gemini-cli - 💡(How to fix) Fix Slash commands (e.g., /quit, /help) not intercepted by CLI; passed to model as text [3 comments, 2 participants]

What happened?

While using the Gemini CLI in an interactive session, I attempted to use slash commands to manage the session. Specifically, when I typed /quit, the CLI did not terminate the application. Instead, the command string was sent directly to the LLM as a prompt. The model responded with a conversational "See you next time!" message, but the CLI process remained active.

What did you expect to happen?

I expected the CLI to intercept any input starting with a / (slash), identify it as a client-side command, and execute the corresponding action (in this case, exiting the application) immediately without consuming model tokens or generating a response.

Client information

Client Information: Gemini CLI Version:* 0.38.2 Operating System:* Windows 11 x64 version 25H2 (OS Build 26200.8246) Node.js Version:* v24.13.1 Date Observed:* Saturday, April 22, 2026

Login information

Login via Google account

Anything else we need to know?

The issue appears to affect multiple slash commands, including /help, which resulted in the model providing a summary of CLI features instead of the CLI displaying its built-in help documentation.
The CLI correctly loaded multiple project-specific GEMINI.md files at the start of the session, indicating the environment was initialized, but the input interceptor for slash commands failed to trigger.
Even when running the "/about" command, it returns:

/about
▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄
Responding with gemini-3-flash-preview

╭─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮ │ ≡ Agent Completed (ctrl+o to expand) │ │ ✓ cli_help · Completed successfully │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯

✦ I am Gemini CLI, a specialized AI agent for software engineering. I can analyze codebases using a massive context window, execute shell commands, and automate complex development tasks through integrated tools and custom skills

Here's what displays in the debug console when i launch. Debug Console (F12 to close) │ │ │ │ ℹ Loaded cached credentials. │ │ │ │ ℹ Detected terminal background color: #0c0c0c │ │ │ │ ℹ Loading extension: caveman │ │ ℹ Loading extension: context7 │ │ ℹ Scheduling MCP context refresh... │ │ ℹ Executing MCP context refresh... │ │ ⚠ [STARTUP] Cannot start phase 'load_builtin_commands': phase is already active. Call end() before starting again. (x2) │ │ ℹ MCP context refresh complete. │ │ ⚠ [STARTUP] Cannot start phase 'load_builtin_commands': phase is already active. Call end() before starting again. │ │ ⚠ Skill "skill-creator" from "C:\Users\luowe.gemini\skills\skill-creator\SKILL.md" is overriding the built-in skill. │ │ ⚠ [STARTUP] Phase 'load_builtin_commands' was started but never ended. Skipping metrics. │ │ ℹ Authenticated via "oauth-personal". │ │ ℹ Registering notification handlers for server 'context7'. Capabilities: { tools: { listChanged: true } } │ │ ℹ Server 'context7' supports tool updates. Listening for changes... │ │ ⚠ [STARTUP] Cannot start phase 'load_builtin_commands': phase is already active. Call end() before starting again. │ │ ℹ Scheduling MCP context refresh... │ │ ℹ Executing MCP context refresh... │ │ ℹ MCP context refresh complete. │ │ ⚠ [STARTUP] Cannot start phase 'load_builtin_commands': phase is already active. Call end() before starting again.