[pull] main from microsoft:main

build/.cachesalt

@@ -1 +1 @@
-2026-06-22T15:10:10.546Z
+2026-07-07T20:00:56.412Z

build/lib/i18n.resources.json

@@ -680,6 +680,10 @@
 			"name": "vs/sessions/contrib/chat",
 			"project": "vscode-sessions"
 		},
+		{
+			"name": "vs/sessions/contrib/promptTimeline",
+			"project": "vscode-sessions"
+		},
 		{
 			"name": "vs/sessions/contrib/providers/copilotChatSessions",
 			"project": "vscode-sessions"

build/lib/stylelint/validateDesignTokens.ts

@@ -121,12 +121,12 @@ export function validateCodiconFontSizes(text: string): IDesignTokenViolation[]
 
 /** Exact px value -> suggested token var(s). Ambiguous values list alternatives. */
 const FONT_SIZE_RAMP: ReadonlyMap<number, string> = new Map([
-	[26, 'var(--vscode-agents-fontSize-heading1)'],
-	[18, 'var(--vscode-agents-fontSize-heading2)'],
-	[13, 'var(--vscode-bodyFontSize) or var(--vscode-agents-fontSize-heading3) or var(--vscode-agents-fontSize-body1)'],
-	[12, 'var(--vscode-bodyFontSize-small) or var(--vscode-agents-fontSize-label1)'],
-	[11, 'var(--vscode-bodyFontSize-xSmall) or var(--vscode-agents-fontSize-body2) or var(--vscode-agents-fontSize-label2)'],
-	[10, 'var(--vscode-agents-fontSize-label3)'],
+	[26, 'var(--vscode-fontSize-heading1)'],
+	[18, 'var(--vscode-fontSize-heading2)'],
+	[13, 'var(--vscode-fontSize-body1) or var(--vscode-fontSize-heading3)'],
+	[12, 'var(--vscode-fontSize-label1)'],
+	[11, 'var(--vscode-fontSize-body2) or var(--vscode-fontSize-label2)'],
+	[10, 'var(--vscode-fontSize-label3)'],
 ]);
 
 /**
@@ -255,7 +255,7 @@ export function validateCornerRadiusTokens(text: string): IDesignTokenViolation[
 // Font-weight token suggestions (sessions design-system area only)
 // ---------------------------------------------------------------------------
 //
-// The agents font ramp defines exactly two weights: regular (400) and
+// The font ramp defines exactly two weights: regular (400) and
 // semiBold (600). Any other numeric weight (e.g. 500, 700) is off the ramp and
 // snaps to the nearer of the two. The CSS keywords `normal` (400) and `bold`
 // (700) are normalised before snapping. `inherit`, `lighter`, `bolder` and
@@ -268,7 +268,7 @@ interface IFontWeightToken {
 	readonly name: string;
 }
 
-/** The only two weights in the agents ramp. */
+/** The only two weights in the font ramp. */
 const FONT_WEIGHT_TOKENS: readonly IFontWeightToken[] = [
 	{ weight: 400, name: 'regular' },
 	{ weight: 600, name: 'semiBold' },
@@ -303,7 +303,7 @@ function snapFontWeight(weight: number): IFontWeightToken {
 }
 
 /**
- * Finds hardcoded `font-weight` values and suggests the agents weight-ramp var.
+ * Finds hardcoded `font-weight` values and suggests the weight-ramp var.
  * Exact ramp values (400 / 600, plus `normal` = 400) are flagged as drop-in
  * replacements; off-ramp values (e.g. 500, 700, `bold`) snap to the nearer
  * token and call out that the value is off the two-weight ramp. `inherit`,
@@ -332,7 +332,7 @@ export function validateFontWeightTokens(text: string): IDesignTokenViolation[]
 		const note = exact ? '' : ' (off-ramp, 400/600 only)';
 		violations.push({
 			line,
-			message: `${shown} -> var(--vscode-agents-fontWeight-${token.name})${note}`
+			message: `${shown} -> var(--vscode-fontWeight-${token.name})${note}`
 		});
 	});
 
