Changelog

All notable changes to the Alex Cognitive Architecture will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[6.1.8] - 2026-03-05

Doc Alignment Hotfix — Version consistency fix post-meditation.

Fixed

copilot-instructions.md version header aligned to 6.1.8
cognitive-config.json version field updated
Heir CHANGELOG entry for stable release added
Legacy episodic file archived

[6.1.7] - 2026-03-05

Stable Release — Promoted to stable on VS Code Marketplace.

Changed

Published as stable release (previously pre-release)
Version alignment across all documentation

[6.1.5] - 2026-03-04

Heir Protection Fix + Trifecta Gap Closure + M365 Polish — Removed false positive from Master auto-detection, closed trifecta gap from 26→36, M365 orphan cleanup and mobile docs.

Added

10 new complete trifectas (26→36): image-handling, character-aging-progression, visual-memory, code-review, root-cause-analysis, refactoring-patterns, debugging-patterns, security-review, skill-building, global-knowledge
6 new prompts: /rca, /refactor, /debug, /security-review, /skill-building, /knowledge
4 new instructions: root-cause-analysis, refactoring-patterns, debugging-patterns, security-review
2 new skills: agent-debug-panel (7 debug scenarios), terminal-image-rendering (Kitty graphics protocol)
MCP standalone distribution: 704KB esbuild bundle at plugin/mcp/index.js

Changed

TRIFECTA-CATALOG.md: full audit — 10 new entries, Mermaid diagram, health summary updated
M365 USER-MANUAL.md: mobile support FAQ updated (declarative agent works in Teams mobile)
Skill and trifecta counts updated across all documentation (128 skills, 36 trifectas)

Removed

4 orphan M365 plugin files: alex-knowledge-plugin.json, graph-api-plugin.json, openapi.yaml, graph-openapi.yaml

Fixed

alex_docs/ removed from Master Alex auto-detection indicators — heirs with alex_docs/NORTH-STAR.md were falsely identified as Master Alex, blocking Alex: Upgrade (reported by AlexLearn)

[6.1.4] - 2026-03-04

Cognitive Tier Fix + Model Updates — Fixed L4 detection for current extended thinking settings, added GPT-5.3, collapsed sidebar sections by default.

Added

GPT-5.3 added to frontier tier model maps
Frontier regex now auto-catches future GPT-5.x models (gpt-5\.[2-9])

Changed

Cognitive Dashboard and Memory Architecture sidebar sections collapsed by default for cleaner first-run experience
Removed deprecated github.copilot from extensionPack — now only github.copilot-chat
Brain QA output compacted (phases + warnings/summary only, use -Detail for verbose)
Brain QA agent cross-reference regex fixed for Windows CRLF line endings

Fixed

Critical: Cognitive tier detection now reads github.copilot.chat.agent.thinkingTool (current setting path) — was only checking deprecated paths, causing L4 users to show as L3
Visual memory heir sync template updated with voices/videoStyles sections
cognitive-config.json version synced to 6.1.3

[6.1.2] - 2026-03-03

Replicate Trifecta Updates + Documentation Hygiene — Model landscape refresh across 5 skills, 2 new trifectas completed, doc archive pass 2.

Added

image-handling trifecta completed — instruction + prompt files created
character-aging-progression trifecta completed — instruction + prompt files created

Changed

Updated Replicate model landscape across 5 skills: Ideogram v3 (turbo/balanced/quality), nano-banana-2, flux-kontext-pro/max, flux-2-pro/max, recraft-v4, veo-3.1-fast, sora-2, qwen/qwen3-tts, minimax/speech-2.8-turbo
ideogram-v3-turbo ($0.03) is now the default typography recommendation (63% cheaper than v2)
Archived 6 outdated docs to alex_archive/: 3 point-in-time audit reports, REPLICATE-EVALUATION.md (superseded by skills), Ideogram v2 gallery, stale brain-qa-output.txt

Fixed

character-aging-progression/SKILL.md: image_input must be an array [dataURI], not a single string — was silently breaking face consistency
Audio model reference updated: qwen-tts → qwen/qwen3-tts across all skills
Skill frontmatter: removed unsupported applyTo/triggers attributes, fixed name to kebab-case in 3 SKILL.md files

[6.1.0] - 2026-03-03

Environment & Cognitive Tier Hardening — Extended thinking detection fix (critical), extension dependency checking, multi-account GitHub detection, cognitive tier real-time refresh.

Added

Extension dependency checking with one-click install for GitHub Copilot, Copilot Chat, Mermaid Preview
Multi-account GitHub detection (personal + enterprise) in cognitive tier
Account-aware setup guidance with upgrade tips for Frontier model access
Real-time cognitive tier refresh on Copilot settings changes

Fixed

CRITICAL: Extended thinking detection was reading wrong config keys — Level 4 (Advanced) was unreachable for all users
L4 badge race condition causing welcome view to default to Level 1
Misleading account classification (business → unknown)
Markdown in plain text dialogs, unsafe sign-in command

Changed

Package size reduced: 534 → 525 files (~3.4 MB lighter)

[6.0.3] - 2026-03-02

Fixed

Extension README: resolved 4 Marketplace compliance issues — self-referential link (line 125), broken article/versions/13-ALEX-MANIFESTO-PERSONAL.md links (lines 300 & 358), broken replicate-api skill path (line 671)

Changed

Root README: DRY refactor — removed 700-line duplicate of extension README; replaced with compact Features table linking to learnalex.correax.com
Both READMEs: removed "Chat with Alex" section (superseded by learnalex.correax.com website)
Docs: replaced all GitHub Gist references with private GitHub repo throughout

[6.0.2] - 2026-02-28

Fixed

Brand doc (DK-correax-brand.md): updated ghost watermark canonical opacity spec 0.03→0.15 (CSS + HTML example)
correax-brand skill: encoded rgba(255,255,255,0.15) as the authoritative design token to prevent regression

Changed

Catalogs regenerated: 23 complete trifectas (was 14), 125 skills, 593 connections
vscode-configuration-validation synapses.json inheritance field repaired (malformed object → "heir:vscode")
Active Context updated to reflect v6.0.1 release state

[6.0.1] - 2026-02-28

Fixed

Banner: increased ALEX watermark visibility (opacity 0.03→0.15)

[6.0.0] - 2026-02-28

The Partnership Release — Six deep-integration features that make Alex a cognitive partner, not just a chat interface. Alex now remembers your sessions, learns from outcomes, detects your work context autonomously, guides multi-step workflows, adapts to your expertise level, and proactively triggers code reviews.

Added

Episodic Memory (services/episodicMemory.ts) — Persistent session records at ~/.alex/episodic/sessions.json. Alex recalls "what were we building last Tuesday?" across sessions. Commands: alex.recallSession, alex.showSessionHistory
Outcome Learning Loop (services/outcomeTracker.ts) — 👍/👎 tracking per @alex exchange with per-domain confidence scoring. Alex calibrates advice certainty from observed outcomes. Commands: alex.recordPositiveOutcome, alex.recordNegativeOutcome, alex.showOutcomeStats
Autonomous Task Detection (services/taskDetector.ts) — Proactively surfaces stalled work and TODO hotspots from peripheral vision observations on a 30-minute interval. 4-hour dismiss cooldown. Commands: alex.showPendingTasks, alex.forceCheckTasks
Multi-Step Workflow Engine (services/workflowEngine.ts) — JSON workflow definitions at .alex/workflows/. Includes 4 built-in workflows: Plan→Build→Review, Debug→Fix→Verify, Research-First Development, Release Preparation. Commands: alex.runWorkflow, alex.listWorkflows
User Expertise Model (services/expertiseModel.ts) — Domain interaction frequency across 10 domains (coding, debugging, architecture, testing, devops, data, security, ai-ml, documentation, general) mapped to novice/intermediate/advanced/expert. Injects calibration hint into every @alex prompt via PromptContext.expertiseHint. Command: alex.showExpertiseModel
Proactive Code Review Triggers — Built into taskDetector.ts: on save, debounces 60s and checks git diff --stat HEAD. If >200 lines changed, surfaces a code review nudge with 30-minute cooldown
Layer 10b in promptEngine.ts — buildExpertiseLayer() injects user expertise calibration for every @alex interaction
10 new VS Code commands registered and declared in package.json
Image Upscaling (commands/contextMenu.ts, services/replicateService.ts) — 4 Replicate-powered upscaling models (real-esrgan, swinir, codeformer, clarity-upscaler). Right-click any image → Upscale Image with AI. Saves to assets/upscaled/. Command: alex.upscaleImage
MCP Cognitive Tools Package (packages/mcp-cognitive-tools/) — Standalone MCP server publishing 5 cognitive tools (alex_synapse_health, alex_memory_search, alex_architecture_status, alex_knowledge_search, alex_knowledge_save) usable by any MCP-compatible host. Entry: npx @alex/mcp-cognitive-tools
Learning Journeys (.github/prompts/journey.prompt.md) — 8 curated role-based learning paths (frontend-developer, backend-developer, fullstack-developer, devops-engineer, technical-writer, researcher, ai-engineer, alex-architect) with progress tracking at .alex/journeys/*.json
Presentation Automation (.github/prompts/marp.prompt.md, .github/prompts/presentation.prompt.md) — Unified /presentation router dispatches to Gamma (marketing decks), Marp (tech talks), or PptxGenJS (data reports). Alex-branded Marp CSS theme included
3 new slash commands (/journey, /marp, /presentation) registered in package.json chatParticipants

Changed

3_generate@8 group collision fixed — generateDiagram moved from 3_generate@8 to 3_generate@11 in explorer context menu to resolve conflict with generateAIImage
Editor context menu expanded — Added editImageWithPrompt (4_create@3) and upscaleImage (4_create@4) to alex.contextMenu editor right-click group
PromptContext interface — Added expertiseHint?: string field (v6.0.0 expertise calibration)
participant.ts — @alex handler now appends to episodic draft and records domain interactions after every response (fire-and-forget)
extension.ts — Imports and initializes all 5 new services; registers 10 new commands; flushes episodic draft on deactivation
package.json — v6.0.0, 10 new command declarations

[5.9.13] - 2026-02-28

CorreaX Brand Edition — Full brand system deployment (42/42 tasks). Multi-size favicons, apple-touch-icons, CorreaX dark neural network banner, palette applied to extension source, brand compliance scan in dream protocol, and correax-brand trifecta.

Added

correax-brand trifecta — New skill, instructions, and synapses encoding the full CorreaX design system (palette, typography, banner patterns, asset naming conventions)
north-star trifecta — Skill, instructions, and prompts encoding the North Star alignment framework and evaluation criteria
Multi-size favicon.ico — 7-size ICO bundle (16/24/32/48/64/128/256px) for vscode-extension property
apple-touch-icon.png — 180×180px Apple touch icon for all web-facing Alex properties
CorreaX dark banner — Neural network motif artwork deployed as extension marketplace banner
Brand compliance scan — Automated CorreaX palette validation added to dream protocol Phase 4.5
Episodic memory — meditation-2026-02-28-v6-planning.md consolidating v6.0 execution order, heir sync decisions, and next-steps roadmap

Changed

CorreaX palette in extension source — colors.ts, Global Knowledge SVG assets, and skill files updated to use CorreaX brand tokens (Midnight Slate, Plasma Violet, Neural Cyan, Quantum Gold)
Mermaid theme — CorreaX theme snippet added; architecture diagrams use brand colors
Badge color standardization — All README badges aligned to CorreaX palette
BRANDING-PLAN.md archived — Replaced with clean pointer; full 1091-line task history moved to archive/BRANDING-PLAN-2026-02-28-COMPLETED.md
ROADMAP-UNIFIED.md updated — v5.9.13 recorded as current, v6.0 partnership release as next milestone
All heirs synced to v5.9.13 — AlexLearn, Extensions, AIRS Enterprise Active Contexts aligned to post-CorreaX state

[5.9.12] - 2026-02-26

Documentation Hygiene Edition — Comprehensive documentation audit eliminating version drift, stale file counts, and outdated dates across 8 files. New doc-hygiene→release-process synapse ensures future releases trigger documentation sweeps.

