lonny-agent 0.2.4 → 0.2.6

package/src/tools/__tests__/edit.test.ts

@@ -1,19 +1,410 @@
 import * as fs from 'node:fs'
+import * as os from 'node:os'
 import * as path from 'node:path'
 import { afterAll, beforeAll, describe, expect, it } from 'vitest'
 import { FileReadTracker } from '../../diff/apply.js'
-import { createEditTool } from '../edit.js'
+import {
+  computeDiff,
+  createEditTool,
+  escapeHtml,
+  findAllLinesTolerant,
+  generateDiff,
+  normalizeLine,
+  parseMarkdownEdit,
+  renderDiffTerminal,
+} from '../edit.js'
 import { makeTempDir } from './helpers.js'
 
-describe('edit tool', () => {
+// ── Minimal mock for FileReadTracker ─────────────────────────────────────
+function makeApplier(): FileReadTracker {
+  const readSet = new Set<string>()
+  return {
+    markRead(f: string) {
+      readSet.add(f)
+    },
+    checkModified(f: string): string | null {
+      return readSet.has(f) ? null : `Warning: file was not read first.`
+    },
+  } as FileReadTracker
+}
+
+// ── Tests for pure functions ─────────────────────────────────────────────
+describe('normalizeLine', () => {
+  it('trims leading whitespace', () => {
+    expect(normalizeLine('  hello')).toBe('hello')
+  })
+
+  it('trims trailing whitespace', () => {
+    expect(normalizeLine('hello  ')).toBe('hello')
+  })
+
+  it('collapses internal space runs', () => {
+    expect(normalizeLine('a    b')).toBe('a b')
+  })
+
+  it('collapses internal tab runs', () => {
+    expect(normalizeLine('a\t\tb')).toBe('a b')
+  })
+
+  it('handles mixed spaces and tabs', () => {
+    expect(normalizeLine('  a \t b  ')).toBe('a b')
+  })
+
+  it('reduces blank line with spaces to empty string', () => {
+    expect(normalizeLine('   ')).toBe('')
+  })
+
+  it('preserves single spaces', () => {
+    expect(normalizeLine('a b c')).toBe('a b c')
+  })
+
+  it('handles empty string', () => {
+    expect(normalizeLine('')).toBe('')
+  })
+})
+
+describe('findAllLinesTolerant', () => {
+  it('finds exact match', () => {
+    const result = findAllLinesTolerant('hello\nworld\nfoo\n', 'hello\nworld')
+    expect(result).toHaveLength(1)
+    expect(result[0]!.index).toBe(0)
+    expect(result[0]!.length).toBe('hello\nworld'.length)
+  })
+
+  it('finds match with trailing whitespace in content', () => {
+    const result = findAllLinesTolerant('hello \nworld\n', 'hello\nworld')
+    expect(result).toHaveLength(1)
+    // "hello \n" = 7 chars, "world" = 5 chars, total match = 12
+    expect(result[0]!.length).toBe(12)
+  })
+
+  it('finds match with leading whitespace in content', () => {
+    const result = findAllLinesTolerant('  hello\nworld\n', 'hello\nworld')
+    expect(result).toHaveLength(1)
+    // "  hello\n" = 8 chars, "world" = 5, total = 13
+    expect(result[0]!.length).toBe(13)
+  })
+
+  it('finds match with extra internal spaces', () => {
+    const result = findAllLinesTolerant('foo  bar\nbaz\n', 'foo bar\nbaz')
+    expect(result).toHaveLength(1)
+  })
+
+  it('returns empty array for empty oldString', () => {
+    expect(findAllLinesTolerant('abc', '')).toEqual([])
+  })
+
+  it('returns empty when oldString longer than content', () => {
+    expect(findAllLinesTolerant('a', 'a\nb\nc')).toEqual([])
+  })
+
+  it('returns empty when not found', () => {
+    expect(findAllLinesTolerant('abc\ndef\n', 'xyz')).toEqual([])
+  })
+
+  it('finds multiple matches', () => {
+    const result = findAllLinesTolerant('a\nb\na\nb\n', 'a\nb')
+    expect(result).toHaveLength(2)
+  })
+
+  it('computes correct index for non-first match', () => {
+    const result = findAllLinesTolerant('x\na\nb\n', 'a\nb')
+    expect(result).toHaveLength(1)
+    expect(result[0]!.index).toBe(2) // "x\n" = 2 chars
+  })
+
+  it('computes correct length matching original whitespace', () => {
+    const result = findAllLinesTolerant('hello \nworld\n', 'hello\nworld')
+    expect(result).toHaveLength(1)
+    const match = contentSlice('hello \nworld\n', result[0]!)
+    expect(match).toBe('hello \nworld')
+  })
+})
+
+function contentSlice(s: string, m: { index: number; length: number }): string {
+  return s.slice(m.index, m.index + m.length)
+}
+
+describe('computeDiff', () => {
+  it('returns empty array for empty strings', () => {
+    expect(computeDiff('', '')).toEqual([])
+  })
+
+  it('returns delete lines when new is empty', () => {
+    const result = computeDiff('a\nb', '')
+    expect(result).toHaveLength(2)
+    expect(result.every(l => l.type === 'delete')).toBe(true)
+  })
+
+  it('returns insert lines when old is empty', () => {
+    const result = computeDiff('', 'a\nb')
+    expect(result).toHaveLength(2)
+    expect(result.every(l => l.type === 'insert')).toBe(true)
+  })
+
+  it('returns equal lines for unchanged content', () => {
+    const result = computeDiff('hello\nworld', 'hello\nworld')
+    expect(result).toHaveLength(2)
+    expect(result.every(l => l.type === 'equal')).toBe(true)
+  })
+
+  it('produces mixed diff for changed content', () => {
+    const result = computeDiff('hello\nworld', 'hi\nworld')
+    // hello→hi should be delete+insert, world should be equal
+    expect(result).toHaveLength(3)
+    expect(result[0]!.type).toBe('delete')
+    expect(result[0]!.content).toBe('hello')
+    expect(result[1]!.type).toBe('insert')
+    expect(result[1]!.content).toBe('hi')
+    expect(result[2]!.type).toBe('equal')
+    expect(result[2]!.content).toBe('world')
+  })
+})
+
+describe('renderDiffTerminal', () => {
+  it('returns empty string for empty input', () => {
+    expect(renderDiffTerminal([])).toBe('')
+  })
+
+  it('renders delete lines in red with - prefix', () => {
+    const lines = computeDiff('old', '')
+    const output = renderDiffTerminal(lines)
+    expect(output).toContain('\x1b[38;2;255;80;80m')
+    expect(output).toContain('- 1  old')
+    expect(output).toContain('\x1b[0m')
+  })
+
+  it('renders insert lines in green with + prefix', () => {
+    const lines = computeDiff('', 'new')
+    const output = renderDiffTerminal(lines)
+    expect(output).toContain('\x1b[38;2;0;200;100m')
+    expect(output).toContain('+ 1  new')
+  })
+
+  it('renders equal lines in dim with space prefix', () => {
+    const lines = computeDiff('same', 'same')
+    const output = renderDiffTerminal(lines)
+    expect(output).toContain('\x1b[38;2;100;100;100m')
+    expect(output).toContain('  same')
+    expect(output).toContain('\x1b[0m')
+  })
+})
+
+describe('generateDiff', () => {
+  it('returns terminal-colored unified diff with line numbers', () => {
+    const output = generateDiff('old', 'new')
+    expect(output).toContain('\x1b[38;2;255;80;80m')
+    expect(output).toContain('- 1  old')
+    expect(output).toContain('\x1b[38;2;0;200;100m')
+    expect(output).toContain('+ 1  new')
+    expect(output).toContain('1')
+  })
+
+  it('returns terminal-colored unified diff with custom start line number', () => {
+    const output = generateDiff('old', 'new', 10)
+    expect(output).toContain('- 10  old')
+    expect(output).toContain('+ 10  new')
+  })
+})
+
+describe('escapeHtml', () => {
+  it('escapes ampersands', () => {
+    expect(escapeHtml('a&b')).toBe('a&amp;b')
+  })
+
+  it('escapes less-than', () => {
+    expect(escapeHtml('<tag>')).toBe('&lt;tag&gt;')
+  })
+
+  it('escapes greater-than', () => {
+    expect(escapeHtml('a > b')).toBe('a &gt; b')
+  })
+
+  it('escapes double quotes', () => {
+    expect(escapeHtml('say "hello"')).toBe('say &quot;hello&quot;')
+  })
+
+  it('handles empty string', () => {
+    expect(escapeHtml('')).toBe('')
+  })
+
+  it('handles string with no special chars', () => {
+    expect(escapeHtml('hello world')).toBe('hello world')
+  })
+})
+
+describe('parseMarkdownEdit', () => {
+  it('parses single edit block', () => {
+    const input = '```edit\nfile: src/test.ts\nold: |\n  hello\nnew: |\n  world\n```'
+    const edits = parseMarkdownEdit(input)
+    expect(edits).toHaveLength(1)
+    expect(edits[0]!.file_path).toBe('src/test.ts')
+    // Pipe format preserves leading whitespace
+    expect(edits[0]!.old_string).toBe('  hello')
+    expect(edits[0]!.new_string).toBe('  world')
+  })
+
+  it('parses create file (empty old)', () => {
+    const input = '```edit\nfile: src/new.ts\nold:\nnew: |\n  const x = 1\n```'
+    const edits = parseMarkdownEdit(input)
+    expect(edits).toHaveLength(1)
+    expect(edits[0]!.file_path).toBe('src/new.ts')
+    expect(edits[0]!.old_string).toBe('')
+    // Pipe format preserves leading whitespace
+    expect(edits[0]!.new_string).toBe('  const x = 1')
+  })
+
+  it('parses create file with pipe (old: | empty content)', () => {
+    const input = [
+      '```edit',
+      'file: src/new-pipe.ts',
+      'old: |',
+      'new: |',
+      '  const x = 1',
+      '```',
+    ].join('\n')
+    const edits = parseMarkdownEdit(input)
+    expect(edits).toHaveLength(1)
+    expect(edits[0]!.file_path).toBe('src/new-pipe.ts')
+    expect(edits[0]!.old_string).toBe('')
+    // Pipe format preserves leading whitespace
+    expect(edits[0]!.new_string).toBe('  const x = 1')
+  })
+
+  it('parses multiple edit blocks', () => {
+    const input = [
+      '```edit',
+      'file: a.ts',
+      'old: |',
+      '  foo',
+      'new: |',
+      '  bar',
+      '```',
+      '```edit',
+      'file: b.ts',
+      'old: |',
+      '  baz',
+      'new: |',
+      '  qux',
+      '```',
+    ].join('\n')
+    const edits = parseMarkdownEdit(input)
+    expect(edits).toHaveLength(2)
+    expect(edits[0]!.file_path).toBe('a.ts')
+    // Pipe format preserves leading whitespace
+    expect(edits[0]!.old_string).toBe('  foo')
+    expect(edits[0]!.new_string).toBe('  bar')
+    expect(edits[1]!.file_path).toBe('b.ts')
+    expect(edits[1]!.old_string).toBe('  baz')
+    expect(edits[1]!.new_string).toBe('  qux')
+  })
+
+  it('returns empty array when no edit blocks found', () => {
+    expect(parseMarkdownEdit('no code blocks here')).toEqual([])
+  })
+
+  it('returns empty array when block has no file path', () => {
+    const input = '```edit\nold: |\n  a\nnew: |\n  b\n```'
+    expect(parseMarkdownEdit(input)).toEqual([])
+  })
+
+  it('parses edit block without pipe (single line)', () => {
+    const input = '```edit\nfile: a.ts\nold: hello\nnew: world\n```'
+    const edits = parseMarkdownEdit(input)
+    expect(edits).toHaveLength(1)
+    // Non-pipe format: "old: hello" → content after "old: " includes the leading space
+    expect(edits[0]!.old_string).toBe(' hello')
+    expect(edits[0]!.new_string).toBe(' world')
+  })
+
+  it('parses old with pipe-newline format', () => {
+    const input = '```edit\nfile: a.ts\nold: |\n  multi\n  line\nnew: |\n  replaced\n```'
+    const edits = parseMarkdownEdit(input)
+    expect(edits).toHaveLength(1)
+    // Pipe format preserves leading whitespace (only strips the first \n)
+    expect(edits[0]!.old_string).toBe('  multi\n  line')
+    expect(edits[0]!.new_string).toBe('  replaced')
+  })
+
+  it('handles extra whitespace after ```edit', () => {
+    const input = '```edit  \nfile: a.ts\nold: |\n  x\nnew: |\n  y\n```'
+    const edits = parseMarkdownEdit(input)
+    expect(edits).toHaveLength(1)
+    expect(edits[0]!.file_path).toBe('a.ts')
+  })
+
+  it('ignores non-edit code blocks', () => {
+    const input = '```typescript\nconst x = 1\n```'
+    expect(parseMarkdownEdit(input)).toEqual([])
+  })
+
+  it('handles mixed content with edit blocks', () => {
+    const input = [
+      'some text before',
+      '',
+      '```edit',
+      'file: a.ts',
+      'old: |',
+      '  hello',
+      'new: |',
+      '  hi',
+      '```',
+      'some text after',
+    ].join('\n')
+    const edits = parseMarkdownEdit(input)
+    expect(edits).toHaveLength(1)
+    expect(edits[0]!.file_path).toBe('a.ts')
+  })
+
+  it('handles early-``` closing (new: outside the block)', () => {
+    // Model may close ``` before new: — common mistake
+    const input = [
+      '```edit',
+      'file: a.ts',
+      'old: |',
+      '  hello',
+      '```',
+      'new: |',
+      '  world',
+      '```',
+    ].join('\n')
+    const edits = parseMarkdownEdit(input)
+    expect(edits).toHaveLength(1)
+    expect(edits[0]!.file_path).toBe('a.ts')
+    expect(edits[0]!.old_string).toBe('  hello')
+    expect(edits[0]!.new_string).toBe('  world')
+  })
+
+  it('handles raw file:/old:/new: without ``` markers', () => {
+    // Model outputs edit without ```edit / ``` wrappers
+    const input = 'file: a.ts\nold: |\n  hello\nnew: |\n  world\n'
+    const edits = parseMarkdownEdit(input)
+    expect(edits).toHaveLength(1)
+    expect(edits[0]!.file_path).toBe('a.ts')
+    expect(edits[0]!.old_string).toBe('  hello')
+    expect(edits[0]!.new_string).toBe('  world')
+  })
+
+  it('handles edit block without any ``` markers (raw file:/old:/new: content)', () => {
+    const input = 'file: a.ts\nold: |\n  hello\nnew: |\n  world\n'
+    const edits = parseMarkdownEdit(input)
+    expect(edits).toHaveLength(1)
+    expect(edits[0]!.file_path).toBe('a.ts')
+    expect(edits[0]!.old_string).toBe('  hello')
+    expect(edits[0]!.new_string).toBe('  world')
+  })
+})
+
+// ── Markdown format integration tests ────────────────────────────────────
+describe('edit tool — markdown format', () => {
   let tmpDir: string
   let applier: FileReadTracker
 
