[pull] main from microsoft:main

.github/skills/sessions/SKILL.md

@@ -33,9 +33,7 @@ Then read the relevant spec for the area you are changing (see table below). If
 - **Modifying workbench code**: Prefer extending/wrapping workbench classes in the sessions layer over modifying shared workbench components.
 - **Timeouts as fixes**: Never use `setTimeout`/`disposableTimeout`/arbitrary delays to fix bugs or implement behaviour. They are race-prone guesses that mask the real ordering/state problem. Drive logic off deterministic signals instead — observables (`autorun`/`derived`), explicit events (`onDidChange*`), lifecycle phases, or awaiting the actual async operation.
 - **Stashed state read back later (side-channels)**: Never stash a value on a service during one method call and read it back from a separate query later, assuming it is still valid (e.g. a `Set`/flag set in `openSession` and consumed by a `shouldX()` pull-API). This is fragile temporal coupling. Instead, make it reactive state that is set **atomically together with its source of truth** and consumed reactively. Example: per-activation intent like "open in background / preserve focus" is exposed as an `IObservable` set in the **same transaction** as `activeSession` (via a single internal setter so it can never go stale), and read with `.read(reader)` in the consumer's `autorun` — never via a consume-once getter.
-- **Blocking on a "pending/waiting" state instead of creating + upgrading**: When an entity (e.g. a draft session) depends on something that registers asynchronously, don't withhold creation behind a pending/waiting state. Prefer creating immediately with the best available data, then **replace/upgrade** it once the awaited dependency arrives (driven by an `onDidChange*`/observable signal), cancelling the upgrade if the user changes the inputs meanwhile. Do **not** bound the upgrade with a timeout or even a lifecycle milestone like `LifecyclePhase.Eventually` — an agent host connects lazily and can surface its session types after `Eventually` has already fired, which would lock in the wrong fallback. Let the upgrade listener live for the consumer's lifetime instead.
-- **Persisting a picker value only on manual selection**: A picker that writes storage only when the user actively changes the dropdown will lose any auto-selected/defaulted value on reload (the stored preference is empty, so it falls back to a default). Persist on **every** change of the underlying value — exactly like a normal picker — so the last effective value always survives reload. Example: `SessionTypePicker` writes `{ providerId, sessionTypeId }` from both the manual pick handler and the active-session `refresh`, through the single `_writeStoredPick` persistence point.
-- **Pre-validating a stored preference against not-yet-ready state in an event handler**: Don't re-check a restored preference against currently-available data (e.g. validating the stored session-type pick against `getSessionTypesForFolder` inside `onDidSelectWorkspace`) and null it out on mismatch. During reload the awaited provider hasn't registered yet, so a perfectly valid preference is discarded and the wrong default is locked in. Pass the preference straight through to the single place that reconciles preference-vs-availability (which creates now and upgrades later).
+- **Blocking on a "pending/waiting" state instead of creating + upgrading**: When an entity (e.g. a draft session) depends on something that registers asynchronously, don't withhold creation behind a pending/waiting state. Prefer creating immediately with the best available data, then **replace/upgrade** it once the awaited dependency arrives (driven by an `onDidChange*`/observable signal), cancelling the upgrade if the user changes the inputs meanwhile. Do **not** bound the upgrade with a timeout or even a lifecycle milestone like `LifecyclePhase.Eventually` — an agent host connects lazily and can surface its session types arbitrarily late, which would lock in the wrong fallback. Let the upgrade listener live for the consumer's lifetime instead.
 
 ## Capturing Feedback (meta-rule)
 

extensions/copilot/package.json

@@ -3074,6 +3074,12 @@
         "icon": "$(git-pull-request)",
         "category": "GitHub Pull Request"
       },
+      {
+        "command": "github.copilot.chat.cloudSessions.createPullRequestForTask",
+        "title": "%github.copilot.command.cloudSessions.createPullRequestForTask.title%",
+        "icon": "$(git-pull-request-create)",
+        "category": "GitHub Pull Request"
+      },
       {
         "command": "github.copilot.chat.cloudSessions.openRepository",
         "title": "%github.copilot.command.cloudSessions.openRepository.title%",
@@ -5403,6 +5409,11 @@
           "command": "github.copilot.chat.checkoutPullRequestReroute",
           "when": "chatSessionType == copilot-cloud-agent && !github.vscode-pull-request-github.activated && gitOpenRepositoryCount != 0",
           "group": "navigation@0"
+        },
+        {
+          "command": "github.copilot.chat.cloudSessions.createPullRequestForTask",
+          "when": "chatSessionType == copilot-cloud-agent && github.copilot.chat.cloudTaskCanCreatePullRequest && !isSessionsWindow",
+          "group": "navigation@0"
         }
       ],
       "agents/changes/actions/primary": [
@@ -7024,7 +7035,6 @@
     "applicationinsights": "^2.9.7",
     "best-effort-json-parser": "^1.2.1",
     "diff": "^8.0.3",
-    "dompurify": "^3.4.8",
     "express": "^5.2.1",
     "ignore": "^7.0.5",
     "isbinaryfile": "^5.0.4",