@@ -475,3 +475,55 @@ export function validateStrokeTokens(text: string): IDesignTokenViolation[] {
 
 	return violations;
 }
+
+// ---------------------------------------------------------------------------
+// Deprecated token usage
+// ---------------------------------------------------------------------------
+//
+// Some `--vscode-*` size tokens have been superseded by more generic ones and
+// are marked `@deprecated` in the size registry. Any remaining usage of a
+// deprecated token var is reported with its drop-in replacement, regardless of
+// which property it appears in.
+
+interface IDeprecatedToken {
+	readonly deprecated: string;
+	readonly replacement: string;
+}
+
+/** Deprecated token var -> its replacement. */
+const DEPRECATED_TOKENS: readonly IDeprecatedToken[] = [
+	{ deprecated: '--vscode-bodyFontSize', replacement: '--vscode-fontSize-body1' },
+	{ deprecated: '--vscode-bodyFontSize-small', replacement: '--vscode-fontSize-label1' },
+	{ deprecated: '--vscode-bodyFontSize-xSmall', replacement: '--vscode-fontSize-body2' },
+];
+
+// Longest-first so a suffixed token (e.g. `-small`) matches before the base
+// token that is its prefix (`--vscode-bodyFontSize`).
+const DEPRECATED_TOKENS_SORTED = [...DEPRECATED_TOKENS].sort((a, b) => b.deprecated.length - a.deprecated.length);
+
+/**
+ * Finds usages of deprecated token vars and suggests the replacement token.
+ * A match is only reported when the token is not the prefix of a longer token
+ * (the next character does not continue the identifier). Returns one finding
+ * per occurrence so the terminal can linkify each `file(line,col)`.
+ */
+export function validateDeprecatedTokens(text: string): IDesignTokenViolation[] {
+	const violations: IDesignTokenViolation[] = [];
+
+	forEachDeclaration(text, (line, _selector, declaration) => {
+		for (const { deprecated, replacement } of DEPRECATED_TOKENS_SORTED) {
+			for (let idx = declaration.indexOf(deprecated); idx !== -1; idx = declaration.indexOf(deprecated, idx + deprecated.length)) {
+				const nextChar = declaration[idx + deprecated.length] ?? '';
+				if (/[\w-]/.test(nextChar)) {
+					continue; // prefix of a longer token
+				}
+				violations.push({
+					line,
+					message: `${deprecated} -> ${replacement} (deprecated)`
+				});
+			}
+		}
+	});
+
+	return violations;
+}

build/lib/stylelint/vscode-known-variables.json

@@ -955,6 +955,7 @@
 		"--session-view-background",
 		"--session-view-foreground",
 		"--session-view-centered-content-max-width",
+		"--prompt-timeline-bottom",
 		"--chat-editing-last-edit-shift",
 		"--chat-voice-icon-glow-color",
 		"--chat-current-response-min-height",

build/stylelint.ts

@@ -7,7 +7,7 @@ import es from 'event-stream';
 import vfs from 'vinyl-fs';
 import { stylelintFilter } from './filters.ts';
 import { getVariableNameValidator } from './lib/stylelint/validateVariableNames.ts';
-import { validateCodiconFontSizes, validateFontSizeTokens, validateFontWeightTokens, validateCornerRadiusTokens, validateSpacingTokens, validateStrokeTokens } from './lib/stylelint/validateDesignTokens.ts';
+import { validateCodiconFontSizes, validateFontSizeTokens, validateFontWeightTokens, validateCornerRadiusTokens, validateSpacingTokens, validateStrokeTokens, validateDeprecatedTokens } from './lib/stylelint/validateDesignTokens.ts';
 
 interface FileWithLines {
 	__lines?: string[];
@@ -32,7 +32,7 @@ export default function gulpstylelint(reporter: Reporter, designTokensEverywhere
 	const layerCheckerDisablePattern = /\/\*\s*stylelint-disable\s+layer-checker\s*\*\//;
 
