 |
VOOZH | about |
|
VOOZH | about |
We’re so glad you’re here. You can expect all the best TNS content to arrive Monday through Friday to keep you on top of the news and at the top of your game.
Check your inbox for a confirmation email where you can adjust your preferences and even join additional groups.
Follow TNS on your favorite social media networks.
Become a TNS follower on LinkedIn.
Check out the latest featured and trending stories while you wait for your first TNS newsletter.
Google has added hooks to Gemini CLI, its terminal-based competitor to Anthropic’s Claude Code.
Hooks ensure that Gemini CLI runs a given script or program inside of the agentic loop and bring a larger degree of control to the agentic development loop. These could be used, for example, to run security scanners or compliance checks, log tool interactions, inject relevant information into the context window, or even adjust the model’s parameters on the fly.
As the Gemini CLI team notes in the announcement, “efficiency in the age of agents isn’t just about writing code faster; it’s about building custom tools that adapt to your specific environment.”
While a developer could try to instruct the agent to run a specific script at certain times within the loop in the prompt or AGENTS.md file, given the non-deterministic nature of those agent models, there’s no guarantee that this will actually happen or that the agent won’t forget about this instruction over time.
If this sounds familiar, it’s likely because you already know about Claude Code Hooks, which first introduced this idea last September (though there is also a GitHub issue from July 2025 that proposes this feature). Google’s implementation is not quite a one-to-one match to Anthropic’s, but it should only take a few minutes to adapt an existing Claude hook to Gemini CLI.
Like with hooks in Claude Code, Gemini CLI also implements roughly a dozen lifecycle events where a hook can fire. That may be right at the session start, after the user submits a prompt but before the agent starts planning (to add context, for example), before tools are selected (to optimize the tool selection or filter available tools), and similar moments in the agent loop.
The hooks are defined as JSON files that describe when they are invoked and which script they should run. Those scripts are standard Bash scripts and Google notes that it is essential to keep those hooks fast because they do run synchronously and delays in the script will also delay the agent response.
Google recommends that developers use parallel operations and caching when possible to keep the operations fast.
One interesting use case for hooks is to utilize the ‘AfterAgent’ hook, which fires when the agent loop ends, to force the agent into a continuous loop to work on a difficult task — while also refreshing the context between those runs to avoid context rot.
As for security, it’s important to stress that hooks will have the user’s privileges, and Google notes that developers should review the source code of any third-party hooks.
Hooks, which are now available as part of the Gemini CLI v0.26.0 update, can also be packaged inside Gemini CLI extensions. That’s Google’s format for packaging prompts, MCP servers, sub-agents, and agent skills — and now hooks — into a single sharable package.