   beforeAll(() => {
     tmpDir = makeTempDir()
     fs.writeFileSync(path.join(tmpDir, 'a.txt'), 'line one\nline two\nline three\n')
-    fs.writeFileSync(path.join(tmpDir, 'b.txt'), 'foo\nbar\nbaz\n')
-    applier = new FileReadTracker()
+    applier = makeApplier()
+    applier.markRead(path.join(tmpDir, 'a.txt'))
   })
 
   afterAll(() => {
@@ -22,570 +413,775 @@ describe('edit tool', () => {
 
   const tool = () => createEditTool(applier, tmpDir)
 
-  describe('single edit mode', () => {
-    beforeAll(() => {
-      fs.writeFileSync(path.join(tmpDir, 'a.txt'), 'line one\nline two\nline three\n')
-      fs.writeFileSync(path.join(tmpDir, 'dup.txt'), 'abc\ndef\nabc\n')
-    })
+  it('replaces exact string via markdown format', async () => {
+    const input = '```edit\nfile: a.txt\nold: |\n  line two\nnew: |\n  line TWO\n```'
+    const r = await tool().execute({ content: input })
+    expect(r.success).toBe(true)
+    expect(fs.readFileSync(path.join(tmpDir, 'a.txt'), 'utf8')).toContain('line TWO')
+  })
 