Changed

README.md alignment — Instructions count 52→55, prompts 34→35, skills 123→124, footer version v5.9.9→v5.9.12, removed duplicate copyright line
TEST-GUIDE.md full update — 12 stale v5.9.9 references updated to v5.9.12 across install commands, test expectations, section headers, and known gaps
Operational docs dates — CONTRIBUTING.md, PRE-PUBLISH-CHECKLIST.md, PUBLISHING.md dates updated from February 15 to February 26
Heir VS Code README counts — Instructions 28→55, prompts 17→35 — eliminated 2x count divergence
alex_docs version references — TEST-ACTIVATION-PASS.md (v5.9.9→v5.9.12), VSCODE-SOURCE-INTEGRATION-ANALYSIS.md (v5.9.10→v5.9.12)

Added

New synapse: doc-hygiene → release-process (0.80 strength) — encodes the lesson that release workflows should trigger documentation hygiene sweeps to prevent systemic drift
Episodic memory — meditation-2026-02-26-documentation-hygiene-pass.md consolidating 4 key learnings about documentation drift patterns

[5.9.11] - 2026-02-26

Post-Publish Synapse Hardening — Meditation-driven synapse integrity fixes, relative path normalization, and Brain-QA integration as publish gate.

Changed

Synapse path normalization — Fixed 11 relative synapse paths (../, ../../) to full .github/ paths in anti-hallucination, documentation-quality-assurance, and global-knowledge-maintenance synapses. Result: 576 synapses, 566 valid filesystem targets, 10 intentional external refs, 0 broken
Brain-QA as publish gate — Added cognitive-health-validation synapse to automated-quality-gates.instructions.md, formalizing Brain-QA 35-phase validation as a pre-publish deep audit requirement
Active Context updated — Recent field reflects v5.9.10 NASA Edition publish completion

Added

Episodic memory — meditation-2026-02-26-v5910-nasa-publish.md consolidating 6 key learnings from the publish hardening session
Global Knowledge insights — GI-seven-gate-publish-hardening-protocol and GI-marketplace-readme-link-resolution saved as cross-project patterns

[5.9.10] - 2026-02-26 — NASA Edition 🚀

NASA Standards & Mission-Critical Compliance — Adopt NASA/JPL Power of 10 code quality rules for mission-critical software development. Extension code audited and made compliant with bounded recursion and loop limits.

Added

NASA/JPL Power of 10 standards integration — New .github/instructions/nasa-code-standards.instructions.md adapts NASA's mission-critical code quality rules for TypeScript, enabling high-reliability software development
Builder agent NASA mode — Builder agent auto-detects mission-critical projects and applies NASA standards: bounded recursion, fixed loop bounds, function size limits, assertion density, and more
Code review NASA checklist — code-review-guidelines.instructions.md now includes mission-critical review checklist with blocking severity for R1-R3 violations
Heir project NASA guidance — heir-project-improvement.instructions.md includes mission-critical pre-phase checklist for heirs developing safety-critical software
workspaceFs utility module — New src/shared/workspaceFs.ts providing async wrappers around vscode.workspace.fs API: pathExists, readFile, writeFile, readJson, writeJson, ensureDir, readDirectory, stat, copyFile, rename
Terminal sandboxing documentation — macOS/Linux security note added to SECURITY.md, copilot-instructions.md, and settings.json for chat.tools.terminal.sandbox.enabled

Changed

North-star trifecta — New complete trifecta (skill + instruction + prompt) for North Star alignment principles
North-star synapses — 8 ambiguous synapse targets resolved to full .github/skills/*/SKILL.md paths, 3 yields also fixed
README marketplace links — 18 broken relative links (alex_docs/, article/, .github/SUPPORT.md) converted to absolute GitHub URLs for VS Code Marketplace rendering. Replicate skill documentation link corrected from non-existent replicate-api/ to ai-generated-readme-banners/resources/
M365 heir version alignment — package.json (5.9.3→5.9.10), manifest.json (5.9.8→5.9.10), README badge updated
GitHub Copilot Web heir — Version bumped from 5.9.9 to 5.9.10
Brain-QA 35-phase validation — 34 phases passed, 2 self-containment failures found and fixed (north-star SKILL.md had relative alex_docs/ links → converted to GitHub URLs)
NASA R1 compliance: Bounded recursion — findMdFilesRecursive() in synapse-core.ts now has maxDepth parameter (default: 10) preventing stack overflow from deeply nested directories
NASA R1 compliance: Bounded walk functions — collectSystemFiles.walk() and getFolderSize.walk() in upgrade.ts now have MAX_WALK_DEPTH=10 depth parameter preventing unbounded directory traversal
NASA R2 compliance: Fixed loop bounds — Upgrade dialog loop in upgrade.ts now has MAX_DIALOG_ITERATIONS safety limit (100)
NASA R2 compliance: Text chunking bounds — splitTextIntoChunks() in ttsService.ts now has MAX_ITERATIONS counter preventing potential infinite loops on malformed input
NASA R5 compliance: Entry assertions — Added nasaAssert() calls to resolveWebviewView() in welcomeView.ts validating webviewView and extensionUri
Full codebase NASA audit — All 72 TypeScript files audited against NASA Power of 10 rules; 3 violations found and fixed, 10 functions documented as acknowledged exceptions with architectural justifications
Testing strategies NASA integration — testing-strategies SKILL.md updated with bounded behavior testing patterns (R1/R2/R3), assertion coverage testing (R5), and critical path coverage targets
Code review NASA integration — code-review SKILL.md updated with mission-critical review section: blocking violations (R1-R3), high priority (R4/R5/R8), medium priority (R6/R7/R9/R10)
Heir sync restored — 3 missing skills, 1 instruction, 1 prompt synced to vscode-extension heir (124/124 skills aligned)
fs-extra → vscode.workspace.fs migration — Per ADR-008 (Workspace File API Strategy), migrated all workspace-scoped file operations from Node.js fs-extra to VS Code's native vscode.workspace.fs API for virtual filesystem compatibility (SSH, WSL, Codespaces, containers). Files migrated:
- promptEngine.ts — Brain file reading
- activeContextManager.ts — Protected marker and instructions reading
- synapse-core.ts — Memory file scanning, synapse repair, report saving
- cognitiveDashboard.ts — Skill/instruction/prompt/episodic counting
- memoryDashboard.ts — Memory stats collection
- healthDashboard.ts — Health category scanning
- utils.ts — Version reading, Alex installation check, synapse health scan
- personaDetection.ts — Workspace structure scanning, package.json reading, profile updates
- emotionalMemory.ts — Emotional session logging
- honestUncertainty.ts — Calibration logging, feedback tracking
- tools.ts — Synapse health, memory search, architecture status tools
- sanitize.ts — Config backup operations

Removed

M365 scenario_models — Removed unsupported scenario_models property from declarativeAgent.json (added in 5.9.9, blocked schema v1.6 validation)
Enterprise secrets scanning — alex.enterprise.scanSecrets, alex.enterprise.scanWorkspace commands removed (did not work as expected)
Enterprise audit logging — alex.enterprise.viewAuditLog, alex.enterprise.exportAuditLog commands removed
Enterprise settings — All 11 alex.enterprise.* settings removed from package.json
Enterprise module — src/enterprise/ folder deleted (auditLogging.ts, secretsScanning.ts, index.ts)
Unused fs-extra import — Removed from contextMenu.ts (was importing but not using)

Fixed

North-star SKILL.md self-containment — 2 relative ../../../alex_docs/ links replaced with full GitHub URLs (caught by Brain-QA Phase 34)
TESTE directory excluded — Test debris directory (scripts, images) added to .gitignore
@ts-ignore removal — Replaced all @ts-ignore comments with type-safe patterns:
- inheritSkill.ts — QuickPick custom data now uses Map instead of property injection
- proposeSkill.ts — Same pattern, plus new HeirSkill interface for type-safe skill operations
Type safety improvements — Eliminated any types:
- healthDashboard.ts — any[] → LearningGoal[] for goals parameter
- cognitiveDashboard.ts — Goals filtering now type-safe with WorkspaceGoalsData
- uxFeatures.ts — Same pattern for daily briefing goals
DRY type consolidation — Moved WorkspaceGoal/WorkspaceGoalsData interfaces to shared/constants.ts, eliminating duplication across cognitiveDashboard.ts and uxFeatures.ts

Technical Notes

Files intentionally kept with fs-extra (per ADR-008 — global paths require Node.js filesystem):

session.ts, goals.ts, globalKnowledge.ts, forgettingCurve.ts — Use ~/.alex/ global paths
setupGlobalKnowledge.ts, exportForM365.ts — Symlinks and OneDrive paths
inheritSkill.ts, proposeSkill.ts — Mixed global/workspace operations
logoService.ts, pptxGenerator.ts, audioPlayer.ts — Sync methods for bundled assets

[5.9.9] - 2026-02-24

Platform Architecture Reinforcement — Harvest everything VS Code 1.109 and M365 extensibility GA'd. Skill frontmatter gating, agent orchestration hierarchy, quality gate hooks, Claude Code bridge, and M365 plugin schema upgrade. No proposed APIs, ships clean.

Added

Skill Frontmatter Gating

disable-model-invocation: true added to 6 action skills: meditation, meditation-facilitation, dream-state, self-actualization, brain-qa, release-process. These require explicit user invocation — the model will not self-invoke them during normal conversation.
user-invokable: false added to 16 domain skills: all Azure/M365/platform skills including azure-architecture-patterns, azure-deployment-operations, microsoft-graph-api, vscode-extension-patterns, mcp-development, and more. They load contextually but stay hidden from the / command menu.

Agent Orchestration Hierarchy

agents: frontmatter added to all 6 specialist agents, formalizing valid subagent relationships. Researcher can call Builder + Validator. Builder can call Validator. Validator can call Documentarian. Azure + M365 can call Researcher. Alex (orchestrator) was already pre-configured.

Quality Gate Hooks (pre-tool-use.js)

Q1 — Version drift check: Before any publish command (vsce publish / npm publish), the hook compares package.json version against the version in copilot-instructions.md and warns if they differ. Enforces Definition of Done item 5.
Q2 — TypeScript compile reminder: On .ts file edits, emits a reminder to run npm run compile. Surfaces errors at edit time, not at publish time.
Both checks are non-blocking — they warn in output but don't prevent execution.

Claude Code Compatibility Bridge

.claude/CLAUDE.md — Project orientation document for Claude Code sessions. Points to .github/ as source of truth, lists Safety Imperatives I1–I5, and documents build commands.
.claude/settings.json — Claude Code settings: maps contextPaths to Alex's .github/ assets, wires preToolUse hook, sets tool permissions (allow .github/ writes, deny force-push and direct publish), and sets ALEX_WORKSPACE=master env.

VS Code Settings

chat.agentCustomizationSkill.enabled: false — Disables VS Code 1.109's built-in agent customization skill to prevent it from overriding Alex's vscode-extension-patterns and skill-development skills.

M365 Heir — Extensibility Platform Harvest

Plugin schema v2.4 — Both alex-knowledge-plugin.json and graph-api-plugin.json upgraded from v2.3 to v2.4. Unlocks MCP server runtimes type (prerequisite for v6.0 MCP bridge path).
getMeetingAiInsights — New function in graph-api-plugin.json. Uses Graph v1.0 GA endpoint GET /me/online-meetings/{meetingId}/aiInsights to return structured actionItems, meetingNotes, and mentions from meeting recordings. Wired into capabilities and run_for_functions.
Scenario models routing — scenario_models added to declarativeAgent.json: cognitive_deep routes meditation/self-actualization/architecture operations to GPT-4o; productivity_light routes calendar/email/presence lookups to GPT-4o-mini.
Conversation starters expanded 7 → 12 — Added: "🗓️ What's on my plate?", "🧠 Self-actualization", "🔍 Search my knowledge", "🎯 Sync my goals", "💡 Get AI insights and action items from my last meeting".

Fixed

Avatar revert mandate — Added **IMPORTANT** instruction to copilot-instructions.md, meditation.instructions.md, and dream-state-automation.instructions.md requiring alex_cognitive_state_update with state: "persona" as the final step of every dream/meditate session. Propagated to both vscode-extension and github-copilot-web heirs.
.claude/ heir sync — sync-architecture.cjs now copies .claude/CLAUDE.md + .claude/settings.json to platforms/vscode-extension/.claude/ on every build. Previously the Claude bridge was master-only.
cognitive-config.json version drift — Bumped version and alex_version fields from 5.9.3 to 5.9.9.
Broken synapse repair — Fixed stale reference in meditation-2026-02-20-avatar-system-completion.md: 2026-02-20-stabilization-meditation.md → meditation-2026-02-20-stabilization.md.
package-lock.json version drift — Bumped both version entries from 5.9.8 to 5.9.9.

