feat: UI-safe request formatting to avoid embedding file contents

src/core/task/Task.ts

@@ -66,7 +66,8 @@ import { RepoPerTaskCheckpointService } from "../../services/checkpoints"
 
 // integrations
 import { DiffViewProvider } from "../../integrations/editor/DiffViewProvider"
-import { findToolName, formatContentBlockToMarkdown } from "../../integrations/misc/export-markdown"
+import { findToolName } from "../../integrations/misc/export-markdown"
+import { formatContentBlockForUi } from "../../shared/formatContentBlockForUi"
 import { RooTerminalProcess } from "../../integrations/terminal/types"
 import { TerminalRegistry } from "../../integrations/terminal/TerminalRegistry"
 
@@ -1794,7 +1795,7 @@ export class Task extends EventEmitter<TaskEvents> implements TaskLike {
 				"api_req_started",
 				JSON.stringify({
 					request:
-						currentUserContent.map((block) => formatContentBlockToMarkdown(block)).join("\n\n") +
+						currentUserContent.map((block) => formatContentBlockForUi(block)).join("\n\n") +
 						"\n\nLoading...",
 					apiProtocol,
 				}),
@@ -1835,7 +1836,7 @@ export class Task extends EventEmitter<TaskEvents> implements TaskLike {
 			const lastApiReqIndex = findLastIndex(this.clineMessages, (m) => m.say === "api_req_started")
 
 			this.clineMessages[lastApiReqIndex].text = JSON.stringify({
-				request: finalUserContent.map((block) => formatContentBlockToMarkdown(block)).join("\n\n"),
+				request: finalUserContent.map((block) => formatContentBlockForUi(block)).join("\n\n"),
 				apiProtocol,
 			} satisfies ClineApiReqInfo)
 