extensions/copilot/package.nls.json

@@ -487,6 +487,7 @@
 	"github.copilot.chat.createPullRequestCopilotCLIAgentSession.createPR": "Create Pull Request",
 	"github.copilot.chat.createDraftPullRequestCopilotCLIAgentSession.createDraftPR": "Create Draft Pull Request",
 	"github.copilot.command.checkoutPullRequestReroute.title": "Checkout",
+	"github.copilot.command.cloudSessions.createPullRequestForTask.title": "Create Pull Request",
 	"github.copilot.command.cloudSessions.openRepository.title": "Browse repositories...",
 	"github.copilot.command.cloudSessions.clearCaches.title": "Clear Cloud Agent Caches",
 	"github.copilot.command.applyCopilotCLIAgentSessionChanges": "Apply Changes to Workspace",

extensions/copilot/src/extension/chatSessions/claude/common/claudeSessionStateService.ts

@@ -24,6 +24,7 @@ export interface SessionState {
 	folderInfo: ClaudeFolderInfo | undefined;
 	usageHandler: UsageHandler | undefined;
 	reasoningEffort: EffortLevel | undefined;
+	contextSize: number | undefined;
 	traceContext: TraceContext | undefined;
 	turnId: string | undefined;
 }
@@ -106,6 +107,18 @@ export interface IClaudeSessionStateService {
 	 */
 	setReasoningEffortForSession(sessionId: string, effort: EffortLevel | undefined): void;
 
+	/**
+	 * Gets the context size for a session (user's per-request selection from the model picker).
+	 * When set, the proxy reports this value as the endpoint's modelMaxPromptTokens so internal
+	 * accounting reflects the chosen tier.
+	 */
+	getContextSizeForSession(sessionId: string): number | undefined;
+
+	/**
+	 * Sets the context size for a session.
+	 */
+	setContextSizeForSession(sessionId: string, contextSize: number | undefined): void;
+
 	/**
 	 * Gets the OTel trace context for a session (used to parent chat spans to invoke_agent).
 	 */

extensions/copilot/src/extension/chatSessions/claude/node/claudeCodeModels.ts

@@ -8,7 +8,7 @@ import type * as vscode from 'vscode';
 import { IEndpointProvider } from '../../../../platform/endpoint/common/endpointProvider';
 import { ILogService } from '../../../../platform/log/common/logService';
 import { IChatEndpoint } from '../../../../platform/networking/common/networking';
-import { formatPricingLabel, getModelCapabilitiesDescription } from '../../../conversation/common/languageModelAccess';
+import { formatPricingLabel, formatTokenCount, getModelCapabilitiesDescription } from '../../../conversation/common/languageModelAccess';
 import { createServiceIdentifier } from '../../../../util/common/services';
 import { Emitter } from '../../../../util/vs/base/common/event';
 import { Disposable } from '../../../../util/vs/base/common/lifecycle';
@@ -17,6 +17,7 @@ import { tryParseClaudeModelId } from './claudeModelId';
 import type { EffortLevel } from '@anthropic-ai/claude-agent-sdk';
 
 export const CLAUDE_REASONING_EFFORT_PROPERTY = 'reasoningEffort';
+export const CLAUDE_CONTEXT_SIZE_PROPERTY = 'contextSize';
 
 export interface IClaudeCodeModels {
 	readonly _serviceBrand: undefined;
@@ -214,33 +215,57 @@ export function pickReasoningEffort(endpoint: IChatEndpoint | undefined, request
 }
 