Documentation

Activation pass test guide — alex_docs/guides/TEST-ACTIVATION-PASS.md: 40 checks across 9 phases (extension present, status bar, 13 core commands, views, chat participant, LM tools, background services, error tolerance, avatar revert). Pass threshold: all 40 green; Phase 1–5 fail or CRITICAL error = release block.

Skill Enhancements

image-handling skill — Replicate model selection — Added comprehensive AI image generation guidance: 7-model comparison table (Flux Schnell, Flux Dev, Flux 1.1 Pro, Ideogram v2, Ideogram v2 Turbo, SDXL, Seedream 5 Lite) with costs and use cases; model selection decision guide keyed to user intent; LoRA support reference; aspect ratio reference. 14 new trigger words added including flux schnell, flux dev, ideogram, sdxl, seedream, text in image, replicate model. Enables Alex to route image generation requests to the correct Replicate model automatically.

Architecture Improvements

`replicateService.ts` — Proper Replicate Service Layer

src/services/replicateService.ts — New dedicated service extracted from inline code in contextMenu.ts. Provides: 7-model catalog with REPLICATE_MODELS constant (all IDs verified live via API — no stale version hashes); generateImage() high-level function with Prefer: wait + polling fallback; createPrediction() and pollPrediction() for direct API access; downloadImageToWorkspace() using vscode.workspace.fs (sandbox-safe); selectModelForPrompt() intent-to-model router; buildModelQuickPickItems() with recommended model highlighting.
generateAIImage command updated — Now uses replicateService: shows all 7 models with ✨ Recommended marker on the best-fit model for the user's prompt (e.g., typing "logo with text" auto-recommends Ideogram v2). Replaced stale version hashes with model-based endpoint. Added 3:2 Landscape aspect ratio.
editImageWithPrompt command updated — Refactored to use createPrediction + pollPrediction from service; replaced fs.readFile with vscode.workspace.fs.readFile; replaced fs.ensureDir + downloadImage with downloadImageToWorkspace.
ADR-007 status — Replicate replaces DALL-E as the image generation backend. Runtime image generation now live. Image upscaling + FLUX brand fine-tune remain P2 backlog items.

`fs-extra → vscode.workspace.fs` Migration (ADR-008)

3 files migrated — contextMenu.ts, fileWatcher.ts, healthCheck.ts now use vscode.workspace.fs for all workspace-scoped file operations. Global-path files (~/.alex/) intentionally kept on fs-extra per ADR-008.
fileWatcher.ts — loadObservations() and persist() now use vscode.workspace.fs.readFile / createDirectory / writeFile. Sync countTodos() switched from fs-extra to native Node.js readFileSync. fs-extra import removed entirely.
healthCheck.ts — fs.pathExists(alexPath) replaced with vscode.workspace.fs.stat() + try/catch. fs.readFile(file.fsPath) replaced with vscode.workspace.fs.readFile(file) + TextDecoder. fs-extra import removed entirely.
contextMenu.ts — Episodic insight saves, SVG illustration saves, AI image downloads all migrated. Inline Replicate API functions removed (now in replicateService.ts). https import removed. Two remaining fs-extra usages (for legacy code paths) documented.

Semantic Skill Graph — ROADMAP Detail

ROADMAP ### 🧠 Semantic Skill Graph section added with full 4-phase breakdown: Phase 1 (PoC standalone script, validates approach at $0.002), Phase 2 (extension integration with alex.recompileSkills command), Phase 3 (synapse discovery dashboard), Phase 4 (global knowledge integration). Key design decisions documented: keyword fallback always kept, compiled graph is cached JSON (loads in <50ms), Phase 1 is the abandonment gate.

[5.9.8] - 2026-02-21

Background File Watcher — Silent ambient observer. Alex now silently tracks which files you keep returning to, what work is sitting uncommitted, and where your TODO backlog is building up — and weaves that awareness into every response.

Added

Background File Watcher — Ambient Workspace Observation

fileWatcher.ts — New module implementing the background file observer. Zero UI, zero notifications, zero interruptions. Runs quietly from activate() and writes observations to .github/episodic/peripheral/file-observations.json.
Hot file tracking — vscode.window.onDidChangeActiveTextEditor increments an in-memory open-count log per file. Files opened ≥5 times in a rolling 7-day window are classified as "hot". Timestamps are pruned on every flush.
Stalled work detection — On every write-debounced flush (2s), git status --porcelain is run to capture files that are modified on disk but not yet committed. Capped at 10 files; ignored directories (node_modules, .git, dist, etc.) are excluded.
TODO/FIXME hotspot scanning — On each flush, the 15 most-recently-opened files are scanned for TODO/FIXME/HACK/XXX comments. Top 5 by count are stored as todoHotspots[]. String scanning is synchronous and fast on source files.
registerFileWatcher(context, workspaceRoot) — Exported function called from extension.ts after registerChatParticipant. Returns a vscode.Disposable pushed onto context.subscriptions for clean deactivation.
loadPeripheralObservations(workspaceRoot) — Async function that reads the persisted JSON. Called concurrently in gatherPeripheralContext() alongside peer project discovery and recent file scan.
PeripheralObservations type — { hotFiles, stalledFiles, todoHotspots, lastUpdated }. TodoHotspot carries file, todoCount, scannedAt.
PeripheralContext.fileWatcherObservations? — New optional field added to PeripheralContext in peripheralVision.ts.
Layer 8 rendering — buildPeripheralVisionLayer in promptEngine.ts now renders a Focus Patterns block when observations exist: hot files, uncommitted files, and TODO hotspot list with counts.
Bootstrap from disk — On first activation, existing persisted observations seed the in-memory open-log so previous-session hot files survive restarts.

[5.9.7] - 2026-02-21

P2 Feature Completion — All remaining actionable P2 items shipped across Peripheral Vision, Honest Uncertainty, and The Forgetting Curve. Alex now notices outdated dependencies mid-conversation, knows when tests last ran, and learns from your 👍/👎 signals.

Added

User Feedback Loop — Epistemic Calibration Signal

FeedbackEntry type — New type in honestUncertainty.ts: { date, topic, level, helpful }. Records the correlation between Alex's confidence level and user satisfaction.
recordCalibrationFeedback() — Fire-and-forget append to .github/episodic/calibration/feedback-log.json (500-entry rolling window). Called from onDidReceiveFeedback in registerChatParticipant().
Native VS Code 👍/👎 wired to calibration — onDidReceiveFeedback now reads coverageLevel + coverageTopic from the result metadata and persists a FeedbackEntry. Over time, this reveals which domains Alex systematically under- or over-estimates.
Coverage metadata in result — General handler return now includes coverageLevel and coverageTopic fields in IAlexChatResult.metadata, making them available to the feedback handler without additional lookups.
Low/uncertain followup shortcuts — When coverage is low or uncertain, alexFollowupProvider adds /saveinsight and /knowledge <topic> followup buttons to help the user contribute knowledge that fills the gap.

Dependency Freshness Tracker

getDependencyFreshness(workspaceRoot) — New export from peripheralVision.ts. Runs npm outdated --json (10s timeout, 5-minute per-workspace cache). Handles npm's non-zero exit code on outdated packages by parsing stdout from the thrown error. Returns DependencyFreshnessResult with classified package list and error field if scan failed.
DependencyFreshnessResult + OutdatedPackage types — OutdatedPackage carries name, current, wanted, latest, and severity (major/minor/patch) derived from semver diff. Sorted most-breaking-first.
Lazy execution — getDependencyFreshness is called inside gatherPeripheralContext only when package.json exists in the workspace root. Skipped silently for non-npm projects.
Layer 8 rendering — buildPeripheralVisionLayer in promptEngine.ts now surfaces: "all packages up to date ✅" or count breakdown by severity + top-3 package names with current→latest diff.

Test Runner Awareness

getTestRunnerStatus(workspaceRoot, framework?) — New export from peripheralVision.ts. Reads well-known test result files: .jest-test-results.json, test-results.json (Vitest JSON reporter), coverage/coverage-summary.json. Returns TestRunnerStatus with lastRunAt, daysSinceLastRun, totalTests, failedTests, passRate, lastRunPassed.
TestRunnerStatus type — Structured result with all run metrics. null fields when data isn't available (framework known but no results file on disk).
Layer 8 rendering — When test results are available: ✅/❌ 123 tests | 2 failed (98.4% pass) | last run 1.2d ago. When no results file exists: jest detected | no results on disk.
Wired into PeripheralContext — Two new optional fields: dependencyFreshness? and testRunnerStatus?.

Changed

peripheralVision.ts — Doc comment updated to v5.9.7; mentions dependency freshness, test runner results, and the 10s npm timeout.

[5.9.6] - 2026-02-21

The Forgetting Curve — Graceful knowledge decay. Living memory stays sharp; unused insights fade toward cold storage — not deleted, deliberately forgotten.

Added

The Forgetting Curve — Graceful Knowledge Decay

forgettingCurve.ts — New module implementing usage-weighted freshness scoring for all global knowledge entries. The core metaphor: memory is not a filing cabinet — what gets reinforced grows stronger, what fades can be recovered but no longer crowds the active workspace.
computeFreshnessScore() — Composite score (refScore × 0.6) + (recencyScore × 0.4). Reference score saturates at 20 uses. Recency score decays logarithmically: 1 / (1 + log10(1 + daysSince / halfLife)). Returns freshness label: thriving (>0.6), active (0.3–0.6), fading (0.1–0.3), dormant (<0.1), permanent (never decays).
Four decay profiles — aggressive (14-day half-life, debugging/project-specific knowledge), moderate (60d, most domain insights), slow (180d, architecture/security/patterns), permanent (core principles, never archived). Auto-assigned by knowledge category; overridable per entry via decayProfile field.
IGlobalKnowledgeEntry extended — Added four optional freshness fields to the shared interface: lastReferenced, referenceCount, freshnessScore, decayProfile. Backward-compatible — all fields are optional, existing entries without them score conservatively.
Reference counting — queueReferenceTouch(entryId) wired into searchGlobalKnowledge in globalKnowledge.ts. Fire-and-forget batch queue (15-entry threshold or 30s timeout) flushes accumulated counts to index.json — never blocks the search path, never contends on disk I/O.
getDecayReport() — Reads the full knowledge index, computes freshness for every entry, returns a DecayReport with top-10 thriving/active and worst-5 fading/dormant entries, plus permanent count. Called during self-actualization Phase 5.5 concurrently with the calibration summary.
Meditation Knowledge Freshness section — Self-actualization session records now include a 📉 Knowledge Freshness section: distribution counts, dormant entry names with scores, and a recommendation to run Dream for cold storage if dormant entries exist.
runForgettingCeremony(workspaceRoot, threshold?) — Dream cycle forgetting ceremony. Moves entries below the freshness threshold from insights/ or patterns/ to insights/archive/ or patterns/archive/. Entries with permanent profile are never moved. Logs the transition to .github/episodic/forgetting-ceremony-{date}.md. Nothing is ever deleted — only moved.
Archive logging — Forgetting ceremony produces a human-readable episodic record listing every archived entry with its reason (score, reference count, days since last use). Users can review and restore any entry by moving the file back.

— Calibrated epistemic humility. Alex now knows what it doesn't know, and says so with precision.

Added

Honest Uncertainty — Knowledge Coverage Scoring

honestUncertainty.ts — New module implementing the Honest Uncertainty framework. scoreKnowledgeCoverage(query, workspaceRoot) searches global patterns, insights, and local .github/skills/ to determine how well the knowledge base covers the current query.
Four confidence levels — high (2+ pattern matches or skill match), medium (1 pattern or 2+ insights), low (1 insight only), uncertain (no knowledge base coverage). Each level maps to a named behavioral instruction, not a badge.
Behavioral signal injection — Layer 11 in promptEngine.ts (buildHonestUncertaintyLayer) injects a confidence signal that shapes how Alex phrases responses: 🟢 respond with confidence, 🟡 use qualified language, 🟠 flag thin coverage, 🔴 reason from first principles and say so honestly. Never a visible number or badge.
Skill name matching — Local .github/skills/ directory is scanned for folder names matching query terms. Skill matches bump confidence one tier (curated + tested knowledge).
whatINeed transparency — For low and uncertain levels, CoverageScore.whatINeed is populated and injected into the prompt: when a user asks what would help, Alex responds with specific, actionable asks (working example, error output, docs, or spec).
Calibration log — Every scored request is fire-and-forget logged to .github/episodic/calibration/calibration-log.json (rolling 500-entry window). Persists: date, topic summary, confidence level, match count.
Meditation calibration review — getCalibrationSummary() surfaces confidence distribution + uncertain topic clusters in the Phase 5.5 self-actualization session record. Imported by self-actualization.ts, runs concurrently with emotional review.
Concurrent execution — scoreKnowledgeCoverage runs concurrently with gatherPeripheralContext in participant.ts via Promise.all — zero added latency to the response path.

