code review + speed fix

src/core/tools/__tests__/contextValidator.test.ts

@@ -2,6 +2,7 @@ import { describe, it, expect, vi, beforeEach } from "vitest"
 import { validateFileSizeForContext } from "../contextValidator"
 import { Task } from "../../task/Task"
 import { promises as fs } from "fs"
+import * as fsPromises from "fs/promises"
 import { readLines } from "../../../integrations/misc/read-lines"
 import * as sharedApi from "../../../shared/api"
 
@@ -11,12 +12,17 @@ vi.mock("fs", () => ({
 	},
 }))
 
+vi.mock("fs/promises", () => ({
+	stat: vi.fn(),
+}))
+
 vi.mock("../../../integrations/misc/read-lines", () => ({
 	readLines: vi.fn(),
 }))
 
 vi.mock("../../../shared/api", () => ({
 	getModelMaxOutputTokens: vi.fn(),
+	getFormatForProvider: vi.fn().mockReturnValue("anthropic"),
 }))
 
 describe("contextValidator", () => {
@@ -25,6 +31,11 @@ describe("contextValidator", () => {
 	beforeEach(() => {
 		vi.clearAllMocks()
 
+		// Default file size mock (1MB - large enough to trigger validation)
+		vi.mocked(fs.stat).mockResolvedValue({
+			size: 1024 * 1024, // 1MB
+		} as any)
+
 		// Mock Task instance
 		mockTask = {
 			api: {
@@ -55,26 +66,27 @@ describe("contextValidator", () => {
 	})
 
 	describe("validateFileSizeForContext", () => {
-		it("should apply 25% buffer to remaining context and read incrementally", async () => {
+		it("should apply 25% buffer to remaining context and use character-based reading", async () => {
 			const mockStats = { size: 50000 }
 			vi.mocked(fs.stat).mockResolvedValue(mockStats as any)
 
-			// Mock readLines to return content in batches
-			// Each batch is 100 lines, returning content that results in 1200 tokens per batch
+			// Mock readLines to return content in larger batches (500 lines)
 			vi.mocked(readLines).mockImplementation(async (path, endLine, startLine) => {
 				const start = startLine ?? 0
-				const end = endLine ?? 99
-				const lines = end - start + 1
-				return `test content line\n`.repeat(lines)
+				const end = endLine ?? 499
+				const lines = []
+				for (let i = start; i <= end; i++) {
+					// Each line is ~60 chars to simulate real code
+					lines.push(`const variable${i} = "test content line with enough characters";`)
+				}
+				return lines.join("\n")
 			})
 
-			// Mock token count - 12 tokens per line (1200 per 100-line batch)
-			let callCount = 0
+			// Mock token count based on character count (using ~3 chars per token)
 			mockTask.api.countTokens = vi.fn().mockImplementation(async (content) => {
-				callCount++
 				const text = content[0].text
-				const lines = text.split("\n").length - 1
-				return lines * 12 // 12 tokens per line
+				// Approximate 3 characters per token
+				return Math.ceil(text.length / 3)
 			})
 
 			const result = await validateFileSizeForContext(
@@ -88,13 +100,14 @@ describe("contextValidator", () => {
 			// Context window = 100k, current usage = 10k
 			// Remaining = 90k
 			// With 25% buffer on remaining: usable = 90k * 0.75 = 67.5k
-			// Reserved for response ~2k
-			// Available should be around 65.5k tokens
-			// File needs 12k tokens total (1000 lines * 12 tokens)
+			// Reserved for response = 4096
+			// Available = 67.5k - 4096 ≈ 63.4k tokens
+			// Target limit = 63.4k * 0.9 ≈ 57k tokens
+			// File content: 1000 lines * 60 chars = 60k chars ≈ 20k tokens
 			expect(result.shouldLimit).toBe(false)
 
-			// Verify readLines was called multiple times (incremental reading)
-			expect(readLines).toHaveBeenCalled()
+			// Should make fewer API calls with character-based approach
+			expect(mockTask.api.countTokens).toHaveBeenCalledTimes(1)
 
 			// Verify the new calculation approach
 			const remaining = 100000 - 10000 // 90k remaining
@@ -106,19 +119,24 @@ describe("contextValidator", () => {
 			const mockStats = { size: 50000 }
 			vi.mocked(fs.stat).mockResolvedValue(mockStats as any)
 
-			// Mock readLines
+			// Mock readLines with larger batches
 			vi.mocked(readLines).mockImplementation(async (path, endLine, startLine) => {
 				const start = startLine ?? 0
-				const end = endLine ?? 99
-				const lines = end - start + 1
-				return `test content line\n`.repeat(lines)
+				const end = endLine ?? 499
+				const lines = []
+				for (let i = start; i <= end && i < 2000; i++) {
+					// Dense content - 150 chars per line
+					lines.push(
+						`const longVariable${i} = "This is a much longer line of content to simulate dense code with many characters per line";`,
+					)
+				}
+				return lines.join("\n")
 			})
 
-			// Mock token count - 50 tokens per line
+			// Mock token count based on character count
 			mockTask.api.countTokens = vi.fn().mockImplementation(async (content) => {
 				const text = content[0].text
-				const lines = text.split("\n").length - 1
-				return lines * 50
+				return Math.ceil(text.length / 3)
 			})
 
 			// Test with 50% context already used
@@ -134,12 +152,16 @@ describe("contextValidator", () => {
 			)
 
