rtk-ai/rtk

Need fix `grep` savings to much lower

Open

#545 opened on Mar 12, 2026

View on GitHub
 (1 comment) (0 reactions) (0 assignees)Rust (2,914 forks)batch import
area:clieffort-mediumenhancementhelp wantedpriority:medium

Repository metrics

Stars
 (48,085 stars)
PR merge metrics
 (Avg merge 11d 1h) (45 merged PRs in 30d)

Description

Problem

rtk gain shows much lower savings for rtk grep

On my session, rtk gain reported:

  • Command: rtk grep
  • Count: 94
  • Saved: 3.3M
  • Avg%: 17.9%

But the README advertises roughly -80% savings for grep / rg.

Why this is confusing

This makes it look like rtk grep underperforms badly in real usage, even though the likely issue is broader:

  • Claude Code prefers native tools like Grep, Read, and Glob
  • those native tools bypass the Bash hook
  • RTK therefore does not see or compress much of the actual search workload
  • rtk gain ends up showing a result that users can easily interpret as “grep savings are poor”

Expected behavior

One of these should happen:

  1. rtk gain / rtk discover should show “reachable vs unreachable” search activity
  2. the UI should otherwise make it clearer that this is a tooling-path limitation, not necessarily an rtk grep filter-quality problem

Actual behavior

rtk gain shows low rtk grep savings without enough context to explain why.

Environment

  • RTK version: 0.28.2
  • OS: [26.3 (25D125)
  • Agent/client: OpenCode & Codex (CLI & desktop)
  • Hook installed: yes

Contributor guide