 	// Per-category tally of design-token suggestions for the summary footer.
-	const designTokenCounts: Record<string, number> = { codicon: 0, 'font-size': 0, weight: 0, radius: 0, spacing: 0, stroke: 0 };
+	const designTokenCounts: Record<string, number> = { codicon: 0, 'font-size': 0, weight: 0, radius: 0, spacing: 0, stroke: 0, deprecated: 0 };
 	let designTokenFileCount = 0;
 
 	return es.through(function (this, file: FileWithLines) {
@@ -73,6 +73,7 @@ export default function gulpstylelint(reporter: Reporter, designTokensEverywhere
 			for (const v of validateCornerRadiusTokens(contents)) { findings.push({ line: v.line, category: 'radius', message: v.message }); }
 			for (const v of validateSpacingTokens(contents)) { findings.push({ line: v.line, category: 'spacing', message: v.message }); }
 			for (const v of validateStrokeTokens(contents)) { findings.push({ line: v.line, category: 'stroke', message: v.message }); }
+			for (const v of validateDeprecatedTokens(contents)) { findings.push({ line: v.line, category: 'deprecated', message: v.message }); }
 
 			if (findings.length > 0) {
 				findings.sort((a, b) => a.line - b.line);
@@ -91,7 +92,7 @@ export default function gulpstylelint(reporter: Reporter, designTokensEverywhere
 		if (errorCount > 0) {
 			reporter('All valid variable names are in `build/lib/stylelint/vscode-known-variables.json`\nTo update that file, run `./scripts/test-documentation.sh|bat.`', false);
 		}
-		const designTokenTotal = designTokenCounts.codicon + designTokenCounts['font-size'] + designTokenCounts.weight + designTokenCounts.radius + designTokenCounts.spacing + designTokenCounts.stroke;
+		const designTokenTotal = designTokenCounts.codicon + designTokenCounts['font-size'] + designTokenCounts.weight + designTokenCounts.radius + designTokenCounts.spacing + designTokenCounts.stroke + designTokenCounts.deprecated;
 		if (designTokenTotal > 0) {
 			reporter('', false);
 			reporter(
@@ -101,7 +102,8 @@ export default function gulpstylelint(reporter: Reporter, designTokensEverywhere
 				', weight ' + designTokenCounts.weight +
 				', radius ' + designTokenCounts.radius +
 				', spacing ' + designTokenCounts.spacing +
-				', stroke ' + designTokenCounts.stroke + ')',
+				', stroke ' + designTokenCounts.stroke +
+				', deprecated ' + designTokenCounts.deprecated + ')',
 				false
 			);
 		}