 			// With 50k remaining and 25% buffer: 50k * 0.75 = 37.5k usable
-			// Minus ~2k for response = ~35.5k available
-			// File needs 100k tokens (2000 lines * 50 tokens)
+			// Minus 4096 for response = ~33.4k available
+			// Target limit = 33.4k * 0.9 ≈ 30k tokens
+			// File content: 2000 lines * 150 chars = 300k chars ≈ 100k tokens
 			// Should limit the file
 			expect(result.shouldLimit).toBe(true)
 			expect(result.safeMaxLines).toBeLessThan(2000)
 			expect(result.reason).toContain("exceeds available context space")
+
+			// Should use character-based approach with fewer API calls
+			expect(mockTask.api.countTokens).toHaveBeenCalled()
 		})
 
 		it("should limit file when it exceeds available space with buffer", async () => {
@@ -147,19 +169,26 @@ describe("contextValidator", () => {
 			const mockStats = { size: 500000 } // Large file
 			vi.mocked(fs.stat).mockResolvedValue(mockStats as any)
 
-			// Mock readLines to return content in batches
+			// Mock readLines to return dense content
 			vi.mocked(readLines).mockImplementation(async (path, endLine, startLine) => {
 				const start = startLine ?? 0
-				const end = endLine ?? 99
-				const lines = end - start + 1
-				return `large content line\n`.repeat(lines)
+				const end = Math.min(endLine ?? 499, start + 499)
+				const lines = []
+				for (let i = start; i <= end && i < 10000; i++) {
+					// Very dense content - 300 chars per line
+					lines.push(
+						`const veryLongVariable${i} = "This is an extremely long line of content that simulates very dense code with many characters, such as minified JavaScript or long string literals that would consume many tokens";`,
+					)
+				}
+				return lines.join("\n")
 			})
 
-			// Mock large token count - 100 tokens per line
+			// Mock token count based on character count
+			let apiCallCount = 0
 			mockTask.api.countTokens = vi.fn().mockImplementation(async (content) => {
+				apiCallCount++
 				const text = content[0].text
-				const lines = text.split("\n").length - 1
-				return lines * 100 // 100 tokens per line
+				return Math.ceil(text.length / 3)
 			})
 
 			const result = await validateFileSizeForContext(
@@ -173,6 +202,9 @@ describe("contextValidator", () => {
 			expect(result.safeMaxLines).toBeGreaterThan(0)
 			expect(result.safeMaxLines).toBeLessThan(10000) // Should stop before reading all lines
 			expect(result.reason).toContain("exceeds available context space")
+
+			// Should make 1-2 API calls with character-based approach
+			expect(apiCallCount).toBeLessThanOrEqual(2)
 		})
 
 		it("should handle very large files through incremental reading", async () => {
@@ -180,19 +212,25 @@ describe("contextValidator", () => {
 			const mockStats = { size: 60_000_000 } // 60MB file
 			vi.mocked(fs.stat).mockResolvedValue(mockStats as any)
 
-			// Mock readLines to return content in batches
+			// Mock readLines to return dense content in larger batches
 			vi.mocked(readLines).mockImplementation(async (path, endLine, startLine) => {
 				const start = startLine ?? 0
-				const end = endLine ?? 99
-				const lines = end - start + 1
-				return `large file content line\n`.repeat(lines)
+				const end = Math.min(endLine ?? 499, start + 499)
+				const lines = []
+				for (let i = start; i <= end && i < 100000; i++) {
+					// Very dense content - 300 chars per line
+					lines.push(
+						`const veryLongVariable${i} = "This is an extremely long line of content that simulates very dense code with many characters, such as minified JavaScript or long string literals that would consume many tokens";`,
+					)
+				}
+				return lines.join("\n")
 			})
 
-			// Mock very high token count per line (simulating dense content)
+			// Mock token count based on character count
 			mockTask.api.countTokens = vi.fn().mockImplementation(async (content) => {
 				const text = content[0].text
-				const lines = text.split("\n").length - 1
-				return lines * 200 // 200 tokens per line for very large file
+				// Return high token count to trigger limit
+				return Math.ceil(text.length / 2) // More tokens per char for dense content
 			})
 
 			const result = await validateFileSizeForContext(
@@ -205,8 +243,9 @@ describe("contextValidator", () => {
 			expect(result.shouldLimit).toBe(true)
 			// Should have attempted to read the file incrementally
 			expect(readLines).toHaveBeenCalled()
-			// Should stop early due to token limits
-			expect(result.safeMaxLines).toBeLessThan(1000)
+			// With character-based approach, it reads more lines before hitting limit
+			expect(result.safeMaxLines).toBeGreaterThan(0)
+			expect(result.safeMaxLines).toBeLessThan(10000) // But still limited
 			expect(result.reason).toContain("exceeds available context space")
 		})
 
@@ -215,7 +254,9 @@ describe("contextValidator", () => {
 			vi.mocked(fs.stat).mockResolvedValue(mockStats as any)
 
 			// Mock readLines to fail
-			vi.mocked(readLines).mockRejectedValue(new Error("Read error"))
+			vi.mocked(readLines).mockImplementation(async () => {
+				throw new Error("Read error")
+			})
 
 			const result = await validateFileSizeForContext(
 				"/test/problematic.ts",
@@ -236,24 +277,26 @@ describe("contextValidator", () => {
 			// Set very high context usage
 			// With new calculation: 100k - 95k = 5k remaining
 			// 5k * 0.75 = 3.75k usable
-			// Minus ~2k for response = ~1.75k available
+			// Minus 4096 for response = negative available space
 			mockTask.getTokenUsage = vi.fn().mockReturnValue({
 				contextTokens: 95000, // 95% of context used
 			})
 
-			// Mock small token count
+			// Mock token count to exceed available space immediately
 			mockTask.api.countTokens = vi.fn().mockImplementation(async (content) => {
-				const text = content[0].text
-				const lines = text.split("\n").length - 1
-				return lines * 10 // 10 tokens per line
+				// Return tokens that exceed available space
+				return 5000 // More than available
 			})
 