src/shared/formatContentBlockForUi.ts

@@ -0,0 +1,179 @@
+import { Anthropic } from "@anthropic-ai/sdk"
+
+/**
+ * UI-safe formatter for ContentBlockParam that avoids including large or sensitive payloads
+ * such as full file contents, diffs, or oversized text in UI messages (ui_messages.json).
+ *
+ * IMPORTANT:
+ * - This is ONLY for rendering content into UI messages (e.g., api_req_started.request).
+ * - Do NOT use this for API conversation history or export; those need full fidelity.
+ */
+export function formatContentBlockForUi(block: Anthropic.Messages.ContentBlockParam): string {
+	switch (block.type) {
+		case "text":
+			return sanitizeText(block.text ?? "")
+		case "image":
+			return "[Image]"
+		case "tool_use":
+			return summarizeToolUse(block)
+		case "tool_result":
+			if (typeof block.content === "string") {
+				return sanitizeText(block.content)
+			} else if (Array.isArray(block.content)) {
+				// Recursively sanitize nested blocks
+				return block.content.map(formatContentBlockForUi).join("\n")
+			} else {
+				return "[Tool Result]"
+			}
+		default:
+			return `[${block.type}]`
+	}
+}
+
+/**
+ * Summarize tool_use without dumping large params (like diff/content).
+ */
+function summarizeToolUse(block: Anthropic.Messages.ToolUseBlockParam): string {
+	const name = block.name
+	// Try to extract relevant lightweight params for display
+	try {
+		const params = (block as any)?.input ?? (block as any)?.params ?? {}
+		// Prefer path if present
+		const directPath = params?.path as string | undefined
+
+		// For XML args (e.g., read_file, apply_diff multi-file), collect a small summary of paths
+		const xmlArgs = typeof params?.args === "string" ? params.args : undefined
+		const pathsFromXml = xmlArgs ? extractPathsFromXml(xmlArgs) : []
+
+		if (name === "read_file") {
+			const paths = directPath ? [directPath] : pathsFromXml
+			if (paths.length === 0) return `[Tool Use: ${name}]`
+			if (paths.length === 1) return `[Tool Use: ${name}] ${paths[0]}`
+			return `[Tool Use: ${name}] ${paths[0]} (+${paths.length - 1} more)`
+		}
+
+		if (
+			name === "apply_diff" ||
+			name === "insert_content" ||
+			name === "search_and_replace" ||
+			name === "write_to_file"
+		) {
+			const paths = directPath ? [directPath] : pathsFromXml
+			if (paths.length === 0) return `[Tool Use: ${name}]`
+			if (paths.length === 1) return `[Tool Use: ${name}] ${paths[0]}`
+			return `[Tool Use: ${name}] ${paths[0]} (+${paths.length - 1} more)`
+		}
+
+		if (name === "search_files") {
+			const regex = params?.regex ? ` regex="${String(params.regex)}"` : ""
+			const fp = params?.file_pattern ? ` file_pattern="${String(params.file_pattern)}"` : ""
+			const p = params?.path ? ` ${String(params.path)}` : ""
+			return `[Tool Use: ${name}]${p}${regex}${fp}`
+		}
+
+		// Default: show name only
+		return `[Tool Use: ${name}]`
+	} catch {
+		return `[Tool Use: ${block.name}]`
+	}
+}
+
+/**
+ * Sanitize a text chunk for UI:
+ * - Collapse <files> XML to a per-file summary (hide <content> bodies)
+ * - Truncate very long text
+ * - Redact obvious diff blobs
+ */
+function sanitizeText(text: string): string {
+	if (!text) return ""
+
+	// If this looks like a files XML, summarize paths/errors/notices and drop content bodies
+	if (text.includes("<files") || text.includes("<files")) {
+		return summarizeFilesXml(text)
+	}
+
+	// If this contains diff markers, replace with a short placeholder
+	if (looksLikeDiff(text)) {
+		return "[diff content omitted]"
+	}
+
+	// Generic truncation to keep UI light-weight
+	const MAX = 2000
+	if (text.length > MAX) {
+		const omitted = text.length - MAX
+		return `${text.slice(0, MAX)}\n[omitted ${omitted} chars]`
+	}
+
+	return text
+}
+
+function looksLikeDiff(s: string): boolean {
+	return (
+		s.includes("<<<<<<< SEARCH") ||
+		s.includes(">>>>>>> REPLACE") ||
+		s.includes("<<<<<<< SEARCH") ||
+		s.includes(">>>>>>> REPLACE") ||
+		/^diff --git/m.test(s)
+	)
+}
+
+/**
+ * Summarize a <files> XML payload by listing file paths and high-level status,
+ * but never including <content> bodies.
+ */
+function summarizeFilesXml(xmlLike: string): string {
+	// Support both escaped and unescaped tags
+	const decode = (s: string) => s.replace(/</g, "<").replace(/>/g, ">").replace(/&/g, "&")
+
+	const raw = decode(xmlLike)
+	const fileRegex = /<file>([\s\S]*?)<\/file>/g
+	const items: string[] = []
+	let match: RegExpExecArray | null
+
+	while ((match = fileRegex.exec(raw)) !== null) {
+		const fileBlock = match[1]
+		const path = matchOne(fileBlock, /<path>([\s\S]*?)<\/path>/)
+		const error = matchOne(fileBlock, /<error>([\s\S]*?)<\/error>/)
+		const notice = matchOne(fileBlock, /<notice>([\s\S]*?)<\/notice>/)
+		const binary = matchOne(fileBlock, /<binary_file(?:[^>]*)>([\s\S]*?)<\/binary_file>/)
+
+		let line = path ? `- ${path}` : "- [unknown path]"
+		if (error) line += ` [error: ${singleLine(error)}]`
+		if (!error && binary) line += " [binary file]"
+		if (!error && !binary && notice) line += ` [${singleLine(notice)}]`
+
+		items.push(line)
+	}
+
+	if (items.length === 0) {
+		return "[files omitted]"
+	}
+
+	const MAX_ITEMS = 20
+	let output = items.slice(0, MAX_ITEMS).join("\n")
+	if (items.length > MAX_ITEMS) {
+		output += `\n[+${items.length - MAX_ITEMS} more files]`
+	}
+	return output
+}
+
+function extractPathsFromXml(xml: string): string[] {
+	const decode = (s: string) => s.replace(/</g, "<").replace(/>/g, ">").replace(/&/g, "&")
+	const raw = decode(xml)
+	const pathRegex = /<path>([\s\S]*?)<\/path>/g
+	const paths: string[] = []
+	let m: RegExpExecArray | null
+	while ((m = pathRegex.exec(raw)) !== null) {
+		paths.push(m[1])
+	}
+	return paths
+}
+
+function matchOne(source: string, re: RegExp): string | undefined {
+	const m = re.exec(source)
+	return m ? m[1] : undefined
+}
+
+function singleLine(s: string): string {
+	return s.replace(/\s+/g, " ").trim()
+}