extensions/copilot/package.json

@@ -1918,12 +1918,16 @@
                   },
                   "maxInputTokens": {
                     "type": "number",
-                    "description": "Maximum number of input tokens supported by the model"
+                    "markdownDescription": "Maximum number of input (prompt) tokens supported by the model. Optional when `contextWindow` is set, in which case it is derived as `contextWindow - maxOutputTokens`."
                   },
                   "maxOutputTokens": {
                     "type": "number",
                     "description": "Maximum number of output tokens supported by the model"
                   },
+                  "contextWindow": {
+                    "type": "number",
+                    "markdownDescription": "The model's full context window (input + output) in tokens, e.g. `1000000` for a 1M model. When set it is the source of truth for the context window and `maxInputTokens` can be omitted. Otherwise the window is derived as `maxInputTokens + maxOutputTokens`."
+                  },
                   "editTools": {
                     "type": "array",
                     "description": "List of edit tools supported by the model. If this is not configured, the editor will try multiple edit tools and pick the best one.\n\n- 'find-replace': Find and replace text in a document.\n- 'multi-find-replace': Find and replace text in a document.\n- 'apply-patch': A file-oriented diff format used by some OpenAI models\n- 'code-rewrite': A general but slower editing tool that allows the model to rewrite and code snippet and provide only the replacement to the editor.",
@@ -1981,8 +1985,19 @@
                   "url",
                   "toolCalling",
                   "vision",
-                  "maxInputTokens",
                   "maxOutputTokens"
+                ],
+                "anyOf": [
+                  {
+                    "required": [
+                      "maxInputTokens"
+                    ]
+                  },
+                  {
+                    "required": [
+                      "contextWindow"
+                    ]
+                  }
                 ]
               }
             }
@@ -2089,12 +2104,16 @@
                   },
                   "maxInputTokens": {
                     "type": "number",
-                    "description": "Maximum number of input tokens supported by the model"
+                    "markdownDescription": "Maximum number of input (prompt) tokens supported by the model. Optional when `contextWindow` is set, in which case it is derived as `contextWindow - maxOutputTokens`."
                   },
                   "maxOutputTokens": {
                     "type": "number",
                     "description": "Maximum number of output tokens supported by the model"
                   },
+                  "contextWindow": {
+                    "type": "number",
+                    "markdownDescription": "The model's full context window (input + output) in tokens, e.g. `1000000` for a 1M model. When set it is the source of truth for the context window and `maxInputTokens` can be omitted. Otherwise the window is derived as `maxInputTokens + maxOutputTokens`."
+                  },
                   "editTools": {
                     "type": "array",
                     "description": "List of edit tools supported by the model. If this is not configured, the editor will try multiple edit tools and pick the best one.\n\n- 'find-replace': Find and replace text in a document.\n- 'multi-find-replace': Find and replace text in a document.\n- 'apply-patch': A file-oriented diff format used by some OpenAI models\n- 'code-rewrite': A general but slower editing tool that allows the model to rewrite and code snippet and provide only the replacement to the editor.",
@@ -2176,8 +2195,19 @@
                   "url",
                   "toolCalling",
                   "vision",
-                  "maxInputTokens",
                   "maxOutputTokens"
+                ],
+                "anyOf": [
+                  {
+                    "required": [
+                      "maxInputTokens"
+                    ]
+                  },
+                  {
+                    "required": [
+                      "contextWindow"
+                    ]
+                  }
                 ]
               }
             }
@@ -2240,12 +2270,16 @@
                   },
                   "maxInputTokens": {
                     "type": "number",
-                    "description": "Maximum number of input tokens supported by the model"
+                    "markdownDescription": "Maximum number of input (prompt) tokens supported by the model. Optional when `contextWindow` is set, in which case it is derived as `contextWindow - maxOutputTokens`."
                   },
                   "maxOutputTokens": {
                     "type": "number",
                     "description": "Maximum number of output tokens supported by the model"
                   },
+                  "contextWindow": {
+                    "type": "number",
+                    "markdownDescription": "The model's full context window (input + output) in tokens, e.g. `1000000` for a 1M model. When set it is the source of truth for the context window and `maxInputTokens` can be omitted. Otherwise the window is derived as `maxInputTokens + maxOutputTokens`."
+                  },
                   "thinking": {
                     "type": "boolean",
                     "default": false,
@@ -2290,8 +2324,19 @@
                   "url",
                   "toolCalling",
                   "vision",
-                  "maxInputTokens",
                   "maxOutputTokens"
+                ],
+                "anyOf": [
+                  {
+                    "required": [
+                      "maxInputTokens"
+                    ]
+                  },
+                  {
+                    "required": [
+                      "contextWindow"
+                    ]
+                  }
                 ]
               }
             }

extensions/copilot/src/extension/byok/common/byokProvider.ts

@@ -47,8 +47,19 @@ export type BYOKModelConfig = BYOKGlobalKeyModelConfig | BYOKPerModelConfig | BY
 export interface BYOKModelCapabilities {
 	name: string;
 	url?: string;
-	maxInputTokens: number;
+	/**
+	 * The maximum number of prompt (input) tokens. Optional when {@link contextWindow}
+	 * is supplied, in which case it is derived as `contextWindow - maxOutputTokens`.
+	 */
+	maxInputTokens?: number;
 	maxOutputTokens: number;
+	/**
+	 * The model's full context window (input + output) in tokens. Many providers
+	 * publish this directly (e.g. "Context Length: 1M"). When set it is used as the
+	 * source of truth for the context window; otherwise the window is derived as
+	 * `maxInputTokens + maxOutputTokens` for backward compatibility.
+	 */
+	contextWindow?: number;
 	toolCalling: boolean;
 	vision: boolean;
 	thinking?: boolean;
@@ -93,14 +104,43 @@ export function isNoAuthConfig(config: BYOKModelConfig): config is BYOKNoAuthMod
 	return !('apiKey' in config) && !('deploymentUrl' in config);
 }
 
