---
name: editing
description: >-
  Use before making changes to existing files. Routes to Julie's edit_file
  tool without a Read + Edit loop. Trigger on fix, change, update, modify,
  refactor, rename, replace, add, remove, move, or any task involving an
  existing file.
allowed-tools: mcp__julie__edit_file, mcp__julie__get_symbols, mcp__julie__deep_dive, mcp__julie__fast_search, mcp__julie__fast_refs
---

# Editing Files with Julie

Julie's edit tools change existing files without reading them first. This is
the default path for file modifications.

## Which tool do I use?

- **Creating a new file?** Use the Write tool. This skill doesn't apply.
- **Changing any text in a file?** Use `edit_file` with `old_text` and `new_text`.
- **Need to understand the file first?** Use `get_symbols` for structure or `deep_dive` for full context, then use `edit_file`. Not Read.

## Stop and check

If you catch yourself thinking any of these, you're about to waste tokens:

| Thought | What to do instead |
|---------|-------------------|
| "I need to read the file first" | No. `edit_file` uses DMP fuzzy matching on `old_text`. It does not need Read. |
| "It's a quick change" | `edit_file` handles that well. `edit_file(old_text=..., new_text=..., dry_run=true)` is the fast path. |
| "I'm not sure which symbol I need" | Use `fast_search`, `get_symbols`, or `deep_dive`, then use `edit_file`. Still no Read + Edit loop. |
| "This isn't a code file" | `edit_file` works on ANY text file: YAML, TOML, Markdown, .gitignore, configs, everything. |
| "The symbol edit is complex" | Use `dry_run=true` first. |

## Workflow

1. **Always preview first**: `dry_run=true` (the default). Review the diff.
2. **Then apply**: same call with `dry_run=false`.

If a file was only read with offset/limit pagination, treat the view as partial: re-read the whole file before overwriting, or narrow the patch until the match is unambiguous. See `references/partial-read-patch-pitfall.md`.

### edit_file (for any text)

- `old_text`: text to find (DMP fuzzy matched)
- `new_text`: replacement text
- `occurrence`: `"first"` (default), `"last"`, or `"all"`

## Example: the cost of Read + Edit

Changing a version number in Cargo.toml:

**Read + Edit pattern (5 calls, ~800 tokens):**
1. Edit -- fails ("File has not been read yet")
2. Read Cargo.toml -- waste
3. Grep for the version line -- unnecessary
4. Read again with offset -- more waste
5. Edit -- finally works

**edit_file pattern (2 calls, ~200 tokens):**
1. `edit_file(file_path="Cargo.toml", old_text='version = "6.6.2"', new_text='version = "6.6.3"', dry_run=true)` -- preview
2. Same call with `dry_run=false` -- done

4x fewer tokens. 3 fewer round trips.