Skip to content

Modernization continuation: CI unblock, security/privacy hardening, reliability fixes, RAG foundation #63

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Pterjudin merged 4 commits into from
Jun 12, 2026
Merged

Modernization continuation: CI unblock, security/privacy hardening, reliability fixes, RAG foundation #63

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
a6cef82
phase9(ci): unblock Compile & Hygiene (diff .d.ts + appimage hygiene …
Jun 12, 2026
845180d
security/privacy: telemetry opt-IN, local-only source-of-truth, SSRF,…
Jun 12, 2026
7e95d80
reliability: fix SR data-loss, agent-loop bugs, cwd traversal, and re…
Jun 12, 2026
00c2b95
phase4(rag): extract + test the structured retrieval-result core
Jun 12, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions build/filters.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -67,6 +67,10 @@ export const unicodeFilter = Object.freeze<string[]>([
// extensions/copilot has its own code style
'!extensions/copilot/**',

// CortexIDE AppImage packaging script embeds intentional translated (non-ASCII)
// desktop-menu names (Japanese/Korean/Russian/Chinese "New Empty Window" etc.)
'!scripts/appimage/create_appimage.sh',

'!src/vs/base/browser/dompurify/**',
'!src/vs/workbench/services/keybinding/browser/keyboardLayouts/**',
'!src/vs/workbench/contrib/terminal/common/scripts/psreadline/**',
Expand Down Expand Up @@ -143,6 +147,8 @@ export const indentationFilter = Object.freeze<string[]>([
'!**/*.{svg,exe,png,bmp,jpg,scpt,bat,cmd,cur,ttf,woff,eot,md,ps1,psm1,template,yaml,yml,d.ts.recipe,ico,icns,plist,opus,admx,adml,wasm}',
'!build/{lib,download,linux,darwin}/**/*.js',
'!build/**/*.sh',
// CortexIDE AppImage packaging script keeps its own (space) shell indentation
'!scripts/appimage/create_appimage.sh',
'!build/azure-pipelines/**/*.js',
'!build/azure-pipelines/**/*.config',
'!build/npm/gyp/custom-headers/*.patch',
Expand Down
80 changes: 58 additions & 22 deletions src/vs/workbench/contrib/cortexide/browser/chatThreadService.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,7 +29,7 @@ import { CancellationToken } from '../../../../base/common/cancellation.js';
import { ILanguageFeaturesService } from '../../../../editor/common/services/languageFeatures.js';
import { ChatMessage, ChatImageAttachment, ChatPDFAttachment, CheckpointEntry, CodespanLocationLink, StagingSelectionItem, ToolMessage, PlanMessage, PlanStep, StepStatus, ReviewMessage } from '../common/chatThreadServiceTypes.js';
import { selectCompactionWindow } from '../common/compactionPolicy.js';
import { updateConsecutiveToolErrors, computeCompactionOverflowDecision, shouldEscalateModel, decideFileReadGate, type ToolMessageType } from '../common/agentLoopDecisions.js';
import { updateConsecutiveToolErrors, computeCompactionOverflowDecision, shouldEscalateModel, decideFileReadGate, classifyToolStepOutcome, type ToolMessageType } from '../common/agentLoopDecisions.js';
import { isRateLimitErrorMessage } from '../common/providerErrorFormat.js';
import { createSerializer } from '../common/asyncSerializer.js';