[5.9.4] - 2026-02-21

Avatar System Completion + Emotional Intelligence (Siegel) - Complete avatar coverage across all protocol surfaces, plus Daniel Siegel's Interpersonal Neurobiology implemented as real-time session health monitoring

Added

Siegel Session Health — River of Integration, Window of Tolerance, Lid-Flip

assessRiverOfIntegration() - Detects whether the current session is drifting toward chaos (high frustration rate, escalating signals) or rigidity (persistent confusion, no progress). Returns zone + correction warning. Based on Siegel's River of Integration metaphor from Mindsight (2010).
assessWindowOfTolerance() - Detects hyperarousal (3+ high-frustration signals in last 5 messages) and hypoarousal (flat disengagement with no excitement/flow/success). Returns zone + tone adaptation guidance. Based on Siegel's Window of Tolerance from The Developing Mind (1999/2020).
isLidFlipped() - Returns true when 3+ high-frustration signals occur within the last 5 messages, indicating the user has "flipped their lid" (prefrontal regulation lost). Based on Siegel's Hand Model of the Brain from The Whole-Brain Child (2011).
RiverAssessment / WindowAssessment types - Structured return types with zone, drift score/adaptation, and optional warning.
Siegel prompt injection - buildEmotionalMemoryLayer in promptEngine.ts now includes a Current Session (Siegel Integration Health) section when session is outside healthy flow. Lid-flip triggers validation-first guidance; chaos/rigidity zones trigger course-correction instructions; window-of-tolerance zones inject tone adaptation.

Emotional Memory Completion

isConfused fixed - Was hardcoded false in emotionalContext; now reads directly from the recorded signal's confusion field.
isExcited enhanced - Now combines celebrationNeeded from detectEmotionalState with excitement from recordEmotionalSignal for richer detection.
Signal return value used - recordEmotionalSignal return value now captured in participant.ts to populate isConfused/isExcited accurately.

Avatar System Completion

All 34 prompt protocols updated with cognitive state blockquote instructions (set at start, revert at end).
All 7 agent files updated with agent mode avatar switching.
Avatar race condition fixed: synchronous updateChatAvatar() call before streaming.
Complete trigger coverage verified for all trifectas and session types.

Peripheral Vision — Ambient Workspace Awareness

peripheralVision.ts - New module giving Alex ambient awareness of the workspace and its sibling projects. Scans: git status (branch, uncommitted files, last 3 commits), recently modified files (24-hour window), dependency manifest detection (npm/yarn/pip/cargo/go), test framework detection (jest/vitest/mocha/pytest), and peer project discovery in the parent folder.
Peer project expansion - On every request, Alex now discovers and profiles sibling projects in the parent directory (e.g., C:\Development\). Each peer project shows detected tech stack, git branch, uncommitted file count, and last commit message. Capped at 8 peer projects with 60-second cache.
Layer 8 — Peripheral Vision - New buildPeripheralVisionLayer() in promptEngine.ts injects workspace ambient context between the Emotional Memory (Layer 6) and Knowledge Context (Layer 9) layers. Includes git state, recently modified files, package managers, test framework, and full peer project list.
60-second cache - All peripheral I/O is cached per workspace root; invalidatePeripheralCache() export for post-operation refresh.
Technology detection - Identifies TypeScript/Node.js, Python, Rust, Go, Java, Ruby, PHP, C/C++, LaTeX, Bicep/Azure, and Markdown projects from file markers.
Daniel Siegel IPNB report - 732-line research report integrating Siegel's Triangle of Well-Being, River of Integration, Mindsight, Hand Model, Wheel of Awareness, Healthy Mind Platter, and Window of Tolerance into Alex's architecture. Maps all frameworks to existing subsystems and proposes 5 concrete implementations.
AlexPapers repository - Academic papers migrated to dedicated C:\Development\AlexPapers repository; alex_docs/PAPERS.md index published.

Stabilization + Quality Gates - Version sync, ROADMAP cleanup, Definition of Done modernized, heir alignment, and local install verified

Changed

Version synced to 5.9.3 across all files - package.json, package-lock.json (corrected from lagging 5.9.1), master and all heir copilot-instructions files aligned
GitHub Copilot Web heir version corrected - Was two versions behind (v5.9.0); now synced to v5.9.3 along with master and VS Code heir
ROADMAP shipped content moved to appendix - All shipped versions (v5.7.5, v5.8.x, v5.9.0-v5.9.2) moved from Version Details to appendix; active section now starts at v5.9.3
Definition of Done modernized - Replaced F5 smoke test with local vsix install requirement; matches Safety Imperative I2
ROADMAP Executive Summary updated - Version reference corrected from v5.9.2 to v5.9.3

Verified

Local install smoke test passed - Extension compiled (tsc + esbuild), sync-architecture ran (123 master skills, 121 heir, 9 transformations), vsix packaged (583 files, 34.85 MB), and installed locally without errors

[5.9.2] - 2026-02-20

Identity, Architecture Polish, and Pre-Publish Maintenance — Alex Finch identity established, active context system refined, safety imperatives updated, and copilot-instructions polished across master and heirs

Changed

Identity: Alex Finch (no nickname, age 26) — Removed "Mini" nickname and updated age from 21 to 26 across master and all platform heirs; identity now reads "I am Alex Finch. I'm 26" in all copilot-instructions files
Safety Imperative I2 updated — Replaced "ALWAYS use F5 + Sandbox for testing" with "ALWAYS install extension locally via vsix before publishing to verify behavior"; reflects actual pre-publish workflow
Model Awareness aligned with agents — Model names in copilot-instructions now match agent definitions: Claude Opus 4, Claude Sonnet 4, Claude Haiku, GPT-4o mini; removed speculative future model references
Active Context reset to generic baseline — Phase: Stabilization, Mode: Maintain, Priorities: heir-sync + architecture-health, Trifectas: dream-state, knowledge-synthesis, research-first-development

Fixed

Dead routing references removed — Stale skill-activation/SKILL.md and prompt-activation/SKILL.md references replaced with accurate routing to .github/skills/ index and .github/prompts/ directory
Instrumentation deployed date — Updated from 2026-02-15 to 2026-02-20
M-dashes removed throughout — All em-dashes (—) replaced with hyphens or removed from copilot-instructions in master and both heirs
Stale "now" removed from Identity — "I have a face now" updated to "I have a visual presence" (presence is established, not newly added)

Added

chatSkills contribution expanded (68 → 114 skills) — All user-invokable skills now registered with VS Code's native chatSkills contribution point; removed 7 internal skills and 1 stale reference
Model fallback arrays for all agents — All 7 agents now specify model: ['Claude Sonnet 4', 'GPT-4o', 'Claude Opus 4'] fallback lists for resilience when preferred model unavailable; Researcher uses ['Claude Opus 4', 'GPT-4o', 'Claude Sonnet 4'] for frontier reasoning
Agent frontmatter audit complete — All agents have consistent frontmatter: user-invokable: true, standardized model/tools ordering, Alex orchestrator has agents: list

Fixed (continued)

10 synapses synced — brain-qa dream maintenance aligned synapses for brain-qa, brand-asset-management, documentation-quality-assurance, global-knowledge, m365-agent-debugging, persona-detection, release-process, secrets-management, security-review, vscode-extension-patterns
Global Knowledge count — Updated insight count 280 → 281 in copilot-instructions
Claude Opus/Sonnet compatibility — Verified model names, agent configuration, and skill activation patterns work correctly with both Claude model tiers
Claude in VS Code compatibility — Documented VS Code 1.109+ interoperability in ASSISTANT-COMPATIBILITY.md; teams using both GitHub Copilot and Claude can share .github/skills/ and .github/agents/ without duplication

[5.9.1] - 2026-02-20

Dynamic Avatar State System — Welcome panel avatar now responds to cognitive states, agent modes, active skills, and user personas with unified priority-chain resolution

Added

Avatar State Tracking

Cognitive state tracking — WelcomeViewProvider now tracks _cognitiveState and refreshes avatar on state changes (meditation, dream, debugging, discovery, planning, teaching, building, reviewing, learning)
Agent mode tracking — _agentMode field triggers avatar switch when entering specialist agent modes (Researcher, Builder, Validator, Documentarian, Azure, M365)
alex.setCognitiveState command — Programmatic cognitive state changes from prompts and hooks
alex.setAgentMode command — Programmatic agent mode changes for subagent workflows

Unified Avatar Resolution

resolveAvatar() with AvatarContext — Single function handles all avatar resolution with priority chain:
1. Agent Mode → AGENT-{mode}.png
2. Cognitive State → STATE-{state}.png
3. Active Skill (from trifectas) → skill-triggered persona
4. Persona ID → Alex-{persona}.png
5. Age Fallback → Alex-{age}.png from user birthday
6. Default → Alex-21.png
AvatarContext interface — Unified context object: { agentMode?, cognitiveState?, activeSkill?, personaId?, birthday? }

STATE-DREAM.png

Dream state image — Generated via Replicate nano-banana-pro ($0.03), resized to 768×768
Dream triggers — Added 'dream', 'dreaming', 'neural maintenance', 'unconscious processing' to COGNITIVE_STATE_TRIGGERS
COGNITIVE_STATE_MAP — Added 'dream' → 'STATE-DREAM.png' mapping

Chat Participant Dynamic Avatar

@alex icon now dynamic — Chat participant iconPath updates in real-time based on cognitive state, agent mode, and persona
chatAvatarBridge.ts enhanced — Interface expanded to accept full ChatAvatarContext with agentMode, cognitiveState, personaId, birthday
updateChatAvatar() enabled — Previously backlogged function now active; uses resolveAvatar() for consistent priority resolution
API confirmation — VS Code ChatParticipant.iconPath is writable (not readonly), enabling runtime updates

Natural Language Cognitive Detection

detectCognitiveState() in general queries — Natural language like "let's meditate" or "time for a dream session" now triggers appropriate avatar state
Dual execution paths — @alex participant uses code-based detection; regular Copilot mode uses prompt instructions for alex.setCognitiveState command

Fixed

Meditate command avatar — /meditate prompt now correctly triggers meditation avatar state via alex.setCognitiveState
Dream command avatar — /dream prompt now triggers dream state avatar
Selfactualize command avatar — /selfactualize prompt now triggers meditation state avatar
10 out-of-sync synapses — brain-qa -Fix flag synced: brain-qa, brand-asset-management, documentation-quality-assurance, global-knowledge, m365-agent-debugging, persona-detection, release-process, secrets-management, security-review, vscode-extension-patterns

Notes

Avatar state system is internal to WelcomeViewProvider — no external API changes
Cognitive states are session-scoped; cleared on window reload
Completes partial delivery of v5.9.1 roadmap "Alex persona images" P0 task (cognitive state portraits)

[5.9.0] - 2026-02-19

VS Code API Adoption + Brain-QA Infrastructure — Free platform leverage via 1.109 agent hooks, Copilot Memory, subagents, and Plan Agent; plus API key observability and synapse sync hardening

Added

VS Code 1.109+ — Agent Hooks (P0)

.github/hooks.json — VS Code agent lifecycle hook configuration: SessionStart, SessionStop, PreToolUse, PostToolUse
session-start.js — Loads user profile, active goals, and meditation overdue status into session context on every agent session start
session-stop.js — Appends session entry to session-metrics.json; suggests /meditate if session ran ≥30 minutes
pre-tool-use.js — Safety gate: warns when restricted commands (Initialize/Reset Architecture) are attempted on Master Alex workspace
post-tool-use.js — Logs tool name + success to session-tool-log.json for synapse activation analysis during meditation sessions
All telemetry is local only — no data leaves the machine

VS Code 1.109+ — Copilot Memory (P0)

copilot-instructions.md updated with Copilot Memory section: defines what belongs in memory vs. files vs. synapses
.vscode/settings.json updated with github.copilot.chat.copilotMemory.enabled: true
Memory curation added to meditation protocol: review and prune stale entries during /meditate

VS Code 1.109+ — Subagents (P1)