 			// Mock readLines
 			vi.mocked(readLines).mockImplementation(async (path, endLine, startLine) => {
 				const start = startLine ?? 0
-				const end = endLine ?? 99
-				const lines = end - start + 1
-				return `test line\n`.repeat(lines)
+				const end = Math.min(endLine ?? 499, start + 499)
+				const lines = []
+				for (let i = start; i <= end && i < 500; i++) {
+					lines.push(`const var${i} = "test line";`)
+				}
+				return lines.join("\n")
 			})
 
 			const result = await validateFileSizeForContext(
@@ -264,10 +307,10 @@ describe("contextValidator", () => {
 			)
 
 			expect(result.shouldLimit).toBe(true)
-			// With the new calculation using full model max tokens (4096),
-			// we have less space available, so we get the minimum 50 lines
-			expect(result.safeMaxLines).toBe(50)
-			expect(result.reason).toContain("Very limited context space")
+			// With the new implementation, when content exceeds limit even after cutback,
+			// it returns a very small number (10) as specified in the safety check
+			expect(result.safeMaxLines).toBe(10)
+			expect(result.reason).toContain("File too large for available context")
 		})
 
 		it("should handle negative available space gracefully", async () => {
@@ -277,11 +320,25 @@ describe("contextValidator", () => {
 			// Set extremely high context usage
 			// With 100k - 99k = 1k remaining
 			// 1k * 0.75 = 750 tokens usable
-			// Minus 2k for response = negative available space
+			// Minus 4096 for response = negative available space
 			mockTask.getTokenUsage = vi.fn().mockReturnValue({
 				contextTokens: 99000, // 99% of context used
 			})
 
+			// Mock token count to always exceed limit
+			mockTask.api.countTokens = vi.fn().mockResolvedValue(10000)
+
+			// Mock readLines
+			vi.mocked(readLines).mockImplementation(async (path, endLine, startLine) => {
+				const start = startLine ?? 0
+				const end = Math.min(endLine ?? 499, start + 499)
+				const lines = []
+				for (let i = start; i <= end && i < 500; i++) {
+					lines.push(`const var${i} = "test line";`)
+				}
+				return lines.join("\n")
+			})
+
 			const result = await validateFileSizeForContext(
 				"/test/smallfile.ts",
 				500, // totalLines
@@ -290,10 +347,9 @@ describe("contextValidator", () => {
 			)
 
 			expect(result.shouldLimit).toBe(true)
-			expect(result.safeMaxLines).toBe(50) // Should be limited to minimum useful lines
-			expect(result.reason).toContain("Very limited context space")
-			// With negative available space, readLines won't be called
-			expect(readLines).not.toHaveBeenCalled()
+			// When available space is negative, it returns minimal safe value
+			expect(result.safeMaxLines).toBe(10) // Minimal safe value from safety check
+			expect(result.reason).toContain("File too large for available context")
 		})
 
 		it("should limit file when it is too large and would be truncated", async () => {
@@ -306,11 +362,19 @@ describe("contextValidator", () => {
 				contextTokens: 90000, // 90% of context used
 			})
 
-			// Mock token counting to simulate a large file
-			mockTask.api.countTokens = vi.fn().mockResolvedValue(1000) // Each batch is 1000 tokens
+			// Mock token counting to exceed limit on first call
+			mockTask.api.countTokens = vi.fn().mockResolvedValue(20000) // Exceeds available space
 
-			// Mock readLines to return some content
-			vi.mocked(readLines).mockResolvedValue("line content")
+			// Mock readLines to return content
+			vi.mocked(readLines).mockImplementation(async (path, endLine, startLine) => {
+				const start = startLine ?? 0
+				const end = Math.min(endLine ?? 499, start + 499)
+				const lines = []
+				for (let i = start; i <= end && i < totalLines; i++) {
+					lines.push(`line content ${i} with enough characters`)
+				}
+				return lines.join("\n")
+			})
 
 			const result = await validateFileSizeForContext(filePath, totalLines, currentMaxReadFileLine, mockTask)
 
@@ -332,15 +396,24 @@ describe("contextValidator", () => {
 			})
 
 			// Mock token counting to quickly exceed limit
-			mockTask.api.countTokens = vi.fn().mockResolvedValue(500) // Each batch uses a lot of tokens
+			mockTask.api.countTokens = vi.fn().mockResolvedValue(5000) // Exceeds available space immediately
 
-			vi.mocked(readLines).mockResolvedValue("line content")
+			vi.mocked(readLines).mockImplementation(async (path, endLine, startLine) => {
+				const start = startLine ?? 0
+				const end = Math.min(endLine ?? 499, start + 499)
+				const lines = []
+				for (let i = start; i <= end && i < totalLines; i++) {
+					lines.push(`line content ${i}`)
+				}
+				return lines.join("\n")
+			})
 
 			const result = await validateFileSizeForContext(filePath, totalLines, currentMaxReadFileLine, mockTask)
 
 			expect(result.shouldLimit).toBe(true)
-			expect(result.reason).toContain("Very limited context space")
-			expect(result.reason).toContain("Consider using search_files or line_range")
+			// With the new implementation, when space is very limited and content exceeds,
+			// it returns the minimal safe value
+			expect(result.reason).toContain("File too large for available context")
 		})
 
 		it("should not limit when file fits within context", async () => {
@@ -351,7 +424,21 @@ describe("contextValidator", () => {
 			// Mock low token usage
 			mockTask.api.countTokens = vi.fn().mockResolvedValue(10) // Small token count per batch
 
-			vi.mocked(readLines).mockResolvedValue("line content")
+			vi.mocked(readLines).mockImplementation(async (path, endLine, startLine) => {
+				const start = startLine ?? 0
+				const end = endLine ?? 0
+
+				// For sampling phase (first 50 lines), return normal length content
+				if (start === 0 && end === 49) {
+					const lines = []
+					for (let i = 0; i <= end; i++) {
+						lines.push(`line content with enough characters to avoid heuristic skip`)
+					}
+					return lines.join("\n")
+				}
+
+				return "line content"
+			})
 
 			const result = await validateFileSizeForContext(filePath, totalLines, currentMaxReadFileLine, mockTask)
 
