google-gemini/gemini-cli

Agent repeatedly runs broad searches causing massive context overflow

Open

#17448 opened on Jan 24, 2026

View on GitHub
 (8 comments) (0 reactions) (1 assignee)TypeScript (103,992 stars) (13,657 forks)batch import
area/agenthelp wantedkind/bugpriority/p1priority/p2status/bot-triaged

Description

What happened?

This happens fairly often. I will ask the agent to review a repo or find code, and it executes an overly broad SearchText command (e.g., searching 'Gmail' in ./ as shown in the screenshot). This picks up thousands of matches, often including huge debug logs or build artifacts.

Instead of refining the search, the agent attempts to ingest all 5,000+ matches immediately. In the attached example, it tried to send a payload of 41 million tokens against a ~900k limit.

Additionally, the UI becomes very confusing here. It displays a yellow "Sending this message..." warning, but it's unclear if the message has actually been sent, if the CLI is hanging, or how to cancel the operation without killing the terminal.

What did you expect to happen?

Smarter Agent: When a search returns thousands of results, the agent should recognize this is too much and refine its search query—not try to read them all.

Hard Block: The CLI should prevent the agent from attempting to send a request that is 40x over the context limit.

Clearer UI: When a limit is hit, the UI should offer a clear "Cancel" or "Abort" option rather than just saying "Sending...".

Client information

I terminated the session, so I can't run /about.

Version: 0.24.0-preview OS: macOS

#######

Screenshot: https://screenshot.googleplex.com/5HatNtr4PQSP7Em

#######

✓ SearchText 'Gmail' within ./

Found 5006 matches

✓ ReadFile scripts/smart_ingest.py

i Sending this message (41740251 tokens) might exceed the remaining context window limit (904878 tokens).

Login information

API key

Anything else we need to know?

No response

Contributor guide

Tech stack
typescriptnodejs
Domain
cliperformance
Issue type
bug
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
4
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-2 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
clear
Prerequisites
TypeScriptNode.js
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
40
Research direction
Examine the search execution logic in the Gemini CLI to understand how the agent performs SearchText and reads files. Identify the point where context limits are checked and where a hard block can be added. Explore the UI code responsible for displaying the sending warning to implement a cancel/abort button. Review any existing tests or discussions in the issue comments for design constraints.