xtrshow 0.2.1

The bridge between your local codebase and LLMs.

xtrshow

The bridge between your local codebase and LLMs.

xtrshow is a lightweight CLI suite designed to optimize the "Context Injection" and "Code Application" loop when working with AI coding assistants (ChatGPT, Claude, Gemini, etc.).

It consists of two tools:

1. xtrshow: A TUI to select, compress, and format code context for your LLM.
2. xtrpatch: A surgical patching tool to apply LLM-generated changes back to your source code safely.

Installation

pip install xtrshow

The Workflow

1. Extract Context (xtrshow)

Stop manually copying and pasting files.

Run xtrshow in your directory. Navigate with arrow keys, select files with Space, and hit Enter (or use -o to save to a file).

# Open interactive TUI
xtrshow

# Or pipe directly to clipboard (Mac/Linux)
xtrshow | pbcopy 

The output is formatted specifically for LLMs, including file paths and line numbers (enabled by default) to assist in referencing specific code blocks.

2. Prompt the LLM

Paste the output into your LLM. When asking for changes, include the following instruction to get a compatible patch:

"Provide changes using the Multi-File Search and Replace Block format."

3. Apply Changes (xtrpatch)

Save the LLM's response to a file (e.g., response.txt) and apply it:

xtrpatch response.txt

xtrpatch is whitespace-insensitive and uses fuzzy matching, making it significantly more robust than standard git apply when dealing with LLM hallucinations or formatting errors.

---

Features

Safety First (Atomic Backups)

xtrpatch never destroys code.

• Every time a file is patched, the original is backed up to .xtrpatch/<file>.orig.
• Backups are versioned (file.py.orig, file.py.1.orig, file.py.2.orig).
• The patch itself is archived alongside the backup.

Reverting

Made a mistake? You can unwind changes easily:

# Revert a specific file to its state before the last patch
xtrpatch --revert src/main.py

# Revert all files modified by a specific patch file
xtrpatch --revert response.txt

File Creation

If the LLM wants to create a new file, xtrpatch handles that too.

--- a/src/new_feature.py
<<<<
====
def hello():
    print("New file created!")
>>>>

Patch Format Specification

xtrpatch uses a custom, hallucination-resistant format:

--- a/path/to/file.py
<<<< 50
def old_function():
    return False
====
def old_function():
    return True
>>>>

• Header: --- a/path/to/file
• Start: <<<< LINE_HINT or <<<< START~END (e.g., <<<< 50~55). The ~ indicates an approximate range/fuzzy hint.
• Search Block: The existing code to find.
• Divider: ====
• Replace Block: The new code.
• Divider (Optional): ==== (Use a second divider to provide Tail Context).
• Tail Context (Optional): Code that must exist immediately after the block to verify location.
• End: >>>>

Example with Tail Context

--- a/src/main.rs
<<<< 10~15
    let x = 1;
    let y = 2;
====
    let x = 10;
    let y = 20;
====
    // This line must exist after the block for the patch to apply
    println!("Calculating...");
>>>>

License

Apache 2.0