lx

Converts files into simple Markdown-fenced blocks for easy use in LLM chats.

The goal is to make prompt setup repeatable. Instead of letting an agent guess context or manually selecting files in a UI, you define the exact context you want in one shell command and rerun it whenever you need a fresh session. It works smoothly with tools like rg -l, fd, and recursive shell globs.

This gives you a stable, controllable workflow:

• You decide exactly what context the model sees.
• If a conversation drifts, you restart instantly with the same context.
• Adjust the command, rerun it, and paste the updated output.

---

Installation

Via go install:

go install github.com/rasros/lx/cmd/lx@latest

Or via curl into $HOME/.local/bin/lx:

curl -fsSL https://raw.githubusercontent.com/rasros/lx/main/install.sh | bash

---

Basic usage

Format a single file as an LLM-ready snippet:

lx cmd/lx/main.go

Example output (trimmed):

cmd/lx/main.go (18 rows)
---
```go
package main
... (rest omitted)
```

You can put it directly in your clipboard by piping it to a copy tool:

# Wayland (Ubuntu, Debian)
lx file.py | wl-copy

# X11
lx file.py | xclip -selection clipboard
# or
lx file.py | xsel --clipboard --input

# macOS
lx file.py | pbcopy

# MSYS2
lx file.py | clip

---

Features

• Generates Markdown headers and fenced blocks for one or many files.
• Automatically detects fenced-code language from file extension.
• Supports lightweight, ergonomic slicing (-h, -t, -n).
• Optional line numbers for precise AI instructions.
• Reads filenames from CLI args or stdin (great with rg, find, etc.).
• Customizable delimiters with placeholders.

---

More examples

Filtering file names

We can select multiple files in shells that allow recursive glob:

lx **/*.py

If you need to exclude certain files we rely on standard tools for file selection.

This example uses includes all python files except those with name ending in _test.py. Here fd or find:

# find using stdin-mode
find . -name '*.py' ! -name '*_test.py' | lx

# fd using stdin-mode
fd -e py -E "*_test.py" | lx

Or through shell glob syntax:

# zsh glob
lx **/*.py~*_test.py 

# bash glob
shopt -s globstar extglob
lx **/!(*_test).py

# fish glob
lx **/*.py ^**/*_test.py

# MSYS2 glob
shopt -s globstar extglob
lx **/!(*_test).py

Pattern search

Searching for patterns is easily done through grep -l and pipe matching files to lx.

This example searches for files with a function starting with save under the src folder:

# grep
grep -rl "def save" src | lx

# ripgrep
rg -l "def save" src | lx

Line numbers: -l

TOON supports line numbers so we do too 🤷.

This is actually very useful for letting the LLM reference where in a big log dump something is wrong. Or if you specifically include prompting about instructions about line number references.

lx -l server.log

Slicing: -t -h -n

While iterating on the command it's convenient to slice files so you can more easily see what's included:

# First 40 lines
lx -h40 server.log

# Last 80 lines
lx -t80 server.log

# Both ends (split 60 lines between head & tail)
lx -n60 server.log

Short forms like -h5, -t10, -n2 are supported.

Custom delimiters and placeholders

Default delimiters:

{filename} ({row_count} rows)
---
```{language}
...file contents...
```

Override them:

lx \
  --prefix-delimiter="### {filename}\n```{language}\n" \
  --postfix-delimiter="```\n\n" \
  file.go

Placeholders:

• {filename}
• {row_count}
• {byte_size}
• {last_modified}
• {language}

Use these to enforce consistent prompt structure and regenerate identical context anytime.