-    it('replaces an exact string', async () => {
-      const r = await tool().execute({
-        edits: [{ file_path: 'a.txt', old_string: 'line two', new_string: 'line TWO' }],
-      })
-      expect(r.success).toBe(true)
-      expect(fs.readFileSync(path.join(tmpDir, 'a.txt'), 'utf8')).toContain('line TWO')
-    })
+  it('replaces multi-line via markdown format', async () => {
+    // Use a fresh file to avoid cross-test contamination
+    const file = path.join(tmpDir, 'multi-md.txt')
+    fs.writeFileSync(file, 'line one\nline two\nline three\n')
+    applier.markRead(file)
 
-    it('replaces multi-line string', async () => {
-      const r = await tool().execute({
-        edits: [
-          {
-            file_path: 'a.txt',
-            old_string: 'line one\nline TWO',
-            new_string: 'line 1\nline 2',
-          },
-        ],
-      })
-      expect(r.success).toBe(true)
-      const content = fs.readFileSync(path.join(tmpDir, 'a.txt'), 'utf8')
-      expect(content).toContain('line 1')
-      expect(content).toContain('line 2')
-    })
+    const input = [
+      '```edit',
+      'file: multi-md.txt',
+      'old: |',
+      '  line one',
+      '  line two',
+      'new: |',
+      '  line 1',
+      '  line 2',
+      '```',
+    ].join('\n')
+    const r = await tool().execute({ content: input })
+    expect(r.success).toBe(true)
+    // Pipe format preserves leading whitespace from the block
+    const content = fs.readFileSync(file, 'utf8')
+    expect(content).toContain('  line 1')
+    expect(content).toContain('  line 2')
+  })
 
-    it('reports old_string not found', async () => {
-      const r = await tool().execute({
-        edits: [{ file_path: 'a.txt', old_string: 'nonexistent', new_string: 'x' }],
-      })
-      expect(r.success).toBe(false)
-      expect(r.error).toContain('not found')
-    })
+  it('creates file via markdown format', async () => {
+    const input = '```edit\nfile: markdown-new.txt\nold:\nnew: |\n  created via markdown\n```'
+    const r = await tool().execute({ content: input })
+    expect(r.success).toBe(true)
+    // Pipe format preserves leading whitespace from the block
+    expect(fs.readFileSync(path.join(tmpDir, 'markdown-new.txt'), 'utf8')).toBe(
+      '  created via markdown',
+    )
+  })
 
-    it('reports duplicate old_string', async () => {
-      const r = await tool().execute({
-        edits: [{ file_path: 'dup.txt', old_string: 'abc', new_string: 'xyz' }],
-      })
-      expect(r.success).toBe(false)
-      expect(r.error).toContain('MULTIPLE times')
-    })
+  it('reports parse error for malformed markdown', async () => {
+    const r = await tool().execute({ content: 'not a valid edit block' })
+    expect(r.success).toBe(false)
+    expect(r.error).toContain('Failed to parse')
+  })
 
-    it('auto-corrects flat file_path/old_string/new_string into edits array', async () => {
-      // The tool auto-corrects { file_path, old_string, new_string }
-      // into { edits: [{ file_path, old_string, new_string }] }
-      // So the error should be about file not found, not 'edits' missing
-      const r = await tool().execute({ file_path: 'x', old_string: 'y', new_string: 'z' })
-      expect(r.success).toBe(false)
-      expect(r.error).toContain('not found')
-    })
+  it('applies multiple markdown blocks', async () => {
+    fs.writeFileSync(path.join(tmpDir, 'multi-a.txt'), 'foo\nbar\n')
+    fs.writeFileSync(path.join(tmpDir, 'multi-b.txt'), 'baz\nqux\n')
+    applier.markRead(path.join(tmpDir, 'multi-a.txt'))
+    applier.markRead(path.join(tmpDir, 'multi-b.txt'))
 
-    it('reports file not found', async () => {
-      const r = await tool().execute({
-        edits: [{ file_path: 'nonexistent.txt', old_string: 'x', new_string: 'y' }],
-      })
-      expect(r.success).toBe(false)
-      expect(r.error).toContain('not found')
-    })
+    const input = [
+      '```edit',
+      'file: multi-a.txt',
+      'old: |',
+      '  foo',
+      'new: |',
+      '  FOO',
+      '```',
+      '```edit',
+      'file: multi-b.txt',
+      'old: |',
+      '  baz',
+      'new: |',
+      '  BAZ',
+      '```',
+    ].join('\n')
+    const r = await tool().execute({ content: input })
+    expect(r.success).toBe(true)
+    expect(fs.readFileSync(path.join(tmpDir, 'multi-a.txt'), 'utf8')).toContain('FOO')
+    expect(fs.readFileSync(path.join(tmpDir, 'multi-b.txt'), 'utf8')).toContain('BAZ')
   })
+})
 
-  describe('validation', () => {
-    beforeAll(() => {
-      fs.writeFileSync(path.join(tmpDir, 'a.txt'), 'test content\n')
-    })
+// ── Advanced edge cases ──────────────────────────────────────────────────
+describe('edit tool — edge cases', () => {
+  let tmpDir: string
+  let applier: FileReadTracker
 
-    it('rejects edit missing old_string', async () => {
-      const r = await tool().execute({
-        edits: [{ file_path: 'a.txt', new_string: 'replacement' }],
-      })
-      expect(r.success).toBe(false)
-      expect(r.error).toContain('missing')
-      expect(r.error).toContain('old_string')
-    })
+  beforeAll(() => {
+    tmpDir = makeTempDir()
+    applier = makeApplier()
+  })
 
-    it('rejects edit missing new_string', async () => {
-      const r = await tool().execute({
-        edits: [{ file_path: 'a.txt', old_string: 'original' }],
-      })
-      expect(r.success).toBe(false)
-      expect(r.error).toContain('missing')
-      expect(r.error).toContain('new_string')
-    })
+  afterAll(() => {
+    fs.rmSync(tmpDir, { recursive: true, force: true })
+  })
+
+  const tool = () => createEditTool(applier, tmpDir)
 
-    it('rejects edit missing file_path', async () => {
+  describe('special characters', () => {
+    it('handles unicode characters', async () => {
+      const file = path.join(tmpDir, 'unicode.txt')
+      fs.writeFileSync(file, 'héllo wörld\n')
+      applier.markRead(file)
       const r = await tool().execute({
-        edits: [{ old_string: 'a', new_string: 'b' }],
+        edits: [{ file_path: 'unicode.txt', old_string: 'héllo wörld', new_string: 'HELLO WORLD' }],
       })
-      expect(r.success).toBe(false)
-      expect(r.error).toContain('missing')
-      expect(r.error).toContain('file_path')
+      expect(r.success).toBe(true)
+      expect(fs.readFileSync(file, 'utf8')).toBe('HELLO WORLD\n')
     })
 
-    it('reports all missing fields in one error', async () => {
+    it('handles emoji characters', async () => {
+      const file = path.join(tmpDir, 'emoji.txt')
+      fs.writeFileSync(file, 'hello 👍 world\n')
+      applier.markRead(file)
       const r = await tool().execute({
         edits: [
-          { file_path: 'a.txt', old_string: 'a', new_string: 'b' },
-          { file_path: 'a.txt', new_string: 'c' },
-          { file_path: 'a.txt', old_string: 'd' },
+          { file_path: 'emoji.txt', old_string: 'hello 👍 world', new_string: 'hello world' },
         ],
       })
-      expect(r.success).toBe(false)
-      // Should report BOTH errors, not just the first
-      expect(r.error).toContain('2 of 3')
-      expect(r.error).toContain('edit #2')
-      expect(r.error).toContain('edit #3')
-      // Should mention not to split across edits
-      expect(r.error).toContain('COMPLETE')
+      expect(r.success).toBe(true)
+      expect(fs.readFileSync(file, 'utf8')).toBe('hello world\n')
     })
 
