diff --git a/packages/coding-agent/src/core/tools/truncate.ts b/packages/coding-agent/src/core/tools/truncate.ts
index c638ee480..05cded0ef 100644
--- a/packages/coding-agent/src/core/tools/truncate.ts
+++ b/packages/coding-agent/src/core/tools/truncate.ts
@@ -9,7 +9,13 @@
  */
 
 export const DEFAULT_MAX_LINES = 2000;
-export const DEFAULT_MAX_BYTES = 50 * 1024; // 50KB
+// 20KB per tool result: bounds how much any single read/find/grep can drag
+// into billed model context (each result rides along with every subsequent
+// call in the session). Offset/limit pagination remains available for more —
+// the truncation notices tell the model how to continue. Lowered from the
+// upstream 50KB after a production session accumulated ~450KB of context via
+// nine max-size reads of a legal document (CD-1284 follow-up).
+export const DEFAULT_MAX_BYTES = 20 * 1024; // 20KB
 export const GREP_MAX_LINE_LENGTH = 500; // Max chars per grep match line
 
 export interface TruncationResult {
diff --git a/packages/coding-agent/src/linc/casedev-vault-api.ts b/packages/coding-agent/src/linc/casedev-vault-api.ts
index 6ae9e487f..9a592a5a6 100644
--- a/packages/coding-agent/src/linc/casedev-vault-api.ts
+++ b/packages/coding-agent/src/linc/casedev-vault-api.ts
@@ -48,10 +48,6 @@ export interface CaseDevVaultReadTextParams {
 	filename?: string;
 }
 
-export const CASEDEV_VAULT_TEXT_MAX_CHARS = 10_000;
-export const CASEDEV_VAULT_TEXT_TRUNCATION_WARNING =
-	"this document is too long to view in full. Use targeted vault searches to retrieve more context";
-
 export interface CaseDevVaultSearchParams {
 	query: string;
 	method?: "hybrid" | "global" | "entity" | "fast" | "vector" | "graph" | "local";
@@ -287,24 +283,25 @@ export async function readCaseDevVaultObjectText(
 	const filename = baseName.endsWith(".txt") ? baseName : `${baseName}.txt`;
 	await mkdir(params.outDir, { recursive: true });
 	const outputPath = safeOutputPath(params.outDir, filename);
-	const truncated = text.length > CASEDEV_VAULT_TEXT_MAX_CHARS;
-	const visibleTextLimit = truncated
-		? CASEDEV_VAULT_TEXT_MAX_CHARS - CASEDEV_VAULT_TEXT_TRUNCATION_WARNING.length - 2
-		: CASEDEV_VAULT_TEXT_MAX_CHARS;
-	const visibleText = text.slice(0, visibleTextLimit);
-	const output = truncated ? `${CASEDEV_VAULT_TEXT_TRUNCATION_WARNING}\n\n${visibleText}` : visibleText;
-	await writeFile(outputPath, output, "utf8");
-	const pageMarkers = (visibleText.match(/^--- Page \d+ ---$/gm) ?? []).length;
+	// The full text is written to disk deliberately: files on disk are outside
+	// model context and cost nothing, and grep over the complete text is what
+	// makes comprehensive large-document analysis possible. Context safety is
+	// enforced where the cost is — per-result output caps on read/grep/bash —
+	// not by truncating the file (which silently breaks any search past the
+	// cut and does nothing for other large files in the workspace).
+	await writeFile(outputPath, text, "utf8");
+	const pageMarkers = (text.match(/^--- Page \d+ ---$/gm) ?? []).length;
+	const usage =
+		'Locate content with `grep -n "pattern" <path>` and read only narrow line ranges around matches. Never page through the full text — it is too large for the conversation.';
 	return {
 		objectId: params.objectId,
 		path: outputPath,
 		chars: text.length,
 		pageMarkers,
-		note: truncated
-			? CASEDEV_VAULT_TEXT_TRUNCATION_WARNING
-			: pageMarkers > 0
-				? "Text contains --- Page N --- markers; cite findings by those page numbers."
-				: "This document has no page markers; cite findings by section or heading.",
+		note:
+			(pageMarkers > 0
+				? "Text contains --- Page N --- markers; cite findings by those page numbers. "
+				: "This document has no page markers; cite findings by section or heading. ") + usage,
 	};
 }
 