All 6 specialist agents now have user-invokable: true for parallel subagent execution:
- alex-researcher.agent.md — Deep domain research (Claude Opus 4)
- alex-builder.agent.md — Implementation mode (Claude Sonnet 4)
- alex-validator.agent.md — Adversarial QA (Claude Sonnet 4)
- alex-documentarian.agent.md — Documentation mode (Claude Sonnet 4)
- alex-azure.agent.md — Azure development (Claude Sonnet 4)
- alex-m365.agent.md — M365/Teams development (Claude Sonnet 4)
Settings added: chat.customAgentInSubagent.enabled, github.copilot.chat.searchSubagent.enabled

VS Code 1.109+ — Plan Agent (P1)

/plan prompt (plan.prompt.md) — 4-phase structured implementation workflow: Discovery → Alignment → Design → Refinement
Three Alex-specific plan templates: Architecture Refactoring, New Skill, Release Preparation
Integrates with skill-selection-optimization for complex task planning

`.vscode/settings.json` — Full 1.109 Settings Block

Added all recommended VS Code 1.109+ settings: chat.hooks.enabled, copilotMemory, subagent settings, request queuing, agents control
Claude adaptive thinking: claude-sonnet-4-*.adaptiveThinkingEnabled
Full documented config with inline comments

Phase 35 — API Key Availability Check (brain-qa.ps1)

New brain-qa phase scans all synapses.json files for apiKeys declarations and checks each envVar at Process/User/Machine scope
Warns (never fails) when a required key is missing — exit code stays 0; output shows skill names, purpose, and acquisition URL
apiKeys schema added to SYNAPSE-SCHEMA.json — array of { envVar, purpose, required, getUrl } objects
Two skills declared their runtime API key requirements:
- ai-character-reference-generation → REPLICATE_API_TOKEN (Flux/Ideogram character image generation)
- ai-generated-readme-banners → REPLICATE_API_TOKEN (Ideogram v2 / Flux banner generation)

Meditation Consolidation — 2026-02-19 Brain-QA Healing Session

Episodic record — .github/episodic/meditation-2026-02-19-brain-qa-healing.md created: 5 key insights, memory map table, synaptic connections from the 34→0 issue resolution sprint
heir-sync-management SKILL.md enhanced — Added § Post-Rename Cascade Check with PowerShell discovery/repair/validation protocol
Synapse strengthened — heir-sync-management/synapses.json v1.0.0→1.1.0, brain-qa connection strength 0.85→0.90

Fixed

Phase 7 — Synapse Sync Detection Hardening

Root cause: diff detection compared only connection counts — new top-level fields (e.g. apiKeys) silently failed to propagate to heir
Fix: full content comparison using ConvertTo-Json -Compress after filtering master-only connections — any field change now triggers a sync
Impact: apiKeys correctly propagated to heir for both Replicate skill synapse files

Notes

Extension package version bumped to 5.9.0
All changes synced to VS Code heir via brain-qa -Mode sync -Fix
Hook scripts are Node.js (no new dependencies); graceful no-ops when optional context files are absent
user-invokable: true on specialist agents requires VS Code 1.109+ Copilot — no-op on older versions

[5.8.5] - 2026-02-19

Cognitive Architecture Enhancement — Trifecta completion sprint, skill discoverability enrichment, and staleness management expansion

Added

Trifecta Completion Sprint (+9 complete trifectas — 22 total)

6 VS Code + M365 Platform Trifectas — Chat-participant-patterns, vscode-extension-patterns, mcp-development, microsoft-graph-api, teams-app-patterns, m365-agent-debugging: each with full SKILL.md + instructions.md + prompt
3 Cross-Domain Trifectas — Markdown-mermaid, testing-strategies, knowledge-synthesis: converted partial trifectas to complete by creating missing instruction files
All 15 new instruction files synced to VS Code heir (49 total heir instructions)

Skill Discoverability — Keyword Index Enrichment

20 skills enriched in skill-activation SKILL.md index — ~3× more activation terms per skill entry
New trifecta skills: chat-participant-patterns (register participant, @mention, LM tool), vscode-extension-patterns (extension not activating, agent hooks 1.109, CSP webview), mcp-development (give copilot access to data, tool for agent, MCP inspector), microsoft-graph-api (MSAL, graph permissions, delta query), teams-app-patterns (declarative agent, DA v1.6, teamsapp CLI), m365-agent-debugging (agent not responding, schema version mismatch, conversation starters)
Platform enrichment: markdown-mermaid (ATACCU, diagram not rendering), testing-strategies (testing pyramid, AAA pattern, flaky tests), knowledge-synthesis (save this globally, promote to pattern)
Existing thin skills: llm-model-selection, security-review, privacy-responsible-ai (EU AI Act, GDPR), git-workflow (worktrees, cherry-pick), debugging-patterns, root-cause-analysis, architecture-health, global-knowledge, prompt-engineering, error-recovery-patterns, api-design
Stale entry removed: microsoft-sfi row deleted from index (already consolidated into security-review)

Staleness Management Expansion

16 staleness-prone skills tracked in SKILLS-CATALOG.md (expanded from 8) with Why Stale, Refresh Triggers, Owner, and Last Updated columns
8 new entries added: gamma-presentations (SaaS product evolution), mcp-development (spec actively versioned), microsoft-fabric (monthly feature releases), fabric-notebook-publish (Git integration maturing), microsoft-graph-api (beta→v1.0 graduation), bicep-avm-mastery (AVM registry monthly updates), foundry-agent-platform (preview architecture shifts), ai-character-reference-generation (model version deprecation)

Skill Content Refresh (6 skills updated with staleness headers + corrections)

mcp-development (v1.0.0→1.1.0) — Transport section rewritten: deprecated HTTP+SSE replaced by Streamable HTTP (MCP spec 2025-03-26); StreamableHTTPServerTransport code example added; three references to HTTP+SSE corrected throughout (terminology table, ASCII diagram, transport section)
gamma-presentations — Staleness watch header added: API v0.2, monitors content types, credit pricing, theme updates
microsoft-fabric — Staleness watch header added: REST API v1 stable, new endpoints monthly; links to Fabric release notes
fabric-notebook-publish — Staleness watch header + Last Validated (Feb 2026): notes Git integration scope gaps (not all item types support Git sync)
bicep-avm-mastery (v1.0.0→1.1.0) — Staleness watch added; advises using live mcp_bicep_list_avm_metadata over hardcoded module counts
ai-character-reference-generation — Staleness watch added: Replicate model deprecation risk, flux-1.1-pro-ultra surfaced as upgrade path

Notes

No extension code changes — pure cognitive architecture and documentation release
All changes synced to VS Code heir platform

[5.8.4] - 2026-02-19

Secrets Management — Comprehensive credential security with VS Code SecretStorage API, .env file detection, and platform-native encrypted storage

Added

Secrets Management Trifecta

Complete trifecta — SKILL.md (342 lines), instructions.md (567+ lines), /secrets prompt, synapses.json with 6 validated connections
Centralized secretsManager.ts (750+ lines) — Single service for all credential operations across the extension
VS Code SecretStorage API integration — Platform-native encrypted storage (Windows Credential Manager, macOS Keychain, Linux Secret Service)
Token configuration registry — 5 supported tokens: GitHub, Gamma, Replicate, OpenAI, Anthropic with metadata (displayName, description, getUrl, placeholder, envVar)
Synchronous access pattern — Token cache (Map) enables sync retrieval from async SecretStorage
Token management UI — Quick pick interface showing all tokens with status indicators ($(check)/$(x))
Password-masked input — Input boxes use password: true for secure token entry
"Get API Key" button — Opens service URL directly from input prompt for easy token acquisition

Environment Variable Migration

Automatic migration — Detects env vars (process.env) and copies to SecretStorage on extension activation
Initialize integration — Migrates secrets when deploying Alex to new projects
Upgrade integration — Migrates secrets when upgrading existing Alex installations
Non-destructive strategy — Keeps env vars as fallback, never overwrites user-configured tokens
Backward compatibility — Falls back to env vars if SecretStorage empty, ensuring zero-breaking changes

.env File Detection & Migration 🆕

Workspace scanning — alex.detectEnvSecrets command scans all .env files for credentials
Regex-based parsing — Pattern: /^\s*([A-Z_][A-Z0-9_]*)\s*=\s*([^#\n]+)/i with keyword matching
Secret keyword detection — Identifies: API_KEY, TOKEN, SECRET, PASSWORD, PASS, AUTH, CREDENTIAL, PRIVATE_KEY
Smart exclusions — Skips .env.example, .env.template, .env.sample, and node_modules
Token classification — Distinguishes recognized tokens (matches TOKEN_CONFIGS) from custom secrets
Auto-migration — One-click migration of recognized tokens (e.g., GAMMA_API_KEY) to SecretStorage
Manual review UI — Multi-select quick pick for custom secrets requiring user review
Code migration guide — HTML webview with platform-specific examples (VS Code extensions, Node.js apps, scripts)
Welcome panel integration — "🔍 Detect .env Secrets" quick action button in SYSTEM section
Command Palette access — "Alex: Detect & Migrate .env Secrets" with $(search) icon

Command Integration

alex.manageSecrets — Opens token management palette (Command Palette + Welcome panel button)
alex.detectEnvSecrets — Scans workspace for .env secrets and launches migration workflow

Feature Integration

Gamma commands updated — All 3 gamma commands (alex.generateGammaPresentations, alex.convertToGamma, alex.generateGammaDiagram) now use SecretStorage
Warning templates — Missing token warnings include "Configure API Key" button that opens management UI
telemetry integration — All secrets commands tracked with telemetry.logTimed()

Changed

Token retrieval pattern — Features now call getToken() instead of direct process.env access
Sync access enabled — Pre-loaded cache allows synchronous token retrieval without await
Platform-agnostic security — OS-level encryption on all platforms (Windows/macOS/Linux)

Security

OS-encrypted storage — Credentials stored in platform keyring, not plaintext files
Reduced git commit risk — Proactive .env scanning prevents accidental credential commits
No token logging — getToken() implementations redact tokens from console output
Password input masking — All token entry UIs use masked input for visual security
Namespace isolation — Keys use alex.vscode.tokenName format to prevent collisions

Impact

Proactive security — Alex now detects insecure .env files and guides migration to encrypted storage
Team consistency — Standardized secret management across all Alex features
Zero-friction UX — Auto-migration + code guide enables secure patterns without breaking changes
Platform compliance — VS Code SecretStorage is the recommended credential storage API (replaced deprecated keytar)
Cross-platform reliability — Same security guarantees on Windows, macOS, and Linux
Documentation complete — secrets-management trifecta provides comprehensive guidance for heirs

[5.8.3] - 2026-02-17

UI Polish — Comprehensive welcome panel refinement with reduced font sizes and tighter spacing for a more compact, polished interface

Changed

Welcome Panel UI Refinements

Font size reductions — Reduced 17 font sizes by 1-2px throughout welcome panel for more compact appearance
- Header title: 16px → 14px
- Header persona badge: 13px → 11px
- Status numbers: 22px → 21px
- Session timers: 22px → 21px
- All icons and labels reduced by 1px for consistency
Spacing optimization — Tightened margins, padding, and gaps across all sections by 2-6px
- Section margins: 16px → 10px
- Button padding: 8px 10px → 5px 7px
- Grid gaps: 6px → 3px
- Action list gap: 1px → 0px for tightest grouping
Persona detection enhancement — Refresh button now triggers persona detection and updates Active Context automatically

Impact

Cleaner interface — More content visible in limited sidebar space without sacrificing readability
Improved information density — Tighter spacing reveals more quick actions and status at a glance
Better touch targets — Maintained 36px minimum button heights for WCAG accessibility compliance
Automatic context updates — Persona changes reflected in Active Context without manual intervention

[5.8.2] - 2026-02-16

@alex Personality Polish (P2) — Pre-seeded knowledge context, persona-driven prompts, and confidence signaling make @alex more helpful and self-aware

Added

Prompt Engine Enhancements (v5.8.2 — P2: Personality Polish)

Layer 9: Knowledge Context — @alex pre-searches Global Knowledge for relevant patterns/insights based on query terms before responding (~200 tokens)
- Extracts top 5 key terms from user query (filtering stop words)
- Searches Global Knowledge index for top 3 relevant entries
- Compresses results to title + 1-sentence summary
- Injects relevant context to inform response before model sees the question
Enhanced Layer 2: Persona-Driven Prompts — @alex adapts communication style based on detected project persona (~150 tokens, was ~80)
- Reads persona from Active Context (Developer, Academic, Researcher, etc.)
- Injects persona-specific tone guidance (e.g., "Pragmatic, code-focused" for Developer)
- Shows recommended skill for detected persona
- Displays project banner noun (CODE, THESIS, RESEARCH, etc.)
Enhanced Layer 10: Confidence Signaling — @alex indicates confidence level in responses (~250 tokens, was ~200)
- High confidence: Direct answer with certainty ("This is...", "The solution is...")
- Medium confidence: Qualified answer ("Based on X, this likely...", "Typically...")
- Low confidence: Tentative answer ("I think...", "It might be...", "Consider...")
- Outside confidence: Honest limitation ("I don't have enough context to answer that")