-    it('reports which fields each malformed edit has', async () => {
+    it('handles Chinese characters', async () => {
+      const file = path.join(tmpDir, 'chinese.txt')
+      fs.writeFileSync(file, '你好世界\n')
+      applier.markRead(file)
       const r = await tool().execute({
-        edits: [{ file_path: 'a.txt', new_string: 'c' }],
+        edits: [{ file_path: 'chinese.txt', old_string: '你好世界', new_string: 'Hello World' }],
       })
-      expect(r.success).toBe(false)
-      expect(r.error).toContain('has: file_path, new_string')
+      expect(r.success).toBe(true)
+      expect(fs.readFileSync(file, 'utf8')).toBe('Hello World\n')
     })
 
-    it('includes rawInput in the error message', async () => {
+    it('handles tab characters', async () => {
+      const file = path.join(tmpDir, 'tabs.txt')
+      fs.writeFileSync(file, 'a\tb\nc\td\n')
+      applier.markRead(file)
       const r = await tool().execute({
-        edits: [{ file_path: 'a.txt', new_string: 'c' }],
+        edits: [{ file_path: 'tabs.txt', old_string: 'a\tb', new_string: 'A\tB' }],
       })
-      expect(r.success).toBe(false)
-      expect(r.error).toContain('a.txt')
-      expect(r.error).toContain('new_string')
+      expect(r.success).toBe(true)
+      expect(fs.readFileSync(file, 'utf8')).toBe('A\tB\nc\td\n')
     })
   })
 
-  describe('auto-correction', () => {
-    it('wraps top-level file_path/old_string/new_string into edits array', async () => {
+  describe('empty and boundary files', () => {
+    it('edits a file with single line', async () => {
+      const file = path.join(tmpDir, 'single.txt')
+      fs.writeFileSync(file, 'only line\n')
+      applier.markRead(file)
       const r = await tool().execute({
-        file_path: 'nonexistent.ts',
-        old_string: 'x',
-        new_string: 'y',
+        edits: [{ file_path: 'single.txt', old_string: 'only line', new_string: 'replaced line' }],
       })
-      // Should be auto-corrected so the error is about file not found, not about missing edits
-      expect(r.success).toBe(false)
-      expect(r.error).toContain('not found')
+      expect(r.success).toBe(true)
+      expect(fs.readFileSync(file, 'utf8')).toBe('replaced line\n')
     })
 
-    it('wraps flat file_path + old_string (missing new_string) with default empty new_string', async () => {
+    it('edits a file with no trailing newline', async () => {
+      const file = path.join(tmpDir, 'no-nl.txt')
+      fs.writeFileSync(file, 'line one\nline two')
+      applier.markRead(file)
       const r = await tool().execute({
-        file_path: path.join(tmpDir, 'new.txt'),
-        old_string: '',
+        edits: [{ file_path: 'no-nl.txt', old_string: 'line two', new_string: 'line TWO' }],
       })
-      // Auto-corrected to { edits: [{ file_path, old_string: '', new_string: '' }] }
-      // This creates a file with empty content
-      expect(r.success).toBe(true)
-      expect(fs.existsSync(path.join(tmpDir, 'new.txt'))).toBe(true)
-      expect(fs.readFileSync(path.join(tmpDir, 'new.txt'), 'utf8')).toBe('')
-      fs.unlinkSync(path.join(tmpDir, 'new.txt'))
-    })
-
-    it('handles edits passed directly as an array (not wrapped)', async () => {
-      // @ts-expect-error — runtime accepts array, TS expects object
-      const r = await tool().execute([
-        { file_path: 'a.txt', old_string: 'test', new_string: 'TEST' },
-      ])
       expect(r.success).toBe(true)
+      expect(fs.readFileSync(file, 'utf8')).toBe('line one\nline TWO')
     })
 
-    it('handles file_path + new_string (no old_string) at top level as create', async () => {
+    it('appends to end of file', async () => {
+      const file = path.join(tmpDir, 'append.txt')
+      fs.writeFileSync(file, 'line one\n')
+      applier.markRead(file)
       const r = await tool().execute({
-        file_path: path.join(tmpDir, 'auto-create.txt'),
-        new_string: 'auto created',
+        edits: [
+          {
+            file_path: 'append.txt',
+            old_string: 'line one',
+            new_string: 'line one\nline two',
+          },
+        ],
       })
       expect(r.success).toBe(true)
-      expect(fs.readFileSync(path.join(tmpDir, 'auto-create.txt'), 'utf8')).toBe('auto created')
-      fs.unlinkSync(path.join(tmpDir, 'auto-create.txt'))
+      expect(fs.readFileSync(file, 'utf8')).toBe('line one\nline two\n')
     })
 
-    it('handles only file_path at top level as empty file creation', async () => {
+    it('replaces entire file content', async () => {
+      const file = path.join(tmpDir, 'full-replace.txt')
+      fs.writeFileSync(file, 'old content\n')
+      applier.markRead(file)
       const r = await tool().execute({
-        file_path: path.join(tmpDir, 'empty-file.txt'),
+        edits: [
+          {
+            file_path: 'full-replace.txt',
+            old_string: 'old content',
+            new_string: 'brand new content',
+          },
+        ],
       })
       expect(r.success).toBe(true)
-      expect(fs.existsSync(path.join(tmpDir, 'empty-file.txt'))).toBe(true)
-      expect(fs.readFileSync(path.join(tmpDir, 'empty-file.txt'), 'utf8')).toBe('')
-      fs.unlinkSync(path.join(tmpDir, 'empty-file.txt'))
-    })
-
-    it('rejects nonsensical top-level keys with helpful error', async () => {
-      const r = await tool().execute({ foo: 'bar' })
-      expect(r.success).toBe(false)
-      expect(r.error).toContain('edits')
-      expect(r.error).toContain('array')
+      expect(fs.readFileSync(file, 'utf8')).toBe('brand new content\n')
     })
   })
 
-  describe('batch mode', () => {
-    beforeAll(() => {
-      fs.writeFileSync(path.join(tmpDir, 'a.txt'), 'one\ntwo\nthree\nfour\n')
-      fs.writeFileSync(path.join(tmpDir, 'b.txt'), 'foo\nbar\nbaz\n')
-    })
-
-    it('applies multiple edits to one file', async () => {
+  describe('file path edge cases', () => {
+    it('creates file in subdirectory that does not exist', async () => {
       const r = await tool().execute({
         edits: [
-          { file_path: 'a.txt', old_string: 'one', new_string: 'ONE' },
-          { file_path: 'a.txt', old_string: 'three', new_string: 'THREE' },
+          {
+            file_path: 'deep/nested/dir/new-file.txt',
+            old_string: '',
+            new_string: 'created in nested dir',
+          },
         ],
       })
       expect(r.success).toBe(true)
-      const content = fs.readFileSync(path.join(tmpDir, 'a.txt'), 'utf8')
-      expect(content).toContain('ONE')
-      expect(content).toContain('THREE')
-      expect(content).toContain('two')
+      expect(fs.readFileSync(path.join(tmpDir, 'deep/nested/dir/new-file.txt'), 'utf8')).toBe(
+        'created in nested dir',
+      )
     })
 
-    it('applies edits across multiple files', async () => {
+    it('handles file path with special characters', async () => {
       const r = await tool().execute({
         edits: [
-          { file_path: 'a.txt', old_string: 'ONE', new_string: 'one' },
-          { file_path: 'b.txt', old_string: 'foo\nbar', new_string: 'FOO\nBAR' },
+          {
+            file_path: 'file-with-dashes_and_underscores.txt',
+            old_string: '',
+            new_string: 'special path',
+          },
         ],
       })
       expect(r.success).toBe(true)
-      expect(fs.readFileSync(path.join(tmpDir, 'a.txt'), 'utf8')).toContain('one')
-      expect(fs.readFileSync(path.join(tmpDir, 'b.txt'), 'utf8')).toContain('FOO')
-      expect(fs.readFileSync(path.join(tmpDir, 'b.txt'), 'utf8')).toContain('BAR')
+      expect(
+        fs.readFileSync(path.join(tmpDir, 'file-with-dashes_and_underscores.txt'), 'utf8'),
+      ).toBe('special path')
     })
+  })
 
