ghostwriter

name	ghostwriter
description	Targeted code reducer for minimizing token usage in Codex. Use it before reading or editing large files so analysis stays focused on the failing logic.

Core Rules

1. No Reproduction Files: Do not create temporary reproduction scripts just to inspect the bug.
2. Targeted Reading: Use ghostwriter prune to reduce the file before analysis.
3. In-Memory Output: Treat the reduced output as analysis context only. Do not save it as a new source file.
4. Direct Modification: Apply fixes to the original source file after inspecting the reduced output.

Execution Pattern

1. Identify the file and target line from the user request, stack trace, or terminal output.
2. Run ghostwriter prune <file> --line <number> --stdout --silent.
3. Analyze the reduced code in-context.
4. Edit the original source file normally.

Best Uses

• Debugging stack traces in large files
• Explaining one method inside a very large class or module
• Reducing token usage before patching a narrow code path