Summary

Vibe currently has no mechanism to intercept or transform bash commands before they are executed. Adding a BeforeTool hook (similar to Gemini CLI's BeforeTool hook) would allow external tools to transparently rewrite commands before execution.

Use Case: RTK (Rust Token Killer)

RTK is a CLI proxy that saves 60-90% of LLM tokens by filtering and compressing command output before it reaches the model context. It works by rewriting commands:

git status     → rtk git status      # ~70% fewer tokens
ls src/        → rtk ls src/         # ~60% fewer tokens
cargo test     → rtk cargo test      # ~90% fewer tokens
gh pr list     → rtk gh pr list      # ~87% fewer tokens

RTK already has transparent hook support for:

• Claude Code — via PreToolUse hook in settings.json
• Gemini CLI — via BeforeTool hook in settings.json

For Vibe, the current workaround is a system prompt instructing the model to use rtk commands. This is unreliable — the model sometimes ignores it.

Proposed API

In ~/.vibe/config.toml or a new hooks section:

[hooks]
before_tool = "~/.vibe/hooks/before-tool.sh"

The hook receives tool call info via stdin (JSON) and can return a rewritten command or allow/block execution:

{
  "tool": "bash",
  "input": { "command": "git status" }
}

Output (stdout):

{ "action": "rewrite", "input": { "command": "rtk git status" } }

or

{ "action": "allow" }

Reference Implementations

• Claude Code: PreToolUse hook docs
• Gemini CLI: BeforeTool hook in ~/.gemini/settings.json
• RTK hook script: rtk-rewrite.sh

Impact

This would make Vibe a first-class citizen in token-optimized AI workflows, and would enable the RTK integration to work transparently without relying on model instruction-following.