Expand Down Expand Up @@ -465,6 +465,8 @@ class ChatThreadService extends Disposable implements IChatThreadService {
@IWorkspaceTrustManagementService private readonly _workspaceTrustService: IWorkspaceTrustManagementService,
) {
super()
// Register the streaming-RAF cancellation ONCE (was re-registered per batch -> unbounded growth).
this._register({ dispose: () => { if (this._streamStateRafId !== undefined) { cancelAnimationFrame(this._streamStateRafId); this._streamStateRafId = undefined; } } })
this.state = { allThreads: {}, currentThreadId: null as unknown as string, openTabs: [] } // default state
// When set for a thread, the next call to _shouldGeneratePlan will return false and clear the flag
this._suppressPlanOnceByThread = {}
Expand Down Expand Up @@ -694,8 +696,7 @@ class ChatThreadService extends Disposable implements IChatThreadService {
this._pendingStreamStateUpdates.clear()
this._streamStateRafId = undefined
})
// Register cancellation so RAF never fires after the service is disposed
this._register({ dispose: () => { if (this._streamStateRafId !== undefined) { cancelAnimationFrame(this._streamStateRafId); this._streamStateRafId = undefined; } } })
// (RAF cancellation is registered ONCE in the constructor -- not per batch.)
}
} else {
// For non-streaming updates (idle, error, etc.), fire immediately
Expand Down Expand Up @@ -3061,6 +3062,7 @@ Output ONLY the JSON, no other text. Start with { and end with }.`
this._fileReadCacheLRU.delete(childId)
this._planCache.delete(childId)
this._pendingStreamStateUpdates.delete(childId)
this._fileOpJournalByThread.delete(childId) // before-content snapshots can be large; don't leak them
delete this._suppressPlanOnceByThread[childId]
delete this.streamState[childId]
}
Expand Down Expand Up @@ -3323,7 +3325,9 @@ Output ONLY the JSON, no other text. Start with { and end with }.`
// Plan mode always generates a plan; agent mode uses heuristics
const shouldGeneratePlan = chatMode === 'plan' || this._shouldGeneratePlan(threadId)
if (shouldGeneratePlan) {
await this._generatePlanFromUserRequest(threadId, modelSelection, modelSelectionOptions)
// Use the RESOLVED selection: prepareLLMChatMessages throws on a literal 'auto', which
// dead-ended Plan mode when Auto was selected (the default).
await this._generatePlanFromUserRequest(threadId, resolvedModelSelection, resolvedModelSelectionOptions)
// CRITICAL: Force cache refresh ONLY here after plan generation
this._planCache.delete(threadId)
const planAfterGen = this._getCurrentPlan(threadId, true) // Force refresh
Expand Down Expand Up @@ -3443,9 +3447,21 @@ Output ONLY the JSON, no other text. Start with { and end with }.`
refreshPlanStep()
}
}
} else {
// Mark step as completed on success
if (activePlanTracking?.currentStep) {
} else if (activePlanTracking?.currentStep) {
// Mark the step from the ACTUAL tool outcome (mirrors the main loop): an errored
// pre-approved callThisToolFirst must be a FAILED step, not silently completed.
const preLoopTailMsg = this.state.allThreads[threadId]?.messages.slice(-1)[0]
const preLoopToolMsg = preLoopTailMsg?.role === 'tool' ? (preLoopTailMsg as ToolMessage<ToolName>) : undefined
const preLoopOutcome = classifyToolStepOutcome(preLoopToolMsg?.type ?? null)
if (preLoopOutcome === 'failed') {
const updatedStep = this._markStepCompletedInternal(threadId, activePlanTracking.currentStep, false, (preLoopToolMsg?.type === 'tool_error' && preLoopToolMsg.result) || 'Tool execution failed')
if (updatedStep) {
activePlanTracking.currentStep = updatedStep
activePlanTracking.planInfo = { plan: updatedStep.plan, planIdx: updatedStep.planIdx }
} else {
refreshPlanStep()
}
} else if (preLoopOutcome === 'succeeded') {
// PERFORMANCE: Use returned step info instead of re-looking up
const updatedStep = this._markStepCompletedInternal(threadId, activePlanTracking.currentStep, true)
if (updatedStep) {
Expand Down Expand Up @@ -3482,6 +3498,7 @@ Output ONLY the JSON, no other text. Start with { and end with }.`
}
}
}
// 'indeterminate' (e.g. invalid_params / rejected): leave the step status as-is, matching the main loop
}
}
this._setStreamState(threadId, { isRunning: 'idle', interrupt: 'not_needed' }) // just decorative, for clarity
Expand Down Expand Up @@ -3677,14 +3694,24 @@ Output ONLY the JSON, no other text. Start with { and end with }.`
contextSize = cached.contextSize;
} else {
// Prepare messages (expensive operation)
const prepResult = await this._convertToLLMMessagesService.prepareLLMChatMessages({
chatMessages: preprocessedMessages,
modelSelection,
chatMode,
repoIndexerPromise: repoIndexerResults ? Promise.resolve(repoIndexerResults) : repoIndexerPromise,
subagentSystemPrompt: runCtx?.systemPromptOverride,
allowedToolNames: runCtx?.allowedToolNames
});
let prepResult: Awaited<ReturnType<typeof this._convertToLLMMessagesService.prepareLLMChatMessages>>;
try {
prepResult = await this._convertToLLMMessagesService.prepareLLMChatMessages({
chatMessages: preprocessedMessages,
modelSelection,
chatMode,
repoIndexerPromise: repoIndexerResults ? Promise.resolve(repoIndexerResults) : repoIndexerPromise,
subagentSystemPrompt: runCtx?.systemPromptOverride,
allowedToolNames: runCtx?.allowedToolNames
});
} catch (prepErr) {
// The first prompt assembly can throw (and has no prior messages to fall back to);
// surface it like the LLM error path instead of ending the turn as a silent "Done".
this._setStreamState(threadId, { isRunning: undefined, error: { message: `Failed to assemble the request: ${getErrorMessage(prepErr)}`, fullError: prepErr instanceof Error ? prepErr : null } });
chatLatencyAudit.releaseContext(finalRequestId); // started at 3659; release so the render-monitor interval can stop
this._addUserCheckpoint({ threadId });
return;
}
messages = prepResult.messages;
separateSystemMessage = prepResult.separateSystemMessage;

