rtk

name	rtk
description	Token optimizer for AI coding agents. Wraps common CLI commands with rtk to reduce LLM token consumption by 60-90%.
compatibility	OpenCode with RTK binary installed
metadata	{"author":"Sisyphus","version":"1.0.0"}

RTK — Rust Token Killer

Purpose: Reduce token consumption by 60-90% on shell commands. RTK filters and compresses command output before it reaches your context.

When This Skill Activates

This skill is loaded automatically. Once active, always prefix supported commands with rtk.

Command Mappings (MANDATORY)

When running shell commands via Bash tool, use rtk prefix for these commands:

Instead of	Use	Savings
git status	rtk git status	~80%
git diff	rtk git diff	~75%
git log	rtk git log	~80%
git add/commit/push/pull	rtk git add/commit/push/pull	~90%
ls / ls -la	rtk ls	~80%
cat file	rtk read file	~70%
grep pattern .	rtk grep pattern .	~80%
rg pattern	rtk grep pattern .	~80%
npm test / cargo test / pytest	rtk test <cmd>	~90%
npm run build / cargo build	rtk err <cmd>	~80%
gh pr list/view	rtk gh pr list/view	~70%
docker ps	rtk docker ps	~80%
eslint / tsc	rtk lint / rtk tsc	~80%

Searching Inside node_modules / Ignored Directories

By default, rtk grep respects .gitignore rules — meaning node_modules, .nuxt, dist, etc. are excluded. This is the right behavior 99% of the time.

When you need to search inside ignored directories (debugging a library, checking an API signature, tracing a dependency bug):

# Search all files including node_modules (--no-ignore bypasses .gitignore)
rtk grep "defineStore" . --no-ignore

# Search a specific package only (combine --no-ignore with --glob)
rtk grep "defineStore" . --no-ignore --glob 'node_modules/pinia/**'

What does NOT work:

• rtk grep "pattern" node_modules/pinia/ — still excluded even with direct path
• rtk grep "pattern" . --glob 'node_modules/**' — glob alone doesn't override .gitignore

Key flag: --no-ignore — this is the ONLY way to search ignored directories with rtk grep.

Other useful rtk grep flags

rtk grep "pattern" . -t ts          # Filter by file type (ts, py, rust, etc.)
rtk grep "pattern" . -m 100         # Increase max results (default: 50)
rtk grep "pattern" . -u             # Ultra-compact mode (even fewer tokens)
rtk grep "pattern" . -l 120         # Max line length before truncation (default: 80)

Commands to NOT Wrap

Do NOT prefix these with rtk (unsupported or counterproductive):

• npx, npm install, pip install (package managers)
• node, python3, ruby (interpreters)
• nano-brain, openspec, opencode (custom tools)
• Heredocs (<<EOF)
• Piped commands (cmd1 | cmd2) — wrap only the first command if applicable
• Commands already prefixed with rtk

How RTK Works

Without RTK:  git status → 50 lines raw output → 2,000 tokens
With RTK:     rtk git status → "3 modified, 1 untracked ✓" → 200 tokens

RTK runs the real command, then filters/compresses the output. The agent sees a compact summary instead of verbose raw output.

Detection

Before using RTK commands, verify it's installed:

rtk --version

If rtk is not found, skip this skill — run commands normally without the rtk prefix.

Token Savings Reference

Typical 30-min coding session:

• Without RTK: ~150,000 tokens
• With RTK: ~45,000 tokens
• Savings: ~70%

Biggest wins: test output (rtk test — 90%), git operations (rtk git — 80%), file reading (rtk read — 70%).