codex Codex Plugin Directory codex plugin marketplace add firebase/agent-skills, then codex plugin add firebase@firebase; in the Codex app, Plugins, Firebase, + Install, then @Firebase in a new task codex plugin marketplace upgrade firebase
Cursor Cursor Marketplace Marketplace, or npx skills add firebase/agent-skills --agent=cursor npx skills update --all
Any other agent npx skills add firebase/agent-skills Same npx skills update --all

The repository behind the package, firebase/agent-skills, is published under the Apache 2.0 licence. On November 21 it showed 491 stars and 109 forks, along with plugin folders for Claude, Codex, Cursor and Kimi, a Gemini extension file and a Kiro folder. No tagged release was visible.

Nine months of Firebase skills

The Analytics skill is a late addition to a package that has been growing since February, and the pitch has been about efficiency from the start, measured in tokens, the subword units that language models read and write and on which AI providers bill. "Earlier this year, we introduced Firebase agent skills to give AI coding agents deep, domain-specific context and instructions for working with Firebase," the November 10 post says. That launch came in a February 17, 2026 Firebase blog post titled "Better code, fewer tokens: Introducing Agent Skills for Firebase", which said the skills worked with more than 30 agents and that the first release was tuned for full-stack web development. Mobile analytics came nine months later.

The November post repeats the efficiency claim, saying skills help agents "write better code with fewer tokens", and points to an August 11, 2026 post by Liang on what Firebase calls eval-driven development. That post published one comparison table, with results captured in July 2026.

Metric Without skills With skills Change (PPC Land calculation)
Pass rate 31.7% 78.0% +46.3 percentage points
Average input tokens 285.1k 169.5k -40.5%
Average output tokens 9.5k 5.8k -38.9%
Average duration 144.3 seconds 101.2 seconds -29.9%

The August post did not give the sample size behind the table or say which skills were tested, and the figures predate the Analytics skill. Nor does it put a price on the token savings. No equivalent figures have been published for firebase-analytics. Between the two posts, Firebase announced its plugin for Codex on September 28 and published a case study on October 8 about building an iOS app in six hours with Firebase and Antigravity.

Google's growing set of agent tools for ads and measurement

The Firebase release sits inside a wider push by Google to make its advertising and measurement products usable by software agents. In July 2025, Google released an experimental open-source MCP server for Google Analytics with eight tools, authenticated through the read-only Analytics scope. In November 2025, Google said Analytics Advisor would reach all English-language accounts in early December, a conversational assistant inside Analytics properties. At I/O in May 2026, Google introduced Antigravity 2.0, with a desktop app, a command-line tool and an SDK for self-hosted agents.

The pace picked up in the second half of the year. Version 4.0 of the Google Ads API Developer Assistant, published on August 25, targeted Antigravity and Claude Code by name. On November 2, the Google Ads MCP server gained four write tools in version 0.0.6, while campaigns, budgets, bids and keywords stayed out of reach. And on November 16, two days before the Analytics release note, Google said AdMob agent skills were available for coding assistants, with six skills listed in its documentation for the Google Mobile Ads Next-Gen SDK.

Google is not alone. Meta opened a beta Developer Tools MCP server with 10 tools at the end of June, and X gave AI agents 23 tools to run ad campaigns in August. Most of those products act on accounts. The Firebase skill acts on code, at the point where measurement data is first created.

Why the tracking code matters to advertisers

Google Analytics has spent the past two years courting advertisers, a shift PPC Land described in January as Analytics working its way back into advertisers' good graces. In 2026 that effort reached app data. On the same help page that carries the November 18 entry, Google announced on September 29 that "App conversions are now fully supported in your conversion reports, including performance, attribution analysis, and attribution models," with attribution settings adjustable separately for app conversions, a feature it says may not be available to every property. On July 13 it added a native AdMob linking flow in the Analytics admin console. One of the benefits listed for that flow reads: "Leverage ad impressions as conversions for Google Ads tROAS campaigns for ad revenue".

All of that depends on how the SDK is installed. When Google merged app and web reporting in 2019, PPC Land noted that the Firebase SDK was needed to bring app data into the new property type. In August 2025, Google's on-device conversion measurement for iOS app campaigns made Firebase iOS SDK 11.14.0 mandatory across most attribution partners unless a standalone SDK was used. Conversion tracking in Google Analytics rests on events that a property owner marks as key events. The release note names conversion tracking as part of what the skills are for, yet the supported tasks do not include it, and Google has not said when they will.