-    it('rolls back all changes on failure', async () => {
-      const original = fs.readFileSync(path.join(tmpDir, 'a.txt'), 'utf8')
-
+  describe('error message quality', () => {
+    it('includes proximity hint when old_string not found', async () => {
+      const file = path.join(tmpDir, 'proximity.txt')
+      fs.writeFileSync(file, 'keep this line\nfind me\nkeep that line\n')
+      applier.markRead(file)
       const r = await tool().execute({
         edits: [
-          { file_path: 'a.txt', old_string: 'THREE', new_string: 'three' },
-          { file_path: 'a.txt', old_string: 'zzz_nonexistent_zzz', new_string: 'FAIL' },
+          {
+            file_path: 'proximity.txt',
+            old_string: 'find me\nnonexistent',
+            new_string: 'x',
+          },
         ],
       })
       expect(r.success).toBe(false)
-      expect(r.error).toContain('rolled back')
-      expect(fs.readFileSync(path.join(tmpDir, 'a.txt'), 'utf8')).toBe(original)
+      // Should include proximity info
+      expect(r.error).toContain('Near line')
+      expect(r.error).toContain('find me')
     })
 
-    it('rolls back multi-file on any failure', async () => {
-      const origA = fs.readFileSync(path.join(tmpDir, 'a.txt'), 'utf8')
-      const origB = fs.readFileSync(path.join(tmpDir, 'b.txt'), 'utf8')
-
+    it('includes file content preview when no close match found', async () => {
+      const file = path.join(tmpDir, 'preview.txt')
+      fs.writeFileSync(file, 'alpha\nbeta\ngamma\n')
+      applier.markRead(file)
       const r = await tool().execute({
         edits: [
-          { file_path: 'a.txt', old_string: 'four', new_string: 'FOUR' },
-          { file_path: 'missing.txt', old_string: 'x', new_string: 'y' },
+          {
+            file_path: 'preview.txt',
+            old_string: 'zzz_nonexistent_zzz',
+            new_string: 'x',
+          },
         ],
       })
       expect(r.success).toBe(false)
-      expect(fs.readFileSync(path.join(tmpDir, 'a.txt'), 'utf8')).toBe(origA)
-      expect(fs.readFileSync(path.join(tmpDir, 'b.txt'), 'utf8')).toBe(origB)
-    })
-
-    it('rejects empty edits array', async () => {
-      const r = await tool().execute({ edits: [] })
-      expect(r.success).toBe(false)
-      expect(r.error).toContain('empty')
-    })
-  })
-
-  describe('diff output', () => {
-    beforeAll(() => {
-      fs.writeFileSync(path.join(tmpDir, 'diff-test.txt'), 'line one\nline two\nline three\n')
+      // Should include file content preview
+      expect(r.error).toContain('File content')
+      expect(r.error).toContain('alpha')
+      expect(r.error).toContain('beta')
     })
 
-    it('shows green + for new file creation (no red - lines)', async () => {
-      const r = await tool().execute({
-        edits: [{ file_path: 'diff-new.txt', old_string: '', new_string: 'hello\nworld' }],
-      })
-      expect(r.success).toBe(true)
-      // Should contain green ANSI code for added lines
-      expect(r.output).toContain('\x1b[38;2;0;200;100m') // green
-      expect(r.output).toContain('hello')
-      expect(r.output).toContain('world')
-      // Should NOT contain red ANSI code (no removed lines)
-      expect(r.output).not.toContain('\x1b[38;2;255;80;80m')
-      expect(r.output).toContain('Created')
-    })
-
-    it('shows red - for removed lines and green + for added lines on replacement', async () => {
+    it('includes read warning when file not read first', async () => {
+      // Don't call markRead — simulate stale content
+      const file = path.join(tmpDir, 'no-read.txt')
+      fs.writeFileSync(file, 'content\n')
       const r = await tool().execute({
         edits: [
           {
-            file_path: 'diff-test.txt',
-            old_string: 'line two',
-            new_string: 'line TWO',
+            file_path: 'no-read.txt',
+            old_string: 'zzz',
+            new_string: 'x',
           },
         ],
       })
-      expect(r.success).toBe(true)
-      // Should contain red ANSI code for removed content
-      expect(r.output).toContain('\x1b[38;2;255;80;80m')
-      expect(r.output).toContain('line two')
-      // Should contain green ANSI code for added content
-      expect(r.output).toContain('\x1b[38;2;0;200;100m')
-      expect(r.output).toContain('line TWO')
-      expect(r.output).toContain('Edited')
+      expect(r.success).toBe(false)
+      expect(r.error).toContain('was not read first')
     })
 
-    it('shows multi-line diff correctly', async () => {
+    it('includes rawInput in error for easy debugging', async () => {
       const r = await tool().execute({
         edits: [
           {
-            file_path: 'diff-test.txt',
-            old_string: 'line one\nline TWO',
-            new_string: 'line 1\nline 2',
+            file_path: 'nonexistent.txt',
+            old_string: 'hello',
+            new_string: 'world',
           },
         ],
       })
-      expect(r.success).toBe(true)
-      // Removed lines (red)
-      expect(r.output).toContain('\x1b[38;2;255;80;80mline one')
-      expect(r.output).toContain('\x1b[38;2;255;80;80mline TWO')
-      // Added lines (green)
-      expect(r.output).toContain('\x1b[38;2;0;200;100mline 1')
-      expect(r.output).toContain('\x1b[38;2;0;200;100mline 2')
+      expect(r.success).toBe(false)
+      expect(r.error).toContain('nonexistent.txt')
+      expect(r.error).toContain('hello')
+      expect(r.error).toContain('world')
     })
+  })
 
-    it('handles CRLF in old_string (Windows compatibility)', async () => {
-      fs.writeFileSync(path.join(tmpDir, 'crlf-test.txt'), 'a\\r\\nb\\r\\nc\\r\\n')
+  describe('output format', () => {
+    it('prefixes output with "Edited" for modifications', async () => {
+      const file = path.join(tmpDir, 'output-test.txt')
+      fs.writeFileSync(file, 'original\n')
+      applier.markRead(file)
       const r = await tool().execute({
-        edits: [
-          {
-            file_path: 'crlf-test.txt',
-            // Simulate AI sending old_string with \\r\\n (CRLF) on Windows
-            old_string: 'b\\r\\n',
-            new_string: 'B\\r\\n',
-          },
-        ],
+        edits: [{ file_path: 'output-test.txt', old_string: 'original', new_string: 'modified' }],
       })
       expect(r.success).toBe(true)
-      const content = fs.readFileSync(path.join(tmpDir, 'crlf-test.txt'), 'utf8')
-      expect(content).toContain('B')
-      // Should show the normalized diff (with \\n, not \\r\\n)
-      expect(r.output).toContain('\x1b[38;2;255;80;80mb')
-      expect(r.output).toContain('\x1b[38;2;0;200;100mB')
+      expect(r.output).toContain('Edited')
     })
 
-    it('does not show empty red line when old_string is empty', async () => {
+    it('prefixes output with "Created" for new files', async () => {
       const r = await tool().execute({
-        edits: [{ file_path: 'empty-old.txt', old_string: '', new_string: 'only added' }],
+        edits: [{ file_path: 'output-create.txt', old_string: '', new_string: 'new file' }],
       })
       expect(r.success).toBe(true)
-      // Should have green colored line
-      expect(r.output).toContain('\x1b[38;2;0;200;100monly added')
-      // Count red-start markers — should be 0 (no removed content)
-      const redCount = (r.output.match(/\x1b\[38;2;255;80;80m/g) || []).length
-      expect(redCount).toBe(0)
-    })
-  })
-
-  describe('create mode', () => {
-    const createdDir = () => path.join(tmpDir, 'created-test')
-
-    afterAll(() => {
-      fs.rmSync(createdDir(), { recursive: true, force: true })
+      expect(r.output).toContain('Created')
     })
 
