hermes - 💡(How to fix) Fix feat: `!command` prefix for direct shell execution (like Claude Code)

Feature Description

Add a ! prefix feature similar to Claude Code's, where typing !<command> in the chat executes the shell command directly without going through the LLM, and returns the output inline.

Motivation

When working in the terminal, I frequently need to run quick shell commands (e.g. !ls, !git status, !docker ps) to check state. Currently every command must go through the LLM tool-calling pipeline, which:

1. Adds latency (the LLM parses, decides to use terminal, executes, returns)
2. Wastes tokens on trivial operations
3. Breaks flow when you just need a quick !pwd or !cat config.yml

This is one of the most appreciated features in Claude Code.

Proposed Solution

In the CLI input handler, before sending to the LLM, check if the user's message starts with !. If so, strip the ! prefix and execute the remainder as a shell command directly, returning stdout/stderr as if it were a terminal tool result.

Key behaviors:

• !command to execute and return output inline (no LLM)
• !?command to ask LLM to interpret the output
• Only in interactive CLI mode (not gateway)
• Standard TIRITH approval still applies for dangerous commands
• History records both command and output

Code Locations

Alternatives Considered

1. Slash command /run - Works but /run ls is more typing than !ls
2. Just telling the LLM - Adds 2-5s latency on trivial commands
3. Separate REPL - Over-engineered for a simple passthrough

Example Comparison