Firebase CLI Now Detects When an AI Agent Is Running It

Firebase CLI 15.26.0, released Aug 5, now checks whether an AI agent is driving it before it decides how to behave. If it detects one, it automatically switches to non-interactive mode so it stops hanging on prompts nobody is there to answer, according to the GitHub release notes.

---

What changed

Interactive prompts have always been the CLI’s default: run firebase deploy or firebase init and it stops to ask questions a person can answer in a terminal. An agent calling the same commands from a script or sandbox has no terminal to answer from, so the process either hangs until it times out or fails outright waiting on stdin that never arrives. Firebase 15.26.0 closes that gap by detecting an agent in the environment and enabling --non-interactive on its own, per the release notes.

The same release adds a fix for the other half of that problem: authentication. Non-interactive login previously meant provisioning a service account or a pre-existing refresh token, both awkward to bootstrap for an agent starting from nothing in a fresh sandbox. 15.26.0 adds a two-phase flow instead: run firebase login --non-interactive, visit the printed link to get an authorization code, then complete the login with firebase login <auth_code>. No browser redirect back to the CLI required.

A smaller but concrete change in the same release: three new appcheck:services: commands — list, get and set — let you read and change App Check enforcement per Firebase product from the CLI, instead of toggling each service one at a time in the console.

---

Why now

Firebase shipped an official Agent Skills library back in February, packaging Firebase-specific instructions for Claude Code, Cursor, Gemini CLI and other assistants so they write correct Firebase code with less trial and error. That release assumed agents would be writing Firebase code. This one assumes they’ll be running the CLI directly, deploying, initializing projects and managing App Check without a human confirming each step. The two problems this release fixes, a stuck prompt and an unbootstrappable login, are exactly the ones that show up the moment an agent tries to do that unsupervised.

Firerun’s take: treating “an AI agent might be the caller” as a first-class CLI input, on the same footing as a CI runner or a human at a keyboard, shows where developer tooling is heading, not just at Google. A CLI that blocks on stdin because it assumed a human was always on the other end is a design choice from before agents ran build steps unattended. This release retires that assumption.

Key Takeaways

• Firebase CLI 15.26.0 (Aug 5) auto-detects an AI agent in the environment and enables non-interactive mode to prevent the CLI hanging on prompts.
• A new two-phase login flow (firebase login --non-interactive → auth code → firebase login <code>) gives agents a way to authenticate without a service account or browser redirect.
• New appcheck:services:list/get/set commands manage App Check enforcement per service from the CLI.
• The release follows Firebase’s Agent Skills library, shipped in February for Claude Code, Cursor and Gemini CLI. This release is the CLI catching up to the same assumption.