@@ -377,4 +464,158 @@ describe("contextValidator", () => {
 			expect(result.reason).toContain("Large file detected")
 		})
 	})
+
+	describe("heuristic optimization", () => {
+		it("should skip validation for files with less than 100 lines", async () => {
+			const filePath = "/test/small-file.ts"
+			const totalLines = 50 // Less than 100 lines
+			const currentMaxReadFileLine = -1
+
+			// Mock file size to be small (3KB)
+			vi.mocked(fs.stat).mockResolvedValue({
+				size: 3 * 1024, // 3KB
+			} as any)
+
+			const result = await validateFileSizeForContext(filePath, totalLines, currentMaxReadFileLine, mockTask)
+
+			// Should not limit small files
+			expect(result.shouldLimit).toBe(false)
+			expect(result.safeMaxLines).toBe(currentMaxReadFileLine)
+			// Should not call countTokens for small files
+			expect(mockTask.api.countTokens).not.toHaveBeenCalled()
+			// Should not even attempt to read the file
+			expect(readLines).not.toHaveBeenCalled()
+		})
+
+		it("should skip validation for small files", async () => {
+			const filePath = "/test/small-file.ts"
+			const totalLines = 500
+			const currentMaxReadFileLine = -1
+
+			// Mock file size to be small (3KB)
+			vi.mocked(fsPromises.stat).mockResolvedValueOnce({
+				size: 3 * 1024, // 3KB
+			} as any)
+
+			const result = await validateFileSizeForContext(filePath, totalLines, currentMaxReadFileLine, mockTask)
+
+			// Small files should skip validation
+			expect(result.shouldLimit).toBe(false)
+			expect(result.safeMaxLines).toBe(currentMaxReadFileLine)
+			// Should not call readLines for validation
+			expect(readLines).not.toHaveBeenCalled()
+			// Should not call countTokens
+			expect(mockTask.api.countTokens).not.toHaveBeenCalled()
+			// Verify fs.stat was called
+			expect(fsPromises.stat).toHaveBeenCalledWith(filePath)
+		})
+
+		it("should skip validation for moderate files when context is mostly empty", async () => {
+			const filePath = "/test/moderate-file.ts"
+			const totalLines = 2000
+			const currentMaxReadFileLine = -1
+
+			// Mock file size to be moderate (80KB - below 100KB threshold)
+			vi.mocked(fsPromises.stat).mockResolvedValueOnce({
+				size: 80 * 1024, // 80KB
+			} as any)
+
+			// Mock context to be mostly empty (30% used - below 50% threshold)
+			mockTask.getTokenUsage = vi.fn().mockReturnValue({
+				contextTokens: 30000, // 30% of 100000
+			})
+
+			const result = await validateFileSizeForContext(filePath, totalLines, currentMaxReadFileLine, mockTask)
+
+			// Should skip validation when context is mostly empty and file is moderate
+			expect(result.shouldLimit).toBe(false)
+			expect(result.safeMaxLines).toBe(currentMaxReadFileLine)
+			expect(readLines).not.toHaveBeenCalled()
+			expect(mockTask.api.countTokens).not.toHaveBeenCalled()
+			// Verify fs.stat was called
+			expect(fsPromises.stat).toHaveBeenCalledWith(filePath)
+		})
+
+		it("should perform validation for larger files", async () => {
+			const filePath = "/test/large-file.ts"
+			const totalLines = 1000
+			const currentMaxReadFileLine = -1
+
+			// Mock file size to be large (1MB)
+			vi.mocked(fs.stat).mockResolvedValue({
+				size: 1024 * 1024, // 1MB
+			} as any)
+
+			// Mock readLines to return normal content
+			vi.mocked(readLines).mockImplementation(async (path, endLine, startLine) => {
+				const start = startLine ?? 0
+				const end = endLine ?? 0
+
+				// For sampling phase, return normal code lines
+				if (start === 0 && end === 49) {
+					const lines = []
+					for (let i = 0; i <= 49; i++) {
+						lines.push(`const variable${i} = "This is a normal length line of code";`)
+					}
+					return lines.join("\n")
+				}
+
+				// For actual reading
+				const lines = []
+				for (let i = start; i <= end; i++) {
+					lines.push(`const variable${i} = "This is a normal length line of code";`)
+				}
+				return lines.join("\n")
+			})
+
+			// Mock token counting
+			mockTask.api.countTokens = vi.fn().mockResolvedValue(100)
+
+			const result = await validateFileSizeForContext(filePath, totalLines, currentMaxReadFileLine, mockTask)
+
+			// Should perform normal validation
+			expect(readLines).toHaveBeenCalled()
+			expect(mockTask.api.countTokens).toHaveBeenCalled()
+		})
+
+		it("should handle cutback strategy when content exceeds limit", async () => {
+			const filePath = "/test/cutback-test.ts"
+			const totalLines = 1000
+			const currentMaxReadFileLine = -1
+
+			// Mock readLines to return content
+			vi.mocked(readLines).mockImplementation(async (path, endLine, startLine) => {
+				const start = startLine ?? 0
+				const end = Math.min(endLine ?? 499, start + 499)
+				const lines = []
+				for (let i = start; i <= end && i < totalLines; i++) {
+					lines.push(`const variable${i} = "This is a line of content";`)
+				}
+				return lines.join("\n")
+			})
+
+			// Mock token counting to exceed limit on first call, then succeed after cutback
+			let apiCallCount = 0
+			mockTask.api.countTokens = vi.fn().mockImplementation(async (content) => {
+				apiCallCount++
+				const text = content[0].text
+				const charCount = text.length
+
+				// First call: return tokens that exceed the limit
+				if (apiCallCount === 1) {
+					return 70000 // Exceeds available tokens
+				}
+				// After cutback: return acceptable amount
+				return Math.ceil(charCount / 3)
+			})
+
+			const result = await validateFileSizeForContext(filePath, totalLines, currentMaxReadFileLine, mockTask)
+
+			// Should apply cutback strategy
+			expect(mockTask.api.countTokens).toHaveBeenCalledTimes(2) // Initial + after cutback
+			expect(result.shouldLimit).toBe(true)
+			expect(result.safeMaxLines).toBeLessThan(totalLines)
+			expect(result.safeMaxLines).toBeGreaterThan(0)
+		})
+	})
 })

src/core/tools/__tests__/readFileTool.spec.ts

@@ -12,6 +12,10 @@ import { readFileTool } from "../readFileTool"
 import { formatResponse } from "../../prompts/responses"
 import * as contextValidatorModule from "../contextValidator"
 