-    it('creates a new file with empty old_string', async () => {
+    it('includes line count for created files', async () => {
       const r = await tool().execute({
         edits: [
           {
-            file_path: 'created-test/new.ts',
+            file_path: 'output-count.txt',
             old_string: '',
-            new_string: 'const x = 1\nexport { x }',
+            new_string: 'line1\nline2\nline3',
           },
         ],
       })
       expect(r.success).toBe(true)
-      const content = fs.readFileSync(path.join(createdDir(), 'new.ts'), 'utf8')
-      expect(content).toBe('const x = 1\nexport { x }')
+      expect(r.output).toContain('3 lines')
     })
+  })
 
-    it('creates a file in batch mode', async () => {
+  describe('concurrent edits to same line area', () => {
+    it('applies edits bottom-to-top within same file', async () => {
+      const file = path.join(tmpDir, 'concurrent.txt')
+      fs.writeFileSync(file, 'first\nsecond\nthird\n')
+      applier.markRead(file)
       const r = await tool().execute({
         edits: [
-          { file_path: 'created-test/a.ts', old_string: '', new_string: '// a' },
-          { file_path: 'created-test/b.ts', old_string: '', new_string: '// b' },
+          { file_path: 'concurrent.txt', old_string: 'first', new_string: 'FIRST' },
+          { file_path: 'concurrent.txt', old_string: 'third', new_string: 'THIRD' },
         ],
       })
       expect(r.success).toBe(true)
-      expect(fs.readFileSync(path.join(createdDir(), 'a.ts'), 'utf8')).toBe('// a')
-      expect(fs.readFileSync(path.join(createdDir(), 'b.ts'), 'utf8')).toBe('// b')
+      const content = fs.readFileSync(file, 'utf8')
+      expect(content).toBe('FIRST\nsecond\nTHIRD\n')
     })
 
-    it('rejects create when file already exists', async () => {
+    it('applies edits bottom-to-top to avoid line offset issues', async () => {
+      const file = path.join(tmpDir, 'offset.txt')
+      fs.writeFileSync(file, 'a\nb\nc\n')
+      applier.markRead(file)
+      // If applied top-to-bottom: after replacing "a" with "A1\nA2",
+      // "c" would be at line 4, not line 2
       const r = await tool().execute({
         edits: [
-          {
-            file_path: 'created-test/new.ts',
-            old_string: '',
-            new_string: 'x',
-          },
+          { file_path: 'offset.txt', old_string: 'a', new_string: 'A1\nA2' },
+          { file_path: 'offset.txt', old_string: 'c', new_string: 'C' },
         ],
       })
-      expect(r.success).toBe(false)
-      expect(r.error).toContain('already exists')
+      expect(r.success).toBe(true)
+      const content = fs.readFileSync(file, 'utf8')
+      expect(content).toBe('A1\nA2\nb\nC\n')
     })
+  })
 
-    it('rolls back on create failure in batch', async () => {
+  describe('CRLF handling', () => {
+    it('handles CRLF in file content', async () => {
+      const file = path.join(tmpDir, 'crlf-content.txt')
+      fs.writeFileSync(file, 'hello\r\nworld\r\n')
+      applier.markRead(file)
       const r = await tool().execute({
         edits: [
-          { file_path: 'created-test/rollback-1.ts', old_string: '', new_string: 'keep me' },
-          { file_path: 'created-test/rollback-1.ts', old_string: '', new_string: 'duplicate' },
+          { file_path: 'crlf-content.txt', old_string: 'hello\nworld', new_string: 'HELLO\nWORLD' },
         ],
       })
-      expect(r.success).toBe(false)
-      expect(fs.existsSync(path.join(createdDir(), 'rollback-1.ts'))).toBe(false)
+      expect(r.success).toBe(true)
+      const content = fs.readFileSync(file, 'utf8')
+      expect(content).toBe('HELLO\nWORLD\n')
     })
   })
 
-  describe('whitespace tolerance', () => {
-    const wsDir = () => path.join(tmpDir, 'ws-test')
-    const f = (name: string) => path.join(wsDir(), name)
-
-    beforeAll(() => {
-      fs.mkdirSync(wsDir(), { recursive: true })
-    })
-
-    const tool = () => createEditTool(applier, wsDir())
+  describe('rollback behavior', () => {
+    it('restores original file content on failure', async () => {
+      const file = path.join(tmpDir, 'rollback-restore.txt')
+      fs.writeFileSync(file, 'original content\n')
+      applier.markRead(file)
+      const original = fs.readFileSync(file, 'utf8')
 
-    it('handles trailing space in file (AI omits trailing space)', async () => {
-      fs.writeFileSync(f('trailing.txt'), 'hello \nworld\n')
       const r = await tool().execute({
         edits: [
-          { file_path: 'trailing.txt', old_string: 'hello\nworld', new_string: 'HELLO\nWORLD' },
+          {
+            file_path: 'rollback-restore.txt',
+            old_string: 'original content',
+            new_string: 'changed',
+          },
+          { file_path: 'rollback-restore.txt', old_string: 'nonexistent', new_string: 'fail' },
         ],
       })
-      expect(r.success).toBe(true)
-      expect(fs.readFileSync(f('trailing.txt'), 'utf8')).toBe('HELLO\nWORLD\n')
-      // Should indicate whitespace-normalized in output
-      expect(r.output).toContain('whitespace-normalized')
+      expect(r.success).toBe(false)
+      expect(r.error).toContain('rolled back')
+      // File should be restored to original
+      expect(fs.readFileSync(file, 'utf8')).toBe(original)
     })
 
-    it('handles trailing space in old_string (AI adds extra trailing space)', async () => {
-      fs.writeFileSync(f('trailing.txt'), 'hello \nworld\n')
+    it('deletes newly created file on rollback', async () => {
       const r = await tool().execute({
         edits: [
-          { file_path: 'trailing.txt', old_string: 'hello \nworld', new_string: 'HELLO\nWORLD' },
+          { file_path: 'rollback-new.txt', old_string: '', new_string: 'new file' },
+          { file_path: 'rollback-new.txt', old_string: '', new_string: 'duplicate' },
         ],
       })
-      expect(r.success).toBe(true)
-      expect(fs.readFileSync(f('trailing.txt'), 'utf8')).toBe('HELLO\nWORLD\n')
+      expect(r.success).toBe(false)
+      expect(fs.existsSync(path.join(tmpDir, 'rollback-new.txt'))).toBe(false)
     })
+  })
 
-    it('handles leading space in file (AI omits leading indent)', async () => {
-      fs.writeFileSync(f('leading.txt'), '  hello\nworld\n')
+  describe('multi-file scenarios', () => {
+    beforeAll(() => {
+      fs.writeFileSync(path.join(tmpDir, 'mf-a.txt'), 'edit me\n')
+      fs.writeFileSync(path.join(tmpDir, 'mf-b.txt'), 'edit me too\n')
+      applier.markRead(path.join(tmpDir, 'mf-a.txt'))
+      applier.markRead(path.join(tmpDir, 'mf-b.txt'))
+    })
+
+    it('edits multiple existing files in one call', async () => {
       const r = await tool().execute({
         edits: [
-          { file_path: 'leading.txt', old_string: 'hello\nworld', new_string: 'HELLO\nWORLD' },
+          { file_path: 'mf-a.txt', old_string: 'edit me', new_string: 'edited' },
+          { file_path: 'mf-b.txt', old_string: 'edit me too', new_string: 'edited too' },
         ],
       })
       expect(r.success).toBe(true)
-      // Leading spaces in the original file become part of the matched block
-      // and are preserved in the result
-      expect(fs.readFileSync(f('leading.txt'), 'utf8')).toBe('  HELLO\nWORLD\n')
+      expect(fs.readFileSync(path.join(tmpDir, 'mf-a.txt'), 'utf8')).toBe('edited\n')
+      expect(fs.readFileSync(path.join(tmpDir, 'mf-b.txt'), 'utf8')).toBe('edited too\n')
     })
 
