Shell Mode

Shell Mode turns your terminal into an AI-augmented command line. Every command you run is visible to the Agent, which can instantly help you debug failures, explain output, and suggest next steps.

Entering Shell Mode

Start Shell Mode by running:

bash
vyre shell

Your terminal prompt changes to indicate Shell Mode is active:

bash
● vyre  ~/my-project $

To exit, press Ctrl+D or type exit.

How it works

In Shell Mode:

  • Every command you run is visible to Agent — it sees both the command and its full output.
  • When a command fails (non-zero exit code), Agent automatically offers to help.
  • You can address Agent at any time by typing a message starting with >.
  • Agent can run commands on your behalf — it asks for confirmation before executing.
bash
● vyre  ~/my-project $ npm run build
> next build

   ▲ Error: Cannot find module '@/components/Button'

Agent: I can see the import path is wrong. The component is at 
'@/components/ui/Button'. Want me to fix all occurrences?
[y/n] y
→ Fixed 3 import paths.

Shell Mode commands

Type these in the Shell Mode prompt to talk to Agent:

bash
> fix this           # Fix the last command that failed
> explain this       # Explain the last command's output
> what caused this   # Root cause analysis of the last error
> how do I [task]    # Get help with a specific task
> run [description]  # Ask Agent to construct and run a command

Or just type naturally — Shell Mode understands conversational messages.

Use cases

Disabling auto-suggestions

By default, Agent suggests help after every failed command. Disable this in Shell Mode settings:

bash
vyre shell --no-auto-suggest

You can still address Agent manually at any time with >.

Shell Mode in CI

Shell Mode can run in non-interactive CI environments. In CI, suggestions are written to a log file instead of stdout:

bash
vyre shell --ci --log-file ./vyre-shell.log npm run build

Shell Mode is especially powerful during incident response. Start a Shell Mode session when debugging a production issue — every command you run and every output you see is tracked, and Agent builds a growing mental model of the incident as you investigate.