+vi.mock("../../../i18n", () => ({
+	t: vi.fn((key: string) => key),
+}))
+
 vi.mock("path", async () => {
 	const originalPath = await vi.importActual("path")
 	return {
@@ -242,8 +246,7 @@ describe("read_file tool with maxReadFileLine setting", () => {
 			expect(result).toContain(`<list_code_definition_names>`)
 
 			// Verify XML structure
-			expect(result).toContain("<notice>Showing only 0 of 5 total lines")
-			expect(result).toContain("</notice>")
+			expect(result).toContain("<notice>tools.readFile.showingOnlyLines</notice>")
 			expect(result).toContain("<list_code_definition_names>")
 			expect(result).toContain(sourceCodeDef.trim())
 			expect(result).toContain("</list_code_definition_names>")
@@ -269,7 +272,7 @@ describe("read_file tool with maxReadFileLine setting", () => {
 			expect(result).toContain(`<file><path>${testFilePath}</path>`)
 			expect(result).toContain(`<content lines="1-3">`)
 			expect(result).toContain(`<list_code_definition_names>`)
-			expect(result).toContain("<notice>Showing only 3 of 5 total lines")
+			expect(result).toContain("<notice>tools.readFile.showingOnlyLines</notice>")
 		})
 	})
 
@@ -565,11 +568,7 @@ describe("read_file tool XML output structure", () => {
 			// Verify the result contains the inline instructions
 			expect(result).toContain("<notice>")
 			expect(result).toContain("File exceeds available context space")
-			expect(result).toContain("To read specific sections of this file, use the following format:")
-			expect(result).toContain("<line_range>start-end</line_range>")
-			expect(result).toContain("For example, to read lines 2001-3000:")
-			expect(result).toContain("<line_range>2001-3000</line_range>")
-			expect(result).toContain("large-file.ts")
+			expect(result).toContain("tools.readFile.contextLimitInstructions</notice>")
 		})
 
 		it("should not show any special notice when file fits in context", async () => {

src/core/tools/contextValidator.ts

@@ -1,6 +1,7 @@
 import { Task } from "../task/Task"
 import { readLines } from "../../integrations/misc/read-lines"
-import { getModelMaxOutputTokens } from "../../shared/api"
+import { getModelMaxOutputTokens, getFormatForProvider } from "../../shared/api"
+import * as fs from "fs/promises"
 
 /**
  * More aggressive buffer percentage specifically for file reading validation.
@@ -15,9 +16,61 @@ export interface ContextValidationResult {
 	reason?: string
 }
 
+/**
+ * Determines if we should skip the expensive token-based validation.
+ * Returns true if we're confident the file can be read without limits.
+ * Prioritizes accuracy - only skips when very confident.
+ */
+async function shouldSkipValidation(filePath: string, totalLines: number, cline: Task): Promise<boolean> {
+	// Heuristic 1: Very small files by line count (< 100 lines)
+	if (totalLines < 100) {
+		console.log(
+			`[shouldSkipValidation] Skipping validation for ${filePath} - small line count (${totalLines} lines)`,
+		)
+		return true
+	}
+
+	try {
+		// Get file size
+		const stats = await fs.stat(filePath)
+		const fileSizeBytes = stats.size
+		const fileSizeMB = fileSizeBytes / (1024 * 1024)
+
+		// Heuristic 2: Very small files by size (< 5KB) - definitely safe to skip validation
+		if (fileSizeBytes < 5 * 1024) {
+			console.log(
+				`[shouldSkipValidation] Skipping validation for ${filePath} - small file size (${(fileSizeBytes / 1024).toFixed(1)}KB)`,
+			)
+			return true
+		}
+
+		// For larger files, check if context is mostly empty
+		const modelInfo = cline.api.getModel().info
+		const { contextTokens: currentContextTokens } = cline.getTokenUsage()
+		const contextWindow = modelInfo.contextWindow
+
+		// Calculate context usage percentage
+		const contextUsagePercent = (currentContextTokens || 0) / contextWindow
+
+		// Heuristic 3: If context is mostly empty (< 50% used) and file is not too big (< 100KB),
+		// we can skip validation as there's plenty of room
+		if (contextUsagePercent < 0.5 && fileSizeBytes < 100 * 1024) {
+			console.log(
+				`[validateFileSizeForContext] Skipping validation for ${filePath} - context mostly empty (${Math.round(contextUsagePercent * 100)}% used) and file is moderate size (${fileSizeMB.toFixed(2)}MB)`,
+			)
+			return true
+		}
+	} catch (error) {
+		// If we can't check file size or context state, don't skip validation
+		console.warn(`[validateFileSizeForContext] Could not check file size or context state: ${error}`)
+	}
+
+	return false
+}
+
 /**
  * Validates if a file can be safely read based on its size and current runtime context state.
- * Reads lines incrementally and counts tokens as it goes, stopping when reaching the token limit.
+ * Uses a 2-phase approach: character-based estimation followed by actual token validation.
  * Returns a safe maxReadFileLine value to prevent context overflow.
  */
 export async function validateFileSizeForContext(
@@ -27,6 +80,11 @@ export async function validateFileSizeForContext(
 	cline: Task,
 ): Promise<ContextValidationResult> {
 	try {
+		// Check if we can skip validation
+		if (await shouldSkipValidation(filePath, totalLines, cline)) {
+			return { shouldLimit: false, safeMaxLines: currentMaxReadFileLine }
+		}
+
 		// Get actual runtime state from the task
 		const modelInfo = cline.api.getModel().info
 		const { contextTokens: currentContextTokens } = cline.getTokenUsage()
@@ -37,22 +95,8 @@ export async function validateFileSizeForContext(
 		const apiProvider = cline.apiConfiguration.apiProvider
 		const settings = await cline.providerRef.deref()?.getState()
 