-    it('handles leading space in old_string (AI adds extra indent)', async () => {
-      fs.writeFileSync(f('leading.txt'), '  hello\nworld\n')
+    it('mixes create and edit across files', async () => {
       const r = await tool().execute({
         edits: [
-          { file_path: 'leading.txt', old_string: '    hello\nworld', new_string: 'HELLO\nWORLD' },
+          { file_path: 'mf-a.txt', old_string: 'edited', new_string: 'edited again' },
+          { file_path: 'mf-create-new.txt', old_string: '', new_string: 'newly created' },
         ],
       })
       expect(r.success).toBe(true)
-      // The file's leading spaces are part of the matched block, so they
-      // get replaced along with the rest of the matched text
-      expect(fs.readFileSync(f('leading.txt'), 'utf8')).toBe('HELLO\nWORLD\n')
+      expect(fs.readFileSync(path.join(tmpDir, 'mf-a.txt'), 'utf8')).toBe('edited again\n')
+      expect(fs.readFileSync(path.join(tmpDir, 'mf-create-new.txt'), 'utf8')).toBe('newly created')
     })
 
-    it('handles extra internal spaces in file (AI sends single space)', async () => {
-      fs.writeFileSync(f('internal.txt'), 'foo  bar\nbaz\n')
+    it('rolls back mixed create+edit on failure', async () => {
+      const origA = fs.readFileSync(path.join(tmpDir, 'mf-a.txt'), 'utf8')
+
       const r = await tool().execute({
         edits: [
-          { file_path: 'internal.txt', old_string: 'foo bar\nbaz', new_string: 'FOO BAR\nBAZ' },
+          { file_path: 'mf-a.txt', old_string: 'edited again', new_string: 'changed' },
+          { file_path: 'mf-rollback-new.txt', old_string: '', new_string: 'will be deleted' },
+          { file_path: 'mf-nonexistent.txt', old_string: 'nope', new_string: 'x' },
         ],
       })
-      expect(r.success).toBe(true)
-      expect(fs.readFileSync(f('internal.txt'), 'utf8')).toBe('FOO BAR\nBAZ\n')
+      expect(r.success).toBe(false)
+      expect(r.error).toContain('rolled back')
+      // Existing file restored
+      expect(fs.readFileSync(path.join(tmpDir, 'mf-a.txt'), 'utf8')).toBe(origA)
+      // Created file should be deleted
+      expect(fs.existsSync(path.join(tmpDir, 'mf-rollback-new.txt'))).toBe(false)
     })
 
-    it('handles extra internal spaces in old_string (AI adds double space)', async () => {
-      fs.writeFileSync(f('internal.txt'), 'foo  bar\nbaz\n')
+    it('creates multiple files in subdirectories', async () => {
       const r = await tool().execute({
         edits: [
-          { file_path: 'internal.txt', old_string: 'foo  bar\nbaz', new_string: 'FOO BAR\nBAZ' },
+          { file_path: 'multi-dir/sub1/x.ts', old_string: '', new_string: '// x' },
+          { file_path: 'multi-dir/sub2/y.ts', old_string: '', new_string: '// y' },
         ],
       })
       expect(r.success).toBe(true)
-      expect(fs.readFileSync(f('internal.txt'), 'utf8')).toBe('FOO BAR\nBAZ\n')
+      expect(fs.readFileSync(path.join(tmpDir, 'multi-dir/sub1/x.ts'), 'utf8')).toBe('// x')
+      expect(fs.readFileSync(path.join(tmpDir, 'multi-dir/sub2/y.ts'), 'utf8')).toBe('// y')
     })
 
-    it('handles blank line with spaces matching empty blank line', async () => {
-      fs.writeFileSync(f('blank.txt'), 'foo\n   \nbar\n')
-      const r = await tool().execute({
-        edits: [{ file_path: 'blank.txt', old_string: 'foo\n\nbar', new_string: 'FOO\n\nBAR' }],
-      })
+    it('creates multiple files with markdown format', async () => {
+      const input = [
+        '```edit',
+        'file: md-multi-1.txt',
+        'old:',
+        'new: |',
+        '  first file',
+        '```',
+        '```edit',
+        'file: md-multi-2.txt',
+        'old:',
+        'new: |',
+        '  second file',
+        '```',
+      ].join('\n')
+      const r = await tool().execute({ content: input })
       expect(r.success).toBe(true)
-      // The blank line with spaces is part of the matched block and gets
-      // replaced entirely by the new_string's blank line
-      expect(fs.readFileSync(f('blank.txt'), 'utf8')).toBe('FOO\n\nBAR\n')
+      expect(fs.readFileSync(path.join(tmpDir, 'md-multi-1.txt'), 'utf8')).toBe('  first file')
+      expect(fs.readFileSync(path.join(tmpDir, 'md-multi-2.txt'), 'utf8')).toBe('  second file')
     })
 
-    it('handles mixed whitespace across multiple lines', async () => {
-      fs.writeFileSync(f('mixed.txt'), '  a  \nb \nc\n')
-      const r = await tool().execute({
-        edits: [{ file_path: 'mixed.txt', old_string: 'a\nb\nc', new_string: 'A\nB\nC' }],
-      })
-      expect(r.success).toBe(true)
-      expect(fs.readFileSync(f('mixed.txt'), 'utf8')).toBe('A\nB\nC\n')
-    })
+    it('rolls back multi-file via markdown format on failure', async () => {
+      const file = path.join(tmpDir, 'md-rollback-target.txt')
+      fs.writeFileSync(file, 'preserve me\n')
+      applier.markRead(file)
 
-    it('still fails when text content differs (not just whitespace)', async () => {
-      fs.writeFileSync(f('trailing.txt'), 'hello \nworld\n')
-      const r = await tool().execute({
-        edits: [{ file_path: 'trailing.txt', old_string: 'hello\nmundo', new_string: 'x\ny' }],
-      })
+      const input = [
+        '```edit',
+        'file: md-rollback-target.txt',
+        'old: |',
+        '  preserve me',
+        'new: |',
+        '  changed',
+        '```',
+        '```edit',
+        'file: md-nonexistent.txt',
+        'old: |',
+        '  not there',
+        'new: |',
+        '  fail',
+        '```',
+      ].join('\n')
+      const r = await tool().execute({ content: input })
       expect(r.success).toBe(false)
-      expect(r.error).toContain('not found')
+      expect(r.error).toContain('rolled back')
+      // Target file should be restored
+      expect(fs.readFileSync(file, 'utf8')).toBe('preserve me\n')
     })
+  })
+})
 
-    it('still fails for completely wrong old_string', async () => {
-      fs.writeFileSync(f('trailing.txt'), 'hello \nworld\n')
-      const r = await tool().execute({
-        edits: [{ file_path: 'trailing.txt', old_string: 'zzz\nzzz', new_string: 'x\ny' }],
-      })
-      expect(r.success).toBe(false)
-      expect(r.error).toContain('not found')
-    })
+// ── Model-generated edit format integration tests ──────────────────────
+// These tests call a real LLM (DeepSeek-V4-Flash via ~/.lonny/config.json)
+// to generate edit format content, then feed it into the edit tool.
 
-    it('reports duplicate in whitespace-normalized mode', async () => {
-      fs.writeFileSync(f('duplicate.txt'), 'a  \nb\na  \nb\n')
-      const r = await tool().execute({
-        edits: [{ file_path: 'duplicate.txt', old_string: 'a\nb', new_string: 'X\nY' }],
-      })
-      expect(r.success).toBe(false)
-      expect(r.error).toContain('MULTIPLE times')
-      expect(r.error).toContain('whitespace-normalized')
-    })
+const LONNY_CONFIG_PATH = path.join(os.homedir(), '.lonny', 'config.json')
 
-    it('prefers exact match when available', async () => {
-      fs.writeFileSync(f('exact-prefer.txt'), 'hello\nworld\n')
-      const r = await tool().execute({
-        edits: [
-          { file_path: 'exact-prefer.txt', old_string: 'hello\nworld', new_string: 'HELLO\nWORLD' },
-        ],
-      })
-      expect(r.success).toBe(true)
-      expect(fs.readFileSync(f('exact-prefer.txt'), 'utf8')).toBe('HELLO\nWORLD\n')
-      // Exact match should NOT have the whitespace-normalized note
-      expect(r.output).not.toContain('whitespace-normalized')
-    })
+interface LonnyConfig {
+  apiKey?: string
+  baseUrl?: string
+  provider?: string
+  model?: string
+}
 