 function buildConfigurationSchema(endpoint: IChatEndpoint): vscode.LanguageModelConfigurationSchema | undefined {
+	const properties: Record<string, NonNullable<vscode.LanguageModelConfigurationSchema['properties']>[string]> = {};
+
+	// Thinking effort
 	const effortLevels = endpoint.supportsReasoningEffort?.filter(
 		(level): level is typeof SUPPORTED_EFFORT_LEVELS[number] =>
 			(SUPPORTED_EFFORT_LEVELS as readonly string[]).includes(level)
 	);
-	if (!effortLevels) {
-		return;
+	if (effortLevels && effortLevels.length > 0) {
+		const defaultEffort = effortLevels.includes('high') ? 'high' : undefined;
+		properties[CLAUDE_REASONING_EFFORT_PROPERTY] = {
+			type: 'string',
+			title: l10n.t('Thinking Effort'),
+			enum: effortLevels,
+			enumItemLabels: effortLevels.map(level => level.charAt(0).toUpperCase() + level.slice(1)),
+			enumDescriptions: effortLevels.map(level => {
+				switch (level) {
+					case 'low': return l10n.t('Faster responses with less reasoning');
+					case 'medium': return l10n.t('Balanced reasoning and speed');
+					case 'high': return l10n.t('Greater reasoning depth but slower');
+				}
+			}),
+			default: defaultEffort,
+			group: 'navigation',
+		};
 	}
 
-	const defaultEffort = effortLevels.includes('high') ? 'high' : undefined;
-
-	return {
-		properties: {
-			[CLAUDE_REASONING_EFFORT_PROPERTY]: {
-				type: 'string',
-				title: l10n.t('Thinking Effort'),
-				enum: effortLevels,
-				enumItemLabels: effortLevels.map(level => level.charAt(0).toUpperCase() + level.slice(1)),
-				enumDescriptions: effortLevels.map(level => {
-					switch (level) {
-						case 'low': return l10n.t('Faster responses with less reasoning');
-						case 'medium': return l10n.t('Balanced reasoning and speed');
-						case 'high': return l10n.t('Greater reasoning depth but slower');
-					}
-				}),
-				default: defaultEffort,
-				group: 'navigation',
-			}
-		}
-	};
+	// Context size — only when CAPI provides a default context max, indicating
+	// a meaningful distinction between default and long context tiers.
+	const pricing = endpoint.tokenPricing;
+	const defaultContextMax = pricing?.default.contextMax;
+	const fullMax = endpoint.modelMaxPromptTokens;
+	if (defaultContextMax && defaultContextMax < fullMax) {
+		const hasLongContextSurcharge = !!pricing?.longContext;
+		properties[CLAUDE_CONTEXT_SIZE_PROPERTY] = {
+			type: 'number',
+			title: l10n.t('Context Size'),
+			enum: [defaultContextMax, fullMax],
+			enumItemLabels: [formatTokenCount(defaultContextMax), formatTokenCount(fullMax)],
+			enumDescriptions: [
+				l10n.t('Default'),
+				hasLongContextSurcharge
+					? l10n.t('Longer sessions')
+					: l10n.t('Longer sessions without compaction'),
+			],
+			default: defaultContextMax,
+			group: 'tokens',
+		};
+	}
+
+	if (Object.keys(properties).length === 0) {
+		return undefined;
+	}
+	return { properties };
 }

extensions/copilot/src/extension/chatSessions/claude/node/claudeLanguageModelServer.ts

@@ -191,6 +191,17 @@ export class ClaudeLanguageModelServer extends Disposable {
 				tokenSource.cancel();
 			});
 
+			// If the user picked a context tier in the model picker, surface that
+			// value directly as the prompt token budget. CAPI's `tokenPricing.default.contextMax`
+			// is documented as a prompt-token limit (matching `endpoint.modelMaxPromptTokens`),
+			// and CAPI bills the tier based on actual prompt size, so no other signaling is
+			// required. Clamp to the default budget so the override never lowers it.
+			const sessionContextSize = sessionId ? this.sessionStateService.getContextSizeForSession(sessionId) : undefined;
+			const defaultPromptBudget = DEFAULT_MAX_TOKENS - DEFAULT_MAX_OUTPUT_TOKENS;
+			const modelMaxPromptTokens = sessionContextSize !== undefined
+				? Math.max(defaultPromptBudget, sessionContextSize)
+				: defaultPromptBudget;
+
 			const endpointRequestBody = requestBody as IEndpointBody;
 			const streamingEndpoint = this.instantiationService.createInstance(
 				ClaudeStreamingPassThroughEndpoint,
@@ -200,7 +211,7 @@ export class ClaudeLanguageModelServer extends Disposable {
 				headers,
 				'vscode_claude_code',
 				{
-					modelMaxPromptTokens: DEFAULT_MAX_TOKENS - DEFAULT_MAX_OUTPUT_TOKENS,
+					modelMaxPromptTokens,
 					maxOutputTokens: DEFAULT_MAX_OUTPUT_TOKENS
 				},
 				sessionId

extensions/copilot/src/extension/chatSessions/claude/node/claudeSessionStateService.ts