-		// Map apiProvider to the format expected by getModelMaxOutputTokens
-		let format: "anthropic" | "openai" | "gemini" | "openrouter" | undefined
-		if (
-			apiProvider === "anthropic" ||
-			apiProvider === "bedrock" ||
-			apiProvider === "vertex" ||
-			apiProvider === "claude-code"
-		) {
-			format = "anthropic"
-		} else if (apiProvider === "openrouter") {
-			format = "openrouter"
-		} else if (apiProvider === "openai" || apiProvider === "openai-native") {
-			format = "openai"
-		} else if (apiProvider === "gemini" || apiProvider === "gemini-cli") {
-			format = "gemini"
-		}
+		// Use the centralized utility function to get the format
+		const format = getFormatForProvider(apiProvider)
 
 		const maxResponseTokens = getModelMaxOutputTokens({ modelId, model: modelInfo, settings, format })
 
@@ -73,91 +117,151 @@ export async function validateFileSizeForContext(
 		// Calculate available tokens for file content
 		const availableTokensForFile = usableRemainingContext - reservedForResponse
 
-		// Now read lines incrementally and count tokens until we reach the limit
-		const BATCH_SIZE = 100 // Read 100 lines at a time
-		let currentLine = 0
-		let totalTokensSoFar = 0
-		let safeMaxLines = 0
-
 		// Use 90% of available space to leave some margin
 		const targetTokenLimit = Math.floor(availableTokensForFile * 0.9)
 
-		while (currentLine < totalLines && totalTokensSoFar < targetTokenLimit) {
-			// Calculate the end line for this batch
-			const batchEndLine = Math.min(currentLine + BATCH_SIZE - 1, totalLines - 1)
+		// Constants for the 2-phase approach
+		const CHARS_PER_TOKEN_ESTIMATE = 3
+		const CUTBACK_PERCENTAGE = 0.2 // 20% reduction when over limit
+		const READ_BATCH_SIZE = 100 // Read 100 lines at a time for efficiency
+
+		// Phase 1: Read content up to estimated safe character limit
+		const estimatedSafeChars = targetTokenLimit * CHARS_PER_TOKEN_ESTIMATE
+
+		let accumulatedContent = ""
+		let currentLine = 0
+		let lineToCharMap: Map<number, number> = new Map() // Maps line number to character position
+
+		// Track the start position of each line for potential cutback
+		lineToCharMap.set(0, 0)
+
+		// Read until we hit our estimated character limit or EOF
+		while (currentLine < totalLines && accumulatedContent.length < estimatedSafeChars) {
+			const batchEndLine = Math.min(currentLine + READ_BATCH_SIZE - 1, totalLines - 1)
 
 			try {
-				// Read the next batch of lines
 				const batchContent = await readLines(filePath, batchEndLine, currentLine)
 
-				// Count tokens for this batch
-				const batchTokens = await cline.api.countTokens([{ type: "text", text: batchContent }])
-
-				// Check if adding this batch would exceed our limit
-				if (totalTokensSoFar + batchTokens > targetTokenLimit) {
-					// This batch would exceed the limit
-					// Try to find a more precise cutoff within this batch
-					if (batchEndLine - currentLine > 10) {
-						// Read smaller chunks to find a more precise cutoff
-						const FINE_BATCH_SIZE = 10
-						let fineLine = currentLine
-
-						while (fineLine <= batchEndLine && totalTokensSoFar < targetTokenLimit) {
-							const fineEndLine = Math.min(fineLine + FINE_BATCH_SIZE - 1, batchEndLine)
-							const fineContent = await readLines(filePath, fineEndLine, fineLine)
-							const fineTokens = await cline.api.countTokens([{ type: "text", text: fineContent }])
-
-							if (totalTokensSoFar + fineTokens > targetTokenLimit) {
-								// Even this fine batch exceeds the limit
-								break
-							}
-
-							totalTokensSoFar += fineTokens
-							safeMaxLines = fineEndLine + 1 // Convert to 1-based line count
-							fineLine = fineEndLine + 1
-						}
+				// Track line positions within the accumulated content
+				let localPos = 0
+				for (let lineNum = currentLine; lineNum <= batchEndLine; lineNum++) {
+					const nextNewline = batchContent.indexOf("\n", localPos)
+					if (nextNewline !== -1) {
+						lineToCharMap.set(lineNum + 1, accumulatedContent.length + nextNewline + 1)
+						localPos = nextNewline + 1
 					}
-					// Stop processing more batches
-					break
 				}
 
-				// Add this batch's tokens to our total
-				totalTokensSoFar += batchTokens
-				safeMaxLines = batchEndLine + 1 // Convert to 1-based line count
+				accumulatedContent += batchContent
 				currentLine = batchEndLine + 1
 			} catch (error) {
-				// If we encounter an error reading a batch, stop here
+				console.warn(`[validateFileSizeForContext] Error reading batch: ${error}`)
 				break
 			}
 		}
 