Expand Down Expand Up @@ -4072,10 +4099,16 @@ Output ONLY the JSON, no other text. Start with { and end with }.`
const outputTokens = textToCount.length > 0 ? Math.max(1, Math.ceil(textToCount.length / 3.5)) : 0
chatLatencyAudit.markStreamComplete(finalRequestId, outputTokens)
// Log metrics for debugging
// PERFORMANCE: Only compute metrics if audit is enabled (metrics computation has overhead)
const metrics = auditEnabled ? chatLatencyAudit.completeRequest(finalRequestId) : null
if (metrics) {
chatLatencyAudit.logMetrics(metrics)
// Always release the context (else it leaks + the 60Hz render-monitor never stops);
// only compute metrics when auditing (getMetrics has overhead).
let metrics: ReturnType<typeof chatLatencyAudit.completeRequest> = null
if (auditEnabled) {
metrics = chatLatencyAudit.completeRequest(finalRequestId)
if (metrics) {
chatLatencyAudit.logMetrics(metrics)
}
} else {
chatLatencyAudit.releaseContext(finalRequestId)
}

// Audit log: record reply
Expand Down Expand Up @@ -4822,15 +4855,17 @@ Output ONLY the JSON, no other text. Start with { and end with }.`
const lastMsg = thread.messages[thread.messages.length - 1]
if (lastMsg && lastMsg.role === 'tool') {
const toolMsg = lastMsg as ToolMessage<ToolName>
if (toolMsg.type === 'tool_error') {
// Shared, tested classification (same fn the pre-loop callThisToolFirst path uses).
const stepOutcome = classifyToolStepOutcome(toolMsg.type)
if (stepOutcome === 'failed') {
// PERFORMANCE: Use returned step info instead of re-looking up
const updatedStep = this._markStepCompletedInternal(threadId, activePlanTracking.currentStep, false, toolMsg.result || 'Tool execution failed')
const updatedStep = this._markStepCompletedInternal(threadId, activePlanTracking.currentStep, false, (toolMsg.type === 'tool_error' && toolMsg.result) || 'Tool execution failed')
if (updatedStep) {
activePlanTracking.currentStep = updatedStep
} else {
refreshPlanStep()
}
} else if (toolMsg.type === 'success') {
} else if (stepOutcome === 'succeeded') {
// PERFORMANCE: Use returned step info instead of re-looking up
const updatedStep = this._markStepCompletedInternal(threadId, activePlanTracking.currentStep, true)
if (updatedStep) {
Expand Down Expand Up @@ -6089,6 +6124,7 @@ We only need to do it for files that were edited since `from`, ie files between
this._fileReadCacheLRU.delete(threadId)
this._planCache.delete(threadId)
this._pendingStreamStateUpdates.delete(threadId)
this._fileOpJournalByThread.delete(threadId) // rollback before-content snapshots can be large; don't leak them
delete this._suppressPlanOnceByThread[threadId]
delete this.streamState[threadId]