@@ -47,6 +47,7 @@ export class ClaudeSessionStateService extends Disposable implements IClaudeSess
 			folderInfo: existing?.folderInfo,
 			usageHandler: existing?.usageHandler,
 			reasoningEffort: existing?.reasoningEffort,
+			contextSize: existing?.contextSize,
 			traceContext: existing?.traceContext,
 			turnId: existing?.turnId,
 		});
@@ -69,6 +70,7 @@ export class ClaudeSessionStateService extends Disposable implements IClaudeSess
 			folderInfo: existing?.folderInfo,
 			usageHandler: existing?.usageHandler,
 			reasoningEffort: existing?.reasoningEffort,
+			contextSize: existing?.contextSize,
 			traceContext: existing?.traceContext,
 			turnId: existing?.turnId,
 		});
@@ -88,6 +90,7 @@ export class ClaudeSessionStateService extends Disposable implements IClaudeSess
 			folderInfo: existing?.folderInfo,
 			usageHandler: existing?.usageHandler,
 			reasoningEffort: existing?.reasoningEffort,
+			contextSize: existing?.contextSize,
 			traceContext: existing?.traceContext,
 			turnId: existing?.turnId,
 		});
@@ -109,6 +112,7 @@ export class ClaudeSessionStateService extends Disposable implements IClaudeSess
 			folderInfo,
 			usageHandler: existing?.usageHandler,
 			reasoningEffort: existing?.reasoningEffort,
+			contextSize: existing?.contextSize,
 			traceContext: existing?.traceContext,
 			turnId: existing?.turnId,
 		});
@@ -128,6 +132,7 @@ export class ClaudeSessionStateService extends Disposable implements IClaudeSess
 			folderInfo: existing?.folderInfo,
 			usageHandler: handler,
 			reasoningEffort: existing?.reasoningEffort,
+			contextSize: existing?.contextSize,
 			traceContext: existing?.traceContext,
 			turnId: existing?.turnId,
 		});
@@ -149,6 +154,29 @@ export class ClaudeSessionStateService extends Disposable implements IClaudeSess
 			folderInfo: existing?.folderInfo,
 			usageHandler: existing?.usageHandler,
 			reasoningEffort: effort,
+			contextSize: existing?.contextSize,
+			traceContext: existing?.traceContext,
+			turnId: existing?.turnId,
+		});
+	}
+
+	getContextSizeForSession(sessionId: string): number | undefined {
+		return this._sessionState.get(sessionId)?.contextSize;
+	}
+
+	setContextSizeForSession(sessionId: string, contextSize: number | undefined): void {
+		const existing = this._sessionState.get(sessionId);
+		if (existing?.contextSize === contextSize) {
+			return;
+		}
+		this._sessionState.set(sessionId, {
+			modelId: existing?.modelId,
+			permissionMode: existing?.permissionMode ?? 'acceptEdits',
+			capturingToken: existing?.capturingToken,
+			folderInfo: existing?.folderInfo,
+			usageHandler: existing?.usageHandler,
+			reasoningEffort: existing?.reasoningEffort,
+			contextSize,
 			traceContext: existing?.traceContext,
 			turnId: existing?.turnId,
 		});
@@ -167,6 +195,7 @@ export class ClaudeSessionStateService extends Disposable implements IClaudeSess
 			folderInfo: existing?.folderInfo,
 			usageHandler: existing?.usageHandler,
 			reasoningEffort: existing?.reasoningEffort,
+			contextSize: existing?.contextSize,
 			traceContext,
 			turnId: existing?.turnId,
 		});
@@ -185,6 +214,7 @@ export class ClaudeSessionStateService extends Disposable implements IClaudeSess
 			folderInfo: existing?.folderInfo,
 			usageHandler: existing?.usageHandler,
 			reasoningEffort: existing?.reasoningEffort,
+			contextSize: existing?.contextSize,
 			traceContext: existing?.traceContext,
 			turnId,
 		});