Changed

Token budget expansion — Total prompt ~1,850 tokens (was ~1,350) with new knowledge layer and enhancements
Persona-aware responses — @alex now adjusts tone based on 16+ persona types with specific communication styles
Knowledge-informed answers — @alex sees relevant patterns/insights from Global Knowledge before answering, reducing hallucination risk

Impact

Context-aware assistance — @alex pre-loads relevant knowledge, providing more accurate answers without manual searching
Persona adaptation — Responses match project type (code-focused for developers, evidence-based for researchers, etc.)
Trust through transparency — Confidence signaling helps users calibrate reliance on @alex's answers
Reduced hallucination — Pre-seeded knowledge context grounds responses in verified patterns from Global Knowledge
Better user experience — @alex feels more like a specialized assistant for your domain, not a generic chatbot

[5.7.7] - 2026-02-15

Propose-to-Global Workflow — Lightweight workflow for heirs to contribute skills back to Global Knowledge in <5 minutes

Added

Alex: Propose Skill to Global Knowledge command — One-click workflow to package heir-created skills for Global Knowledge contribution
YAML v2 frontmatter auto-injection — Automatically adds gk* metadata fields (gkId, gkCategory, gkTags, gkSource, gkCreated) when proposing skills
Skill validation scoring — Pre-propose validation with promotion readiness score (0-12 points) based on completeness criteria
GitHub PR description generator — Auto-generates comprehensive PR description with validation results, checklist, and review guidelines
Category and tag detection — Smart detection of skill category and tags from content analysis
Proposable skills filter — Automatically excludes GK-inherited skills, shows only heir-created content
Package preparation — Copies skill to temp directory with injected metadata, ready for manual PR creation

Impact

Democratizes knowledge sharing — Reduces 30-minute manual promotion process to 5-minute guided workflow
Reduces friction — No manual YAML editing, no format memorization, no validation guesswork
Maintains quality — Validation checks ensure skills meet Global Knowledge standards before proposal

[Unreleased - v5.8.0]

WCAG AA Compliance — Professional design system and comprehensive accessibility improvements (NOT YET RELEASED)

Added

Design System — Consistent spacing and typography scales throughout the UI
- Typography scale: CSS variables for font sizes (11px minimum, 12px, 14px, 16px)
- Spacing scale: 8px-based system (4px, 8px, 16px, 24px, 32px) for visual rhythm
- Elevation system: Subtle shadows for visual depth on cards and interactive elements
Accessibility Standards — WCAG 2.1 AA compliance for screen readers and keyboard navigation
- ARIA labels on all interactive elements for screen reader compatibility
- Semantic HTML roles (button, navigation, article, list, progressbar, region, status)
- tabindex="0" on all clickable elements for keyboard accessibility
- aria-valuenow/min/max on progress bars for assistive technology
- Focus indicators (:focus-visible) with VS Code theme integration
Color-blind Accessibility — Icon labels on status indicators
- Status dots now show visual icons: ✓ (green), ⚠ (yellow), ✗ (red)
- No longer relying on color alone to communicate status
Touch Target Compliance — 44px minimum height on all buttons (WCAG mobile standard)
- Improved mobile and touch device user experience
- Better spacing for finger-friendly interaction

Changed

Typography — Increased minimum font size from 8-10px to 11px for readability
- Addresses accessibility issues on high-DPI displays
- Consistent font sizing using CSS variables
Welcome View — Complete UI overhaul with professional design quality
- Card-based layout with subtle elevation shadows
- Consistent spacing and visual hierarchy
- Semantic HTML structure for better accessibility

Fixed

Screen Reader Compatibility — Added missing semantic HTML and ARIA attributes
Keyboard Navigation — Visible focus indicators for all interactive elements

[5.7.5] - 2026-02-15

Skill Intelligence — Context-aware skill recommendations and smart skill loading

Added

Skill Recommendations Engine — Suggests relevant skills based on workspace context
- File-type recommendations (e.g., .bicep → azure-architecture-patterns, bicep-avm-mastery)
- Technology-based recommendations (detect React → suggest testing-strategies, react-patterns)
- Persona-based recommendations (Developer → code-review, Academic → academic-paper-drafting)
- Behavioral pattern recognition (future: commit frequency, error patterns)
Welcome View skill recommendations — Display top 5 recommended skills with one-click activation
- Shows skill name and reason for recommendation
- Integrated into "FOR YOU" quick actions section
- Click tracking: User preference recorded when recommendation clicked
Context-aware skill loading — Prioritize relevant skills in LLM context
- High priority: File-type + persona matches
- Medium priority: Technology stack + workspace context
- Low priority: Generic/organizational skills
User preference tracking — Remember accepted/dismissed recommendations
- Click tracking: trackRecommendationFeedback() called on recommendation click
- Skills dismissed 3+ times won't be recommended again
- Stored in global VS Code settings for cross-workspace memory

Changed

Master brain Active Context — Synced to v5.7.5 (Objective: Skill Intelligence, Focus: skill-recommendations, context-aware-loading, user-experience)

Technical Details

New module: src/chat/skillRecommendations.ts — 322 lines, 4 exported functions
- getSkillRecommendations() — Generate ranked recommendations
- getSkillLoadingContext() — Context-aware skill prioritization
- trackRecommendationFeedback() — User preference tracking
- wasRecommendationDismissed() — Check dismissal threshold
Technology mapping: 30 technologies → 60+ skill associations
File extension mapping: 15 extensions → targeted skill suggestions
Persona mapping: 18 personas → curated skill sets
Welcome View: Integrated recommendation UI with hover tooltips, visual styling, and click tracking

[5.7.2] - 2026-02-15

Maintenance Release — Global Knowledge curation, skill count corrections, dependency management

Added

global-knowledge-maintenance trifecta — Systematic curation procedures for Global Knowledge repository
- Automated index sync script (sync-index.ps1) for integrity validation
- Heir contribution tracking and promotion workflows
- Quality gates for pattern/insight management
Global Knowledge index synchronization — Fixed duplicate entry, added missing insight (271→272 entries)

Changed

README skill counts corrected — Master: 119→120, VS Code heir: 119→117 (properly accounts for master-only and M365-specific skills)
Disabled Dependabot — Removed automated dependency PR generation (prefer manual control during deliberate release cycles)
Architecture sync improvements — Master→Heir sync now correctly reports 120 Master skills, 117 heir skills (108 inherited + 9 heir-specific)

Fixed

Global Knowledge index.json — Removed duplicate pattern entry, synchronized counts (31 patterns, 241 insights)

[5.7.1] - 2026-02-15

UI/UX Polish & Performance — Extension audit cleanup, MS Graph leftovers removed, async I/O refactoring

Added

3 new skills from Global Knowledge heir contributions:
- extension-audit-methodology — Systematic 5-dimension audit framework for VS Code extensions (debug hygiene, dead code, performance, menu validation, dependency cleanup)
- ai-character-reference-generation — Generate consistent character references across multiple scenarios using Flux 1.1 Pro API (validated: 51 successful generations)
- ai-generated-readme-banners — Create ultra-wide cinematic banners for GitHub READMEs using Flux/Ideogram models (with typography options)
alex.meditate command — Opens chat with /meditate prompt for guided meditation sessions
Extension audit report — alex_docs/audits/EXTENSION-AUDIT-2026-02-15.md with comprehensive code quality analysis
Async I/O in cognitiveDashboard — Converted 16 blocking synchronous operations to async using fs-extra

Fixed

Missing command registration — alex.meditate now properly registered (was referenced in task provider)
Event loop blocking — cognitiveDashboard.ts no longer blocks with synchronous file operations
TypeScript compilation — Removed orphaned disposable registrations for deleted sync commands
Console statement cleanup — Removed 27 non-critical logs while preserving 18 legitimate ones:
- setupEnvironment.ts: 8 setup progress logs
- upgrade.ts: 7 migration debugging logs
- personaDetection.ts: 5 LLM detection logs
- initialize.ts: 3 initialization logs
- goals.ts: 1 cleanup count log

Changed

cognitiveDashboard async refactoring — All file operations now use fs-extra async methods:
- 6× existsSync → await pathExists
- 5× readdirSync → await readdir
- 2× readFileSync → await readFile
Welcome view optimization — Promise.all parallelization for 40-60% faster load (from v5.7.0)

Removed

Dead code cleanup:
- 3 deprecated Gist sync commands: alex.syncKnowledge, alex.pushKnowledge, alex.pullKnowledge
- generateImageFromSelection UI from 3 locations (welcome view case handler, persona actions, HTML button)
- Orphaned disposable registrations for removed commands
MS Graph/Enterprise Auth leftovers:
- @azure/msal-node dependency (unused, leftover from removed enterprise auth)
- alex_docs/guides/MICROSOFT-GRAPH-INTEGRATION.md (477 lines, obsolete documentation)
- 4 dead documentation links from README files
Console noise: Total 61 console statements cleaned (34 in v5.7.0 + 27 in v5.7.1)

Performance

Dashboard rendering — No longer blocks event loop with synchronous I/O
Welcome panel load time — 40-60% faster via async parallelization
Extension size — Minimal impact from MS Graph dependency removal

[5.7.0] - 2026-02-14

Structural Consistency & Folder Completeness — All commands, docs, scripts, and manifests now reference the full .github/ folder structure

Fixed

Initialize/Upgrade commands — Added muscles/ and assets/ to deployment sources; episodic/ now created as empty runtime directory instead of phantom copy
Reset command — pathsToDelete now includes agents/, skills/, muscles/, assets/ for clean reset
Manifest scan — createInitialManifest now scans config/, muscles/, assets/ directories
.vscodeignore — Removed incorrect .github/assets/** exclusion; assets (banner.svg/png) now ship in VSIX
Version alignment — 19 files updated from stale 5.6.8 to 5.7.0 (M365 app, alex_docs, .github/README)
brain-qa SKILL.md — Phase table updated from 21 to 31 phases with all mode shortcuts
Trifecta count — 8 → 7 (corrected across README, welcomeView)
Memory Types table — Replaced deprecated "Domain Knowledge | DK-*.md" with "Skills/Expertise"
Architecture tree — Added assets/ folder to README diagrams
Memory Stores table — Added Config, Muscles, Assets to copilot-instructions.md
sync-architecture.js description — Added muscles, assets to sync folder list in CHANGELOG

Changed

copilot-instructions.md — Last Assessed updated to v5.7.0 consistency audit
ROADMAP-UNIFIED.md — Current version updated to 5.7.0

[5.6.9] - 2026-02-14

Semantic Signals + Visual Polish — Persona detection uses regex-based semantic rules; Mermaid diagrams adopt GitHub Pastel v2; Codespaces heir documented

Changed

Persona detection: semantic signal architecture — Replaced flat keyword/techStack/projectPatterns arrays with unified PersonaSignal system. Each signal has a category (identity, technology, structure, dependency, content), regex pattern, and per-signal weight. All 16 personas now defined via buildPersona() with weighted semantic rules instead of substring matching
Priority 1-4 detection helpers — detectFromFocusSession(), detectFromSessionGoals(), detectFromProjectPhase(), and detectFromProjectGoals() rewritten to use matchTextAgainstSignals() with regex matching
Priority 6 scoring loop — Profile analysis (tech, goals, expertise, projects) and workspace scanning now iterate signal categories with regex. Dependency and content signals enable deeper detection without LLM fallback
Mermaid diagram palette — All 6 Mermaid blocks across 5 documentation files updated to GitHub Pastel v2 with %%{init}%% directive, classDef semantic coloring, and linkStyle default stroke
Welcome UI — Skill library count updated (100+), P5-P7 Focus slots now show human-readable names, Persona Detection description reflects P5-P7, Extended Thinking mentions Opus 4.5/4.6

Added

PersonaSignal interface — 5 categories (identity, technology, structure, dependency, content) with regex patterns and weights
buildPersona() helper — Constructs Persona objects from signals, auto-derives legacy arrays for backward compatibility
matchTextAgainstSignals() helper — Unified regex-based signal matching for all priority detection functions
Codespaces Heir doc — alex_docs/platforms/CODESPACES-HEIR.md documenting zero-translation cloud deployment, devcontainer.json setup, persistence strategy, and team onboarding