Expand Down
21 changes: 21 additions & 0 deletions src/vs/workbench/contrib/cortexide/browser/react/out/diff/index.d.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,21 @@
/*--------------------------------------------------------------------------------------
* Copyright 2025 Glass Devtools, Inc. All rights reserved.
* Licensed under the Apache License, Version 2.0. See LICENSE.txt for more information.
*--------------------------------------------------------------------------------------*/

// Hand-maintained type declarations for the generated diff bundle
// (browser/react/out/diff/index.js, produced by `npm run buildreact`). Committed
// alongside the sibling react/out *.d.ts files (out/ is gitignored, these are force-added)
// so the type checker can resolve `import { diffLines } from '../react/out/diff/index.js'`
// in findDiffs.ts WITHOUT first building react -- CI runs the typecheck before buildreact.
// The runtime impl re-exports diffLines/Change from the `diff` npm package
// (see react/src/diff/index.tsx).

export interface Change {
count?: number;
value: string;
added?: boolean;
removed?: boolean;
}

export declare function diffLines(oldStr: string, newStr: string): Change[];
56 changes: 16 additions & 40 deletions src/vs/workbench/contrib/cortexide/browser/repoIndexerService.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@ import { ISecretDetectionService } from '../common/secretDetectionService.js';
import { CancellationToken } from '../../../../base/common/cancellation.js';
import { OfflineGate } from '../common/offlineGate.js';
import { canEgress } from '../common/egressPolicy.js';
import { formatRetrievalResult, RetrievalResult } from '../common/retrievalResult.js';
import { ITreeSitterService } from './treeSitterService.js';
import { IVectorStore } from '../common/vectorStore.js';
import { ICortexideSettingsService } from '../common/cortexideSettingsService.js';
Expand Down Expand Up @@ -1257,47 +1258,22 @@ class RepoIndexerService extends Disposable implements IRepoIndexerService {
/**
* Format an index entry as a result string with citations
*/
private _formatResult(entry: IndexEntry, chunk: IndexChunk | undefined, query: string): string {
const parts: string[] = [];

// Format file path with line range citation
let fileHeader = `File: ${entry.uri}`;
let citationStartLine = entry.snippetStartLine || 1;
let citationEndLine = entry.snippetEndLine || 1;

if (chunk) {
// Use chunk line numbers if available
citationStartLine = chunk.startLine;
citationEndLine = chunk.endLine;
}

if (citationStartLine === citationEndLine) {
fileHeader += `:${citationStartLine}`;
} else {
fileHeader += `:${citationStartLine}-${citationEndLine}`;
}
parts.push(fileHeader);

// Add symbols if available
if (entry.symbols.length > 0) {
const qLower = query.toLowerCase();
const matchingSymbols = entry.symbols.filter(s =>
s.toLowerCase().includes(qLower) || qLower.includes(s.toLowerCase())
);
const otherSymbols = entry.symbols.filter(s =>
!matchingSymbols.includes(s)
);
const prioritizedSymbols = [...matchingSymbols.slice(0, 5), ...otherSymbols.slice(0, 5)].slice(0, 10);
parts.push(`Symbols: ${prioritizedSymbols.join(', ')}`);
}

// Add snippet or chunk text with citation
const displayText = chunk ? chunk.text : entry.snippet;
if (displayText) {
parts.push(`Content preview:\n${displayText}`);
}
/** Build the structured retrieval hit from an index entry (+ optional chunk). The file URI is a
* discrete field here -- see retrievalResult.ts for why (the old formatted-string-only shape
* broke the consumers that need the path). */
private _buildResult(entry: IndexEntry, chunk: IndexChunk | undefined): RetrievalResult {
return {
uri: entry.uri,
symbols: entry.symbols,
snippet: chunk ? chunk.text : entry.snippet,
startLine: chunk ? chunk.startLine : (entry.snippetStartLine || 1),
endLine: chunk ? chunk.endLine : (entry.snippetEndLine || 1),
};
}

return parts.join('\n');
private _formatResult(entry: IndexEntry, chunk: IndexChunk | undefined, query: string): string {
// Byte-identical to the previous inline formatting -- now via the pure, tested common module.
return formatRetrievalResult(this._buildResult(entry, chunk), query);
}

private _fallbackToContextCache(text: string, k: number, startTime: number, flags: { timedOut: boolean; earlyTerminated: boolean }): Promise<{ results: string[]; metrics: QueryMetrics }> {
Expand Down
Loading
Loading