-    it('provides context snippet when old_string not found', async () => {
-      fs.writeFileSync(f('trailing.txt'), 'hello \nworld\n')
-      const r = await tool().execute({
-        edits: [{ file_path: 'trailing.txt', old_string: 'nonexistent_line_xyz', new_string: 'x' }],
-      })
-      expect(r.success).toBe(false)
-      expect(r.error).toContain('not found')
-      // Should contain a snippet with the file's actual content
-      expect(r.error).toContain('hello')
-    })
+function loadLonnyConfig(): LonnyConfig {
+  try {
+    return JSON.parse(fs.readFileSync(LONNY_CONFIG_PATH, 'utf-8'))
+  } catch {
+    return {}
+  }
+}
+
+const lonnyConfig = loadLonnyConfig()
+const hasApiKey = !!lonnyConfig.apiKey
+
+/** Call the model with a prompt and return the full text response (non-streaming). */
+async function callModel(prompt: string, signal?: AbortSignal): Promise<string> {
+  const { default: OpenAI } = await import('openai')
+  const client = new OpenAI({
+    apiKey: lonnyConfig.apiKey,
+    baseURL: lonnyConfig.baseUrl,
+  })
+
+  const response = await client.chat.completions.create(
+    {
+      model: lonnyConfig.model || 'deepseek-v4-flash',
+      messages: [{ role: 'user', content: prompt }],
+      stream: false,
+      max_tokens: 4096,
+    },
+    signal ? { signal } : undefined,
+  )
+
+  return response.choices[0]?.message?.content || ''
+}
+
+describe.runIf(hasApiKey)('model-generated edit format', () => {
+  const MODEL_TIMEOUT = 60_000 // Model API calls can take 10-30s
+  let tmpDir: string
+  let applier: FileReadTracker
+
+  beforeAll(() => {
+    tmpDir = makeTempDir()
+    applier = new FileReadTracker()
+  })
+
+  afterAll(() => {
+    fs.rmSync(tmpDir, { recursive: true, force: true })
+  })
+
+  const tool = () => createEditTool(applier, tmpDir)
+
+  it('model replaces exact string in a file', { timeout: MODEL_TIMEOUT }, async () => {
+    const file = path.join(tmpDir, 'simple.txt')
+    fs.writeFileSync(file, 'hello world\nfoo bar\n')
+    applier.markRead(file)
+
+    const prompt = [
+      'Output ONLY an edit code block (```edit ... ```) that replaces "hello world" with "hello universe" in file "simple.txt".',
+      '',
+      'The EXACT format is:',
+      '```edit',
+      'file: simple.txt',
+      'old: |',
+      '  hello world',
+      'new: |',
+      '  hello universe',
+      '```',
+      '',
+      'Do NOT add quotes, do NOT add backticks around the text inside old/new. Output ONLY the code block.',
+    ].join('\n')
+
+    const modelOutput = await callModel(prompt)
+    const edits = parseMarkdownEdit(modelOutput)
+    expect(edits.length).toBeGreaterThanOrEqual(1)
+
+    const result = await tool().execute({ content: modelOutput })
+    expect(result.success).toBe(true)
+    const content = fs.readFileSync(file, 'utf8')
+    expect(content).toContain('hello universe')
+    expect(content).toContain('foo bar')
+  })
+
+  it('model replaces multi-line text in a file', { timeout: MODEL_TIMEOUT }, async () => {
+    const file = path.join(tmpDir, 'multiline.txt')
+    fs.writeFileSync(file, 'line A\nline B\nline C\nline D\n')
+    applier.markRead(file)
+
+    const prompt = [
+      'Output ONLY an edit code block that replaces:',
+      '',
+      '  line B',
+      '  line C',
+      '',
+      'with:',
+      '',
+      '  line X',
+      '  line Y',
+      '',
+      `in file "multiline.txt".`,
+      '',
+      'Format:',
+      '```edit',
+      'file: multiline.txt',
+      'old: |',
+      '  line B',
+      '  line C',
+      'new: |',
+      '  line X',
+      '  line Y',
+      '```',
+      '',
+      'IMPORTANT: old: must contain EXACTLY "line B" and "line C" on separate lines with 2-space indent.',
+      'Do NOT add quotes around the text. Output ONLY the code block.',
+    ].join('\n')
+
+    const modelOutput = await callModel(prompt)
+    const edits = parseMarkdownEdit(modelOutput)
+    expect(edits.length).toBeGreaterThanOrEqual(1)
+
+    const result = await tool().execute({ content: modelOutput })
+    expect(result.success).toBe(true)
+    const content = fs.readFileSync(file, 'utf8')
+    expect(content).toContain('line X')
+    expect(content).toContain('line Y')
+  })
+
+  it('model creates a new file', { timeout: MODEL_TIMEOUT }, async () => {
+    const prompt = [
+      'Output ONLY an edit code block that CREATES a new file called "fresh.txt" with content "hello world".',
+      '',
+      'Format (note: old: has NO content - this means create):',
+      '```edit',
+      'file: fresh.txt',
+      'old:',
+      'new: |',
+      '  hello world',
+      '```',
+      '',
+      'Output ONLY the code block.',
+    ].join('\n')
+
+    const modelOutput = await callModel(prompt)
+    const edits = parseMarkdownEdit(modelOutput)
+    expect(edits.length).toBeGreaterThanOrEqual(1)
+    expect(edits[0]!.file_path).toBe('fresh.txt')
+
+    const result = await tool().execute({ content: modelOutput })
+    expect(result.success).toBe(true)
+    expect(fs.readFileSync(path.join(tmpDir, 'fresh.txt'), 'utf8')).toContain('hello world')
+  })
+
+  it('model edits a file with special characters (double quotes)', {
+    timeout: MODEL_TIMEOUT,
+  }, async () => {
+    const file = path.join(tmpDir, 'quotes.txt')
+    fs.writeFileSync(file, 'alpha\nbeta\ngamma\n')
+    applier.markRead(file)
+
+    const prompt = [
+      'Output ONLY an edit code block that replaces "beta" with "BETA" in file "quotes.txt".',
+      '',
+      '```edit',
+      'file: quotes.txt',
+      'old: |',
+      '  beta',
+      'new: |',
+      '  BETA',
+      '```',
+      '',
+      'Output ONLY the code block. No extra text.',
+    ].join('\n')
+
+    const modelOutput = await callModel(prompt)
+    const edits = parseMarkdownEdit(modelOutput)
+    expect(edits.length).toBeGreaterThanOrEqual(1)
+
+    const result = await tool().execute({ content: modelOutput })
+    expect(result.success).toBe(true)
+    const content = fs.readFileSync(file, 'utf8')
+    expect(content).toContain('BETA')
+    expect(content).toContain('alpha')
+    expect(content).toContain('gamma')
+  })
+
+  it('model generates edit block that round-trips successfully', {
+    timeout: MODEL_TIMEOUT,
+  }, async () => {
+    const file = path.join(tmpDir, 'roundtrip.txt')
+    fs.writeFileSync(file, 'first\nsecond\nthird\nfourth\n')
+    applier.markRead(file)
+
+    const prompt = [
+      'Output ONLY an edit code block that replaces "second" with "SECOND" in file "roundtrip.txt".',
+      'Include 1 line of context before and after to ensure uniqueness.',
+      '',
+      'Format:',
+      '```edit',
+      'file: roundtrip.txt',
+      'old: |',
+      '  first',
+      '  second',
+      '  third',
+      'new: |',
+      '  first',
+      '  SECOND',
+      '  third',
+      '```',
+      '',
+      'Output ONLY the code block.',
+    ].join('\n')
+
+    const modelOutput = await callModel(prompt)
+    const edits = parseMarkdownEdit(modelOutput)
+    expect(edits.length).toBeGreaterThanOrEqual(1)
+
+    const result = await tool().execute({ content: modelOutput })
+    expect(result.success).toBe(true)
+    const content = fs.readFileSync(file, 'utf8')
+    expect(content).toContain('SECOND')
+    expect(content).toContain('fourth')
   })
 })