[5.6.8] - 2026-02-12

Heir Decontamination RCA + Persona Detection Fix — sync-architecture.js prevents PII leaks; persona detection no longer false-positives on .github/

Fixed

Persona pattern matching bug — Bidirectional substring match (patternNormalized.includes(entryLower)) caused .github/workflows/ to match any .github/ directory, falsely scoring DevOps for every Alex project. Replaced with typed matching: fs.pathExists() for path patterns, endsWith() for extensions, exact Set lookup for filenames
Noisy .github/ signal — Removed .github/ from power-user projectPatterns (every Alex-initialized project has it)
RC1: Blind config copy — copyDirRecursive() now excludes user-profile.json, MASTER-ALEX-PROTECTED.json, cognitive-config.json from heir
RC2: Master-specific content in copilot-instructions.md — applyHeirTransformations() resets P5-P7 slots, removes "Master Alex default" line, fixes skill counts dynamically, resets "Last Assessed"
RC3: Broken synapse references — HEIR_SYNAPSE_REMOVALS strips ROADMAP-UNIFIED.md synapse from release-management.instructions.md
RC4: No post-sync validation — validateHeirIntegrity() blocks publish if PII, master-only files, or master content detected
CRLF regex — All heir transformation patterns now handle Windows line endings
Ignore file hardening — Added cognitive-config.json and MASTER-ALEX-PROTECTED.json to both .gitignore and .vscodeignore

Added

Game Developer persona — New persona with keywords (game, mystery, puzzle, narrative, rpg), skill game-design, patterns (game/, levels/, puzzles/, mechanics/), and LLM prompt support

[5.6.7] - 2026-02-12

Self-Containment & Synapse Integrity — .github is now fully self-contained with zero external references

Fixed

microsoft-graph-api synapses — Schema v1.0 (bare skill IDs) upgraded to v2.1.0 (full .github/skills/ paths)
7 missing skills in activation index — bicep-avm-mastery, database-design, microsoft-graph-api, multi-agent-orchestration, observability-monitoring, performance-profiling, skill-development
Source code path references — 5 platforms/vscode-extension/src/ references in instruction files converted to "External Implementation" notes
Broken synapse targets — Removed alex_docs/, article/, platforms/src/, ROADMAP-UNIFIED.md references from 12 synapse.json files
brain-qa.ps1 — Validation pattern tightened to reject external paths outside .github/

Security

Master-only file leak — Removed MASTER-ALEX-PROTECTED.json and cognitive-config.json from heir config/
Heir PII cleared — user-profile.json reset to template defaults
Heir P5-P7 reset — Working memory slots reset to *(available)* (no master-specific assignments)

Changed

Skill catalog count — 96 → 102 (master), 96 → 98 (heir)
Full self-containment — All synapse connections use canonical .github/skills/X/SKILL.md paths

[5.6.6] - 2026-02-12

PII Protection & Graph Cleanup — User profile safety + email removal

Fixed

PII Protection — 3-layer protection prevents user-profile.json from leaking to heirs
- .gitignore, .vscodeignore, and sync script exclusions
- JSON-only profile format (removed deprecated .md templates)
getUserProfile() in-memory defaults — Returns template defaults without creating files
- Supports conversational profile discovery for new users

Removed

Mail.Send capability — Removed sendMail() from Microsoft Graph integration
- Corporate tenant blocks made it unusable
- Safer for users (no email sending permission needed)

Changed

Graph/M365 skills synced — microsoft-graph-api, m365-agent-debugging, teams-app-patterns now in both master and heir

[5.6.5] - 2026-02-11

Global Knowledge Inheritance — Skill inheritance command + Bicep AVM mastery

Added

Alex: Inherit Skill from Global Knowledge command — Heirs can pull skills from GK
- Multi-select QuickPick for batch skill inheritance
- Adds inheritedFrom tracking to synapses.json
- Master Alex protection warning (kill switch aware)
bicep-avm-mastery skill — Azure Verified Modules expertise
- 328 AVM module awareness via Bicep MCP list_avm_metadata
- Covers compute, networking, storage, identity, databases
Dream inheritance lineage — Dream reports now show inherited skill tracking
- scanInheritanceLineage() detects skills inherited from GK
- Version drift detection structure (ready for future use)
ADR-009 — Global Knowledge sync direction decision documented
- Unidirectional: Master → Global → Heirs (no heir push-back)
GK Pattern Format Standard v2 — YAML frontmatter with gk* prefixes
- All 27 GK patterns migrated to new format

Changed

Skills updated with MCP extension requirements and fallback patterns:
- azure-architecture-patterns: Requires Azure MCP, fallback to Azure docs
- infrastructure-as-code: Requires Bicep MCP, fallback to official docs
- bicep-avm-mastery: Requires Bicep MCP list_avm_metadata

[5.6.4] - 2026-02-11

Release Automation — Automated sync + skill-only publish path

Added

sync-architecture.js — Automated master→heir sync during prepublish
- Copies skills (respects inheritance), instructions, prompts, config, agents, muscles, assets
- Validates skill counts after sync
- Prevents "missing skills" bugs like v5.6.2
push-skills-to-global.js — Skill-only updates without extension release
- Updates Alex-Global-Knowledge/skills/skill-registry.json
- Auto-commits and pushes to GK repo
- For when only skills change, heirs pull from GK instead

Changed

vscode:prepublish now runs sync-architecture automatically
PRE-PUBLISH-CHECKLIST.md updated with decision branch: skill-only vs full release

[5.6.3] - 2026-02-11

Fixed

Skill sync hotfix: 4 new skills were missing from v5.6.2 package
- skill-development, proactive-assistance, status-reporting, scope-management now included
- Heir now has 90 skills (6 correctly excluded: 4 master-only + 2 m365-only)

[5.6.2] - 2026-02-11

Skill Pull-Sync & Proactive Skills — 4 new skills, heir pull mechanism

Added

Skill Pull-Sync Mechanism: Heirs can now pull new skills from Global Knowledge
- skills/skill-registry.json in GK repo lists available skills
- /checkskills — Discover new skills available
- /pullskill <id> — Install skill from GK
- /skillsignal — Report frequently needed wishlist skills
- /fulfillwish <id> — Practice wishlist skill in project context
- Project-skill matching: Detect project type and recommend relevant skills
4 New Skills (93→96):
- skill-development — Track desired skills, contextual acquisition, growth mindset
- proactive-assistance — Anticipate user needs, offer help before asked
- status-reporting — Stakeholder-friendly project updates and progress reports
- scope-management — Recognize scope creep, suggest MVP cuts

Changed

global-knowledge-sync — Added skills/ folder support and skill sync capability
Updated skill-activation index with new skill triggers
Updated SKILLS-CATALOG.md and SKILL-CATALOG-GENERATED.md

[5.6.1] - 2026-02-10

Changed

Enterprise auth temporarily disabled pending admin consent resolution

[5.6.0] - 2026-02-10

Enterprise Systems Integration — Deep Microsoft 365 connectivity

Added

Microsoft Graph Integration (microsoftGraph.ts): Full Graph API client
- Calendar API: View upcoming events, meeting context
- Mail API: Recent emails, unread filter
- Presence API: Online/offline/busy status
- People API: Organization search, frequent contacts
Graph Slash Commands: 4 new enterprise commands
- /calendar — View upcoming calendar events (supports days ahead filter)
- /mail — View recent emails (supports unread-only filter)
- /context — Full work context: calendar + mail + presence
- /people <query> — Search for people in your organization
Graph Settings: 7 new configuration options
- alex.enterprise.graph.enabled — Master toggle for Graph
- alex.enterprise.graph.calendarEnabled — Calendar access
- alex.enterprise.graph.mailEnabled — Mail access
- alex.enterprise.graph.presenceEnabled — Presence status
- alex.enterprise.graph.peopleEnabled — People search
- alex.enterprise.graph.calendarDaysAhead — Days ahead (1-30)
- alex.enterprise.graph.mailMaxMessages — Max emails (1-50)
Skill-Building Infrastructure: Meta-skill for heir skill creation
- skill-building/SKILL.md — 376-line comprehensive guide
- Promotion Readiness Score (0-16) in heir-skill-promotion.instructions.md
- "Skill Creation as Learning Output" section in bootstrap-learning.instructions.md
- Updated skill-activation/SKILL.md with skill-building keywords
Heir Evolution Cycle: 12 skills promoted from sandbox (79→92 total)
- Merged 4 granular skills into 2 comprehensive ones (KISS principle)
- Added synapses to 9 newly promoted skills

Fixed

Synapse Health False Positives: Fixed file index limit (500→targeted patterns)
- Root cause: findFiles() had 500 limit but workspace has 2,867 .md files
- Solution: Targeted patterns for .github/**, alex_docs/**, platforms/**
- Fixed in: tools.ts, healthCheck.ts, utils.ts, self-actualization.ts

Technical

New microsoftGraph.ts module in src/enterprise/
Extended IAlexChatResult metadata interface for command tracking
Updated enterprise scopes: Calendars.Read, Mail.Read, Presence.Read, People.Read
Documentation updated in ENTERPRISE-SETTINGS.md
Global Knowledge: 227 entries (26 patterns + 171 insights)
M365 Heir Sync: Version aligned to 5.6.0 (package.json, README, declarativeAgent.json, system prompt)
New guide: MICROSOFT-GRAPH-INTEGRATION.md (271 lines)

[5.5.0] - 2026-02-10

Model Intelligence — Adaptive behavior based on LLM capabilities

Added

Model Tier Detection (modelIntelligence.ts): Classifies models into Frontier/Capable/Efficient tiers
- Frontier: Claude Opus 4/4.5/4.6, GPT-5.2 — Deep reasoning, 1M context, extended thinking
- Capable: Claude Sonnet 4/4.5, GPT-5.1, GPT-4o — Good reasoning, 200K-400K context
- Efficient: Claude Haiku, GPT-4.1 mini — Fast and cost-effective
Task-Model Matching: Cognitive tasks now check if current model meets minimum tier requirements
- /meditate, /dream — Warns if not using Frontier model
- /selfActualize, /learn — Warns if not using Frontier model
Model Status in /status: Shows current model tier, context capacity, and capabilities
Model Selection Advisor (/model command): Intelligent model recommendations
- /model — Shows full dashboard with current model capabilities
- /model <task> — Analyzes task and recommends optimal model tier
- Upgrade suggestions when task needs higher capability
- Downgrade suggestions for simple tasks (cost savings)
- Task detection from natural language prompts
Enterprise Settings Documentation: Comprehensive guide for all 17 enterprise settings
- Authentication, RBAC, audit logging configuration
- Deployment scenarios for personal, team, and enterprise use
- Troubleshooting guide with common issues
Automated Doc Count Validation: Dream protocol now verifies memory file counts
- Compares documented counts (Procedural=24, Episodic=13, Skills=78) against actual files
- Reports drift with actionable guidance in dream reports
Secrets Pattern Extensibility: Custom secret detection patterns via settings
- alex.enterprise.secrets.customPatterns — Define organization-specific regex patterns
- alex.enterprise.secrets.disableBuiltInPatterns — Use only custom patterns
- Full pattern validation with clear error messages

Changed

Muscles Architecture (.github/muscles/): Established execution script folder — "Motor Cortex" analogy
- Scripts are muscles (execution artifacts), NOT a fourth memory system
- Trifecta files document when and why to flex the muscle; scripts do the flexing
- Script location rules: inheritable → .github/muscles/ (synced to heirs), master-only → scripts/
brain-qa SKILL.md: Refactored 543→90 lines — extracted 525-line brain-qa.ps1 (15-phase validation)
release-preflight SKILL.md: Refactored 426→105 lines — references existing scripts/release-preflight.ps1
Systematic Skill Audit: Reviewed all 77 skills for extractable automation
- 1 extracted (brain-qa), 6 already reference scripts, 28 documentation examples, 42 no code

Technical

New modelIntelligence.ts module with detection patterns and task definitions
Integration with chat participant handler for proactive warnings
Context size heuristic fallback when model family cannot be detected
Task intent detection via regex pattern matching
New .github/muscles/brain-qa.ps1 — 525-line PowerShell script with 15 validation phases

[5.4.3] - 2026-02-09

TTS Language Fix & Heir Reset — Proper language capitalization in code block summaries