+		// Phase 2: Validate with actual API and cutback if needed
+		let finalContent = accumulatedContent
+		let finalLineCount = currentLine
+		let apiCallCount = 0
+		const maxApiCalls = 5 // Safety limit to prevent infinite loops
+
+		while (apiCallCount < maxApiCalls) {
+			apiCallCount++
+
+			// Make the actual API call to count tokens
+			const actualTokens = await cline.api.countTokens([{ type: "text", text: finalContent }])
+
+			console.log(
+				`[validateFileSizeForContext] API call ${apiCallCount}: ${actualTokens} tokens for ${finalContent.length} chars (${finalLineCount} lines)`,
+			)
+
+			if (actualTokens <= targetTokenLimit) {
+				// We're under the limit, we're done!
+				break
+			}
+
+			// We're over the limit - cut back by 20%
+			const targetLength = Math.floor(finalContent.length * (1 - CUTBACK_PERCENTAGE))
+
+			// Find the line that gets us closest to the target length
+			let cutoffLine = 0
+			for (const [lineNum, charPos] of lineToCharMap.entries()) {
+				if (charPos > targetLength) {
+					break
+				}
+				cutoffLine = lineNum
+			}
+
+			// Ensure we don't cut back too far
+			if (cutoffLine < 10) {
+				console.warn(
+					`[validateFileSizeForContext] Cutback resulted in too few lines (${cutoffLine}), using minimum`,
+				)
+				cutoffLine = Math.min(50, totalLines)
+			}
+
+			// Get the character position for the cutoff line
+			const cutoffCharPos = lineToCharMap.get(cutoffLine) || 0
+			finalContent = accumulatedContent.substring(0, cutoffCharPos)
+			finalLineCount = cutoffLine
+
+			// Safety check
+			if (finalContent.length === 0) {
+				return {
+					shouldLimit: true,
+					safeMaxLines: 10,
+					reason: `File too large for available context. Even minimal content exceeds token limit.`,
+				}
+			}
+		}
+
+		// Log final statistics
+		console.log(
+			`[validateFileSizeForContext] Final: ${finalLineCount} lines, ${finalContent.length} chars, ${apiCallCount} API calls`,
+		)
+
 		// Ensure we provide at least a minimum useful amount
 		const minUsefulLines = 50
-		const finalSafeMaxLines = Math.max(minUsefulLines, safeMaxLines)
+		const finalSafeMaxLines = Math.max(minUsefulLines, finalLineCount)
 
 		// If we read the entire file without exceeding the limit, no limitation needed
-		if (safeMaxLines >= totalLines) {
+		if (finalLineCount >= totalLines) {
 			return { shouldLimit: false, safeMaxLines: currentMaxReadFileLine }
 		}
 
 		// If we couldn't read even the minimum useful lines
-		if (safeMaxLines < minUsefulLines) {
+		if (finalLineCount < minUsefulLines) {
 			return {
 				shouldLimit: true,
 				safeMaxLines: finalSafeMaxLines,
-				reason: `Very limited context space. Could only safely read ${safeMaxLines} lines before exceeding token limit. Context: ${currentlyUsed}/${contextWindow} tokens used (${Math.round((currentlyUsed / contextWindow) * 100)}%). Limited to ${finalSafeMaxLines} lines. Consider using search_files or line_range for specific sections.`,
+				reason: `Very limited context space. Could only safely read ${finalLineCount} lines before exceeding token limit. Context: ${currentlyUsed}/${contextWindow} tokens used (${Math.round((currentlyUsed / contextWindow) * 100)}%). Limited to ${finalSafeMaxLines} lines. Consider using search_files or line_range for specific sections.`,
 			}
 		}
 
 		return {
 			shouldLimit: true,
 			safeMaxLines: finalSafeMaxLines,
-			reason: `File exceeds available context space. Safely read ${finalSafeMaxLines} lines (${totalTokensSoFar} tokens) out of ${totalLines} total lines. Context usage: ${currentlyUsed}/${contextWindow} tokens (${Math.round((currentlyUsed / contextWindow) * 100)}%). Use line_range to read specific sections.`,
+			reason: `File exceeds available context space. Safely read ${finalSafeMaxLines} lines out of ${totalLines} total lines. Context usage: ${currentlyUsed}/${contextWindow} tokens (${Math.round((currentlyUsed / contextWindow) * 100)}%). Use line_range to read specific sections.`,
 		}
 	} catch (error) {
 		// If we can't get runtime state, fall back to conservative estimation
 		console.warn(`[validateFileSizeForContext] Error accessing runtime state: ${error}`)
 
+		// In error cases, we can't check context state, so use simple file size heuristics
+		try {
+			const stats = await fs.stat(filePath)
+			const fileSizeBytes = stats.size
+
+			// Very small files are safe
+			if (fileSizeBytes < 5 * 1024) {
+				return { shouldLimit: false, safeMaxLines: currentMaxReadFileLine }
+			}
+		} catch (statError) {
+			// If we can't even stat the file, proceed with conservative defaults
+			console.warn(`[validateFileSizeForContext] Could not stat file: ${statError}`)
+		}
+
 		if (totalLines > 10000) {
 			return {
 				shouldLimit: true,