+/**
+ * Resolves a model's token limits from its BYOK capabilities, honoring an explicit
+ * {@link BYOKModelCapabilities.contextWindow} when provided.
+ *
+ * - When `contextWindow` is set it is the source of truth for the full window and, if
+ *   `maxInputTokens` is omitted, the prompt budget is derived as
+ *   `contextWindow - maxOutputTokens`.
+ * - Otherwise the window falls back to `maxInputTokens + maxOutputTokens` for backward
+ *   compatibility.
+ *
+ * The returned limits are always internally consistent: `maxOutputTokens` is clamped so
+ * it never exceeds the context window, and `maxInputTokens` is clamped to the remaining
+ * budget (`contextWindow - maxOutputTokens`). This prevents invalid combinations such as
+ * `maxOutputTokens > contextWindow`, or a `maxInputTokens` supplied alongside a smaller
+ * `contextWindow` overflowing the window.
+ */
+export function resolveModelTokenLimits(capabilities: Pick<BYOKModelCapabilities, 'maxInputTokens' | 'maxOutputTokens' | 'contextWindow'>): { contextWindow: number; maxInputTokens: number; maxOutputTokens: number } {
+	const contextWindow = capabilities.contextWindow ?? ((capabilities.maxInputTokens ?? 0) + capabilities.maxOutputTokens);
+	// The output budget can never exceed the full window.
+	const maxOutputTokens = Math.min(capabilities.maxOutputTokens, contextWindow);
+	// The prompt budget is whatever remains after the output reservation; an explicitly
+	// provided maxInputTokens is clamped to that remaining budget.
+	const remainingInputBudget = Math.max(0, contextWindow - maxOutputTokens);
+	const maxInputTokens = Math.min(capabilities.maxInputTokens ?? remainingInputBudget, remainingInputBudget);
+	return { contextWindow, maxInputTokens, maxOutputTokens };
+}
+
 export function resolveModelInfo(modelId: string, providerName: string, knownModels: BYOKKnownModels | undefined, modelCapabilities?: BYOKModelCapabilities): IChatModelInformation {
 	// Model Capabilities are something the user has decided on so those take precedence, then we rely on known model info, then defaults.
 	let knownModelInfo = modelCapabilities;
 	if (knownModels && !knownModelInfo) {
 		knownModelInfo = knownModels[modelId];
 	}
 	const modelName = knownModelInfo?.name || modelId;
-	const contextWinow = knownModelInfo ? (knownModelInfo.maxInputTokens + knownModelInfo.maxOutputTokens) : 128000;
+	const limits = knownModelInfo
+		? resolveModelTokenLimits(knownModelInfo)
+		: { contextWindow: 128000, maxInputTokens: 100000, maxOutputTokens: 8192 };
 	const modelInfo: IChatModelInformation = {
 		id: modelId,
 		name: modelName,
@@ -119,9 +159,9 @@ export function resolveModelInfo(modelId: string, providerName: string, knownMod
 			},
 			tokenizer: TokenizerType.O200K,
 			limits: {
-				max_context_window_tokens: contextWinow,
-				max_prompt_tokens: knownModelInfo?.maxInputTokens || 100000,
-				max_output_tokens: knownModelInfo?.maxOutputTokens || 8192
+				max_context_window_tokens: limits.contextWindow,
+				max_prompt_tokens: limits.maxInputTokens,
+				max_output_tokens: limits.maxOutputTokens
 			}
 		},
 		is_chat_default: false,
@@ -146,12 +186,13 @@ export function byokKnownModelsToAPIInfo(providerName: string, knownModels: BYOK
 }
 
 export function byokKnownModelToAPIInfo(providerName: string, id: string, capabilities: BYOKModelCapabilities): LanguageModelChatInformation {
+	const limits = resolveModelTokenLimits(capabilities);
 	return {
 		id,
 		name: capabilities.name,
 		version: '1.0.0',
-		maxOutputTokens: capabilities.maxOutputTokens,
-		maxInputTokens: capabilities.maxInputTokens,
+		maxOutputTokens: limits.maxOutputTokens,
+		maxInputTokens: limits.maxInputTokens,
 		// `detail` is intentionally omitted: when this model is resolved
 		// via a configured provider group, `LanguageModelsService` will
 		// fall back to the group name so multiple instances of the same