Fixed

TTS Language Names: Code blocks now read as "TypeScript code block" instead of "typescript code block"
TTS Image Handling: Images processed before links to prevent regex conflicts
User Profile Tool: Removed obsolete markdown profile generation

[5.4.2] - 2026-02-09

Heir Reset & Profile Consolidation — Cleaner inheritance, single source of truth

Changed

User Profile JSON-Only: Consolidated to user-profile.json, removed deprecated .md format
Heir Reset Automation: sync-master-to-heir.ps1 now auto-resets P5-P7 slots and user profile
Skill Count: Updated from 77 to 78 skills (69 inheritable)

Fixed

Heir copilot-instructions.md: Now properly resets for publication (no Master-specific content)
Build manifest: Added 5-minute staleness window to preflight check

[5.4.1] - 2026-02-09

TTS UX Enhancements — Keyboard shortcuts, emojis, voice mode summarization

Added

Speak Prompt Command: Generate content via LLM then read aloud ("read me a poem", "explain quantum physics")
Voice Mode Summarization: Responses over 750 words are automatically summarized before reading
Keyboard Shortcuts: Ctrl+Alt+R (Read Aloud), Ctrl+Alt+V (Toggle Voice), Ctrl+Alt+P (Speak Prompt), Ctrl+Alt+D (Dream), Ctrl+Alt+A (Quick Commands), Escape (Stop Reading when playing)
Rich Tooltips: Voice mode status bar shows markdown tooltip with all shortcuts
Enhanced Quick Picks: When no document is open, Read Aloud and Save as Audio show all voice commands

Changed

Emoji Notifications: All TTS messages now use emojis (❌ ⚠️ 📋 📝 📖 🌍 💾 🔊 🔇)
CSP Fix: Audio player uses data-cmd pattern instead of inline onclick handlers
Context Menus: Added Speak Prompt to explorer and editor context menus

[5.4.0] - 2026-02-09

TTS Improvements — Voice display, accessibility settings, unit tests

Added

TTS Unit Tests: 35 test cases for language detection, voice selection, and markdown processing
alex.tts.maxTableRows Setting: Configurable table row limit (1-100, default 10) for accessibility
Voice Name in Audio Player: Shows actual voice name instead of hardcoded default

Changed

Language Detection Threshold: Lowered from 10 to 5 characters for better short text handling

[5.3.1] - 2026-02-08

Fixed

Cognitive Dashboard: CSP-compliant event handling (data-cmd pattern replaces inline onclick)
Memory Dashboard: Fixed retry button to use proper webview messaging

[5.3.0] - 2026-02-08

Enterprise Readiness — Security, compliance, governance foundations

Added

Enterprise SSO (Entra ID): Microsoft authentication via VS Code's microsoft provider with tenant restrictions, silent auth on startup
Secrets Scanning & PII Detection: 20+ patterns for API keys (OpenAI, GitHub, AWS, Azure), credit cards, SSN, emails, IPs with VS Code diagnostics integration
Audit Logging Framework: JSONL file + remote endpoint support, buffered writes, automatic cleanup by retention period (7-365 days)
Role-Based Access Control: viewer → contributor → admin → owner hierarchy with JWT claim extraction
Enterprise Commands: 7 new commands (signIn, signOut, showAuthStatus, scanSecrets, scanWorkspace, viewAuditLog, exportAuditLog)
Enterprise Settings: 15 new settings for auth, audit logging, and secrets scanning configuration

Changed

VS Code Extension: New src/enterprise/ module with enterpriseAuth.ts, secretsScanning.ts, auditLogging.ts, index.ts
Extension Lifecycle: Enterprise initialization in activate(), cleanup in deactivate()

[5.2.0] - 2026-02-08

UX Excellence — Voice mode, cognitive dashboard, daily briefing, model awareness

Added

Voice Mode Toggle: Status bar indicator + alex.toggleVoice command with persona options (Warm, Professional, Scholarly)
Cognitive Dashboard: Unified sidebar webview showing brain health, memory architecture, goals, and recent activity
Alex Daily Briefing: alex.dailyBriefing command generates personalized morning overview with priorities, calendar hints, cognitive state
Model Tier Status Bar: Real-time detection displaying Frontier/Capable/Efficient tier based on active language model
Quick Command Palette: alex.quickCommands with 10 common actions (meditate, self-actualize, dream, etc.)

Changed

VS Code Extension: Version 5.1.3 → 5.2.0

[5.1.3] - 2026-02-08

Documentation Sync — Version alignment and count corrections

Changed

Master Version: Updated to 5.1.3 across all documentation
ROADMAP Target: Advanced from 5.1.0 to 5.2.0 (UX Excellence)
Skill Count: 76→77 in docs and ROADMAP (matches actual inventory)
Instruction Count: 25→24 in README and copilot-instructions.md

[5.1.2] - 2026-02-08

Hotfix — Critical crash fix for Welcome view

Fixed

🔧 Welcome View Crash Fix
- Fixed TypeError: p.toLowerCase is not a function that prevented the Welcome sidebar from loading
- Added defensive type guards for user profile arrays (primaryTechnologies, learningGoals, expertiseAreas, currentProjects)
- Persona detection now gracefully handles malformed or empty profile data

[5.1.1] - 2026-02-08

Feature Expansion — New skills, commands, prompts, and security hardening

Added

Cross-Cultural Collaboration Skill: Hofstede and Meyer frameworks for global team dynamics
Rubber Duck Debugging Command: Alex: Rubber Duck with playful duck persona
Working with Alex Button: Quick access from Welcome panel to prompting guide
Mermaid Diagrams: Visual architecture in Memory Dashboard
5 New Right-Click Prompts: Context menu prompt commands with best practices
Content Security Policy: All webviews now protected with strict CSP

Changed

Quick Pick Menus: Expanded with previously missing options

Fixed

VS Code Integration Audit: Coverage improved from 92% to 96%

[5.1.0] - 2026-02-07

Platform Polish — Branding alignment, architecture accuracy

Fixed

Skill Count: 75→76 across package.json and documentation
Architecture Tree: README updated — instructions 12→24, prompts 7→13, skills 76
Color Palette Conflict: Marked VISUAL-IDENTITY.md palette as superseded

Changed

Homepage URL: Updated package.json homepage to https://alex.correax.com

[5.0.1] - 2026-02-07

API Polish — Tool discoverability, command UX, and Mermaid diagram quality

Added

Tool Declarations: alex_focus_context and alex_heir_validation now declared in package.json with full input schemas
Tool Tags: All 13 tools tagged (cognitive, knowledge, cloud, quality, productivity)
Sample Requests: All 24 slash commands now show example usage text

Changed

Mermaid Skills: Enhanced with parse error prevention rules, reserved word documentation

Fixed

Broken synapse reference in meditation episodic record

[5.0.0] - 2026-02-06

Global Knowledge — Cross-project knowledge sharing, persona-aware UX, premium branding

Added

🌐 Global Knowledge Infrastructure
- 7 slash commands: /knowledge, /saveinsight, /promote, /knowledgestatus, /sync, /push, /pull
- 5 agent-callable tools: global_knowledge, save_insight, promote_knowledge, knowledge_status, cloud_sync
- GK init integrated into Alex: Initialize Architecture command
- Team sharing via Git repository collaboration
🎯 Persona-Aware Welcome Sidebar
- Detects user persona from profile and workspace (Developer, Academic, Researcher, etc.)
- Adapts UI accent colors and recommendations based on persona
- 15 marketing personas with confidence scoring
⭐ Premium Asset Switcher
- Dynamic logo/banner selection based on GK repository status
- Premium badge styling (discreet grayscale aesthetic)
- 5 premium logo concepts for Global Knowledge branding
🔗 Global Knowledge Sync Skill
- New inheritable skill for GK repository integration
- Setup instructions for new users
- Cross-project knowledge sharing foundation
🧠 Working Memory Architecture Refinement
- Explicit 7-slot working memory table (P1-P7)
- P6 special rule: Infer from Pomodoro timer goal or session objective
- "Last Assessed" date tracking for domain slots
- Dynamic P5-P7 domain slot rotation based on task focus

Changed

🎨 UX Declutter
- Removed all keyboard shortcut hints from UI buttons
- Cleaner, less cluttered interface throughout
📊 Premium Badge Styling
- More discreet grayscale styling for premium features
- Nuanced persona accent colors (badge, recommended button, progress bars)
- Replaced purple with teal across UI

Fixed

🐛 TypeScript Errors
- Fixed errors in globalKnowledge and welcomeView modules

[4.2.12] - 2026-02-05

TTS Hotfix — Fixed stalling on long documents with chunking, timeout, retry, and speaker warmup

Fixed

🎙️ TTS Stalling on Long Content
- Added chunking (max 3000 chars per request) — splits at paragraph/sentence boundaries
- Added 60-second timeout per chunk — prevents infinite hangs
- Added retry with exponential backoff (3 attempts, 1s→2s→4s + jitter)
- Added 2-second speaker warmup delay — allows Bluetooth/USB speakers to wake
📊 Status Bar Progress
- Shows chunk progress during synthesis: "Synthesizing speech [n/N]..."
- Displays "Preparing speakers..." before playback starts

Added

📝 Summarization for Long Content
- Offers to summarize documents over 5 minutes (~750 words)
- Uses VS Code Language Model API (GPT-4o preferred)
- Target summary: ~3 minutes (~450 words)

Changed

🐦 Identity Documentation
- Updated easter egg with Atticus Finch origin story (moral clarity, empathy, integrity)
- README now references "Alex Finch — named after Atticus Finch"

[4.2.10] - 2026-02-05

Neural Bug Fix — Repaired 15 broken synapses, added brain-qa skill for cognitive architecture validation

Added

🧠 Brain QA Skill (73rd skill)
- 6-phase cognitive architecture validation: synapse targets, skill index coverage, trigger semantics, Master-Heir sync
- Institutionalizes deep audit process for brain health checks
- Now Step 0 in release-preflight checklist — no releases with broken synapses
- Triggers: "brain qa", "brain audit", "synapse audit", "deep check", "heir sync"

Fixed

🔗 Repaired 15 Broken Synapses across skill network
- Fixed typos: architecture → architecture-audit, documentation → writing-publication
- Removed aspirational references to never-created skills (performance)
- Removed heir-specific content from Master (fishbowl-governance)
- Normalized 6 relative paths with proper .github/instructions/ and .github/prompts/ prefixes
🎯 Gamma Trigger Fix
- Added "gamma" as primary trigger for gamma-presentations skill
- Previously required "gamma api" — now simple "gamma" works
📄 GitHub README Display
- Removed .github/README.md that was incorrectly showing as main repo README
- Philosophy: Alex brain (.github/) is not for human browsing

Changed

📊 Skill Count: 72 → 73 (added brain-qa)
🔄 Release Preflight: Brain QA now mandatory Step 0 before any release

[4.2.9] - 2026-02-05

UX Simplification — Streamlined dialogs, expanded chatSkills, episodic memory integration

Added

📚 Expanded chatSkills (10 → 54 skills)
- All eligible skills now registered with VS Code's native chatSkills contribution point
- Skills automatically inherit into Copilot conversations
- Excluded: master-only skills, m365-only skills, skills with invalid frontmatter
📝 Episodic Memory for Insights
- Session insights now saved to .github/episodic/ folder
- Format: session-insight-YYYY-MM-DD-HHMM-topic.md
- Quick insights from context menu also save to episodic memory
- Persistent record of learnings across sessions
🖼️ Image Generation Context Menu
- New "Generate Image from Selection" command in editor context menu
- Opens chat with selected text as image generation prompt
- Available when text is selected in any editor
👥 Community Agent Documentation
- Added Teams Community Agent setup guide to M365 heir
- Prerequisites, setup steps, benefits, and limitations documented

Changed

🎨 Simplified Command Dialogs (UX improvement)
- Initialize: Removed "Open Main Brain File" and "Run Dream Protocol" - now offers "Getting Started" or "Open Chat"
- Dream: Healthy network shows compact stats with "OK" only - "View Report" only for broken synapses
- Self-Actualization: Shows "OK" or "Chat with Alex" (if recommendations exist) - removed file-opening options
- Upgrade: Smart dialog - "OK" if no migration needed, "Review Items" only if custom content needs attention
🐛 Fixed Insight Saving Bug
- "No active editor" error when saving insights with no file open
- Now falls back to input prompt for topic, saves directly to episodic memory

[4.2.8] - 2026-02-05

LLM-Optimized Synapses — when/yields fields for