src/shared/__tests__/providerFormat.spec.ts

@@ -0,0 +1,85 @@
+import { describe, it, expect } from "vitest"
+import { getFormatForProvider, isVertexAnthropicModel } from "../api"
+import { ProviderName } from "@roo-code/types"
+
+describe("providerFormat", () => {
+	describe("getFormatForProvider", () => {
+		it("should return 'anthropic' for Anthropic-based providers", () => {
+			const anthropicProviders: ProviderName[] = ["anthropic", "bedrock", "vertex", "claude-code", "requesty"]
+
+			anthropicProviders.forEach((provider) => {
+				expect(getFormatForProvider(provider)).toBe("anthropic")
+			})
+		})
+
+		it("should return 'openai' for OpenAI-based providers", () => {
+			const openaiProviders: ProviderName[] = [
+				"openai",
+				"openai-native",
+				"deepseek",
+				"moonshot",
+				"xai",
+				"groq",
+				"chutes",
+				"mistral",
+				"ollama",
+				"lmstudio",
+				"litellm",
+				"huggingface",
+				"glama",
+				"unbound",
+				"vscode-lm",
+				"human-relay",
+				"fake-ai",
+			]
+
+			openaiProviders.forEach((provider) => {
+				expect(getFormatForProvider(provider)).toBe("openai")
+			})
+		})
+
+		it("should return 'gemini' for Gemini-based providers", () => {
+			const geminiProviders: ProviderName[] = ["gemini", "gemini-cli"]
+
+			geminiProviders.forEach((provider) => {
+				expect(getFormatForProvider(provider)).toBe("gemini")
+			})
+		})
+
+		it("should return 'openrouter' for OpenRouter provider", () => {
+			expect(getFormatForProvider("openrouter")).toBe("openrouter")
+		})
+
+		it("should return undefined for undefined provider", () => {
+			expect(getFormatForProvider(undefined)).toBeUndefined()
+		})
+
+		it("should return undefined for unknown providers", () => {
+			// Test with a provider that doesn't exist in the switch statement
+			// by casting to bypass TypeScript type checking
+			expect(getFormatForProvider("unknown-provider" as ProviderName)).toBeUndefined()
+		})
+	})
+
+	describe("isVertexAnthropicModel", () => {
+		it("should return true for Claude models", () => {
+			expect(isVertexAnthropicModel("claude-3-opus")).toBe(true)
+			expect(isVertexAnthropicModel("claude-3-sonnet")).toBe(true)
+			expect(isVertexAnthropicModel("claude-3-haiku")).toBe(true)
+			expect(isVertexAnthropicModel("CLAUDE-3-OPUS")).toBe(true) // Case insensitive
+			expect(isVertexAnthropicModel("anthropic.claude-v2")).toBe(true)
+		})
+
+		it("should return false for non-Claude models", () => {
+			expect(isVertexAnthropicModel("gemini-pro")).toBe(false)
+			expect(isVertexAnthropicModel("gemini-1.5-pro")).toBe(false)
+			expect(isVertexAnthropicModel("palm-2")).toBe(false)
+			expect(isVertexAnthropicModel("gpt-4")).toBe(false)
+		})
+
+		it("should return false for undefined or empty model ID", () => {
+			expect(isVertexAnthropicModel(undefined)).toBe(false)
+			expect(isVertexAnthropicModel("")).toBe(false)
+		})
+	})
+})

src/shared/api.ts

@@ -1,10 +1,82 @@
 import {
 	type ModelInfo,
 	type ProviderSettings,
+	type ProviderName,
 	ANTHROPIC_DEFAULT_MAX_TOKENS,
 	CLAUDE_CODE_DEFAULT_MAX_OUTPUT_TOKENS,
 } from "@roo-code/types"
 
+// Provider Format Mapping
+
+/**
+ * Maps API provider names to their corresponding format for model parameter handling.
+ * This centralizes the provider-to-format mapping logic used across the codebase.
+ *
+ * @param apiProvider - The API provider name
+ * @returns The format string used by getModelParams and getModelMaxOutputTokens, or undefined if not mapped
+ */
+export function getFormatForProvider(
+	apiProvider: ProviderName | undefined,
+): "anthropic" | "openai" | "gemini" | "openrouter" | undefined {
+	if (!apiProvider) {
+		return undefined
+	}
+
+	switch (apiProvider) {
+		// Anthropic-based providers
+		case "anthropic":
+		case "bedrock":
+		case "vertex": // Note: vertex can use either anthropic or gemini format depending on the model
+		case "claude-code":
+		case "requesty": // Uses anthropic format based on code analysis
+			return "anthropic"
+
+		// OpenAI-based providers
+		case "openai":
+		case "openai-native":
+		case "deepseek":
+		case "moonshot":
+		case "xai":
+		case "groq":
+		case "chutes":
+		case "mistral":
+		case "ollama":
+		case "lmstudio":
+		case "litellm":
+		case "huggingface":
+		case "glama":
+		case "unbound":
+		case "vscode-lm":
+		case "human-relay":
+		case "fake-ai":
+			return "openai"
+
+		// Gemini-based providers
+		case "gemini":
+		case "gemini-cli":
+			return "gemini"
+
+		// OpenRouter
+		case "openrouter":
+			return "openrouter"
+
+		// Providers that don't have a specific format mapping
+		default:
+			return undefined
+	}
+}
+
+/**
+ * Special case: Vertex provider can use either anthropic or gemini format depending on the model.
+ * This function checks if a vertex model should use anthropic format.
+ *
+ * @param modelId - The model ID to check
+ * @returns true if the model should use anthropic format
+ */
+export function isVertexAnthropicModel(modelId?: string): boolean {
+	return modelId?.toLowerCase().includes("claude") ?? false
+}
+
 // ApiHandlerOptions
 
 export type ApiHandlerOptions = Omit<ProviderSettings, "apiProvider">
@@ -70,14 +142,17 @@ export const getModelMaxOutputTokens = ({
 		return settings.claudeCodeMaxOutputTokens || CLAUDE_CODE_DEFAULT_MAX_OUTPUT_TOKENS
 	}
 
+	// If format is not provided, derive it from the provider settings
+	const effectiveFormat = format ?? getFormatForProvider(settings?.apiProvider)
+
 	if (shouldUseReasoningBudget({ model, settings })) {
 		return settings?.modelMaxTokens || DEFAULT_HYBRID_REASONING_MODEL_MAX_TOKENS
 	}
 
 	const isAnthropicContext =
 		modelId.includes("claude") ||
-		format === "anthropic" ||
-		(format === "openrouter" && modelId.startsWith("anthropic/"))
+		effectiveFormat === "anthropic" ||
+		(effectiveFormat === "openrouter" && modelId.startsWith("anthropic/"))
 
 	// For "Hybrid" reasoning models, discard the model's actual maxTokens for Anthropic contexts
 	if (model.supportsReasoningBudget && isAnthropicContext) {
@@ -95,7 +170,7 @@ export const getModelMaxOutputTokens = ({
 	}
 
 	// For non-Anthropic formats without explicit maxTokens, return undefined
-	if (format) {
+	if (effectiveFormat) {
 		return undefined
 	}