The SDK also carries legal weight. In September 2025, a jury ordered Google to pay $425.7 million in Rodriguez v. Google, a case in which the plaintiffs argued that the Web & App Activity setting did not stop data collection through the Firebase SDK inside third-party apps. Consent configuration for apps is a separate process: a Google technical guide published in December 2024 covered validating consent mode settings through Firebase debug mode. Neither the Firebase post nor the release note mentions consent.

Who checks what the agent writes? The Firebase post does not address review of the code an agent produces. A study of professional developers covered by PPC Land in January found that nine of 13 observed developers reviewed every change an agent made, and that experienced developers kept control of design decisions.

The sources leave several questions open. Neither says which declarative frameworks the screen view skill supports beyond naming four as the reason for manual logging, how it handles apps that mix automatic and manual screen reporting, or whether it touches consent settings. No evaluation results are given for the Analytics skill, and no date is set for the "additional development tasks". The blog post refers to skills in the plural while the documentation lists one, and the repository's README uses firebase/skills in its install commands where the post and documentation use firebase/agent-skills.

Key facts

  • Release note: "Google Analytics for Firebase (GA4F) SDK agent skills", What's new in Google Analytics, dated November 18, 2026, https://support.google.com/analytics/answer/9164320
  • Blog post: "Introducing Firebase agent skills for Google Analytics", The Firebase Blog, November 10, 2026, by Don Kahn and Charlotte Liang (Product Managers), https://firebase.blog/posts/2026/11/firebase-skills-for-google-analytics
  • Skill name: firebase-analytics, one of 13 skills in Firebase's agent skills documentation (last updated November 18, 2026), https://firebase.google.com/docs/ai-assistance/agent-skills
  • Supported tasks: initial integration and screen view logging, for iOS and Android apps
  • Example prompts: "Help me set up my app to use Google Analytics for Firebase." and "Help me instrument my app to log screenviews using Google Analytics for Firebase."
  • Agents named: Antigravity, Claude Code, Codex, Cursor; other agents through npx skills add firebase/agent-skills
  • Claude Code commands: claude plugin marketplace add firebase/agent-skills; claude plugin install firebase@firebase
  • Repository: github.com/firebase/agent-skills, Apache 2.0 licence
  • Original Firebase agent skills launch: February 17, 2026
  • Eval figures (July 2026, published August 11, 2026): pass rate 31.7% without skills, 78.0% with skills
  • Related documentation: "Measure screenviews" and the iOS and Android get-started guides, all last updated November 19, 2026
  • Android dependency shown: com.google.firebase:firebase-bom:35.2.0 with com.google.firebase:firebase-analytics

Timeline

Summary

Who: Google, through its Firebase and Google Analytics teams. The Firebase blog post was written by product managers Don Kahn and Charlotte Liang. The tools are aimed at iOS and Android developers using the Google Analytics for Firebase SDK, and at the AI coding agents they work with, including Antigravity, Claude Code, Codex and Cursor.

What: A Firebase agent skill for Google Analytics, listed in the documentation as firebase-analytics, that lets coding agents connect an app to Firebase, set up Google Analytics, add the SDK and its initialisation code, and add manual screen view logging for apps built with declarative frameworks such as SwiftUI and Jetpack Compose. The Analytics release note describes the skills as able to change "behavioral and conversion tracking", but only initial integration and screen view logging are supported so far.

When: The Firebase blog post was published on November 10, 2026. The Google Analytics release note is dated November 18, 2026, the same day the Firebase agent skills documentation was last updated. Further tasks are described as "coming soon", with no date.

Where: In AI coding agents, installed through the Antigravity Marketplace, Claude Code plugin commands, the Codex Plugin Directory, the Cursor Marketplace or the npx skills command, and in the code of iOS and Android apps. No market restrictions are stated.

Why: According to Google, setup involves platform-specific steps that cause build failures when missed, and declarative interface frameworks have made automatic screen view logging less reliable, forcing manual instrumentation. For advertisers, the code these agents write determines the app engagement data that reaches Google Analytics reports, app conversion reporting and linked Google Ads accounts.