diff --git a/packages/coding-agent/src/linc/casedev-vault-tools.ts b/packages/coding-agent/src/linc/casedev-vault-tools.ts
index 667390a30..89e01ff9a 100644
--- a/packages/coding-agent/src/linc/casedev-vault-tools.ts
+++ b/packages/coding-agent/src/linc/casedev-vault-tools.ts
@@ -351,7 +351,7 @@ export function createCaseDevVaultTools(): ToolDefinition[] {
 			name: "casedev_vault_read_text",
 			label: "case.dev vault read text",
 			description:
-				"Write up to the first 10,000 characters of an ingested Case.dev vault object's extracted text into the workspace as a .txt file. Use targeted vault searches for additional context.",
+				"Write the full extracted text of an ingested Case.dev vault object into the workspace as a .txt file, page-numbered when the source is paginated. Locate content with grep -n and read only narrow line ranges around matches; never page the full text into the conversation.",
 			promptSnippet: "Read a bounded excerpt of a Case.dev vault object's extracted text into the workspace.",
 			parameters: vaultReadTextSchema,
 			async execute(_toolCallId, params: VaultReadTextInput, signal, _onUpdate, ctx) {
@@ -439,7 +439,7 @@ export function createCaseDevVaultTools(): ToolDefinition[] {
 			name: "vault_read_text",
 			label: "vault_read_text",
 			description:
-				"Write up to the first 10,000 characters of an ingested matter document's extracted text into the workspace as a .txt file. Use targeted vault searches for additional context.",
+				"Write the full extracted text of an ingested matter document into the workspace as a .txt file, page-numbered when the source is paginated. The right tool for comprehensively analyzing document content at any size: locate content with grep -n and read only narrow line ranges around matches; never page the full text into the conversation.",
 			promptSnippet: "Read a bounded excerpt of a matter document's extracted text into the workspace",
 			promptGuidelines: [
 				"Originals from vault_download are raw bytes and often scanned images; use vault_read_text for a bounded excerpt and targeted vault searches for additional context.",
diff --git a/packages/coding-agent/test/casedev-vault-api.test.ts b/packages/coding-agent/test/casedev-vault-api.test.ts
index 35a9688df..a332fef06 100644
--- a/packages/coding-agent/test/casedev-vault-api.test.ts
+++ b/packages/coding-agent/test/casedev-vault-api.test.ts
@@ -4,8 +4,6 @@ import { join } from "node:path";
 import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 import type { ExtensionContext } from "../src/core/extensions/types.ts";
 import {
-	CASEDEV_VAULT_TEXT_MAX_CHARS,
-	CASEDEV_VAULT_TEXT_TRUNCATION_WARNING,
 	downloadCaseDevVaultObject,
 	listCaseDevVaults,
 	readCaseDevVaultObjectText,
@@ -232,8 +230,8 @@ describe("Case.dev vault REST API helper", () => {
 		});
 	});
 
-	it("limits extracted object text and prepends targeted-search guidance", async () => {
-		const text = "x".repeat(CASEDEV_VAULT_TEXT_MAX_CHARS + 1);
+	it("writes large extracted text to disk in full with grep-first guidance", async () => {
+		const text = "x".repeat(50_000);
 		vi.stubGlobal(
 			"fetch",
 			vi.fn(async () => jsonResponse({ text, metadata: { filename: "long.pdf" } })),
@@ -248,15 +246,13 @@ describe("Case.dev vault REST API helper", () => {
 		expect(result).toMatchObject({
 			objectId: "obj-long",
 			chars: text.length,
-			note: CASEDEV_VAULT_TEXT_TRUNCATION_WARNING,
 		});
+		expect(result.note).toContain("grep -n");
+		expect(result.note).toContain("Never page through the full text");
+		// The on-disk file is complete: disk is outside model context and free,
+		// and grep over the whole text is what enables large-document analysis.
 		const written = await readFile(join(cwd, "long.pdf.txt"), "utf-8");
-		expect(written).toHaveLength(CASEDEV_VAULT_TEXT_MAX_CHARS);
-		expect(written).toBe(
-			`${CASEDEV_VAULT_TEXT_TRUNCATION_WARNING}\n\n${"x".repeat(
-				CASEDEV_VAULT_TEXT_MAX_CHARS - CASEDEV_VAULT_TEXT_TRUNCATION_WARNING.length - 2,
-			)}`,
-		);
+		expect(written).toBe(text);
 	});
 
 	it("rejects text reads for objects with no extracted text", async () => {
diff --git a/packages/coding-agent/test/tools.test.ts b/packages/coding-agent/test/tools.test.ts
index c5e8fa6fb..970b95e92 100644
--- a/packages/coding-agent/test/tools.test.ts
+++ b/packages/coding-agent/test/tools.test.ts
@@ -84,7 +84,7 @@ describe("Coding Agent Tools", () => {
 
 		it("should truncate when byte limit exceeded", async () => {
 			const testFile = join(testDir, "large-bytes.txt");
-			// Create file that exceeds 50KB byte limit but has fewer than 2000 lines
+			// Create file that exceeds the byte limit but has fewer than 2000 lines
 			const lines = Array.from({ length: 500 }, (_, i) => `Line ${i + 1}: ${"x".repeat(200)}`);
 			writeFileSync(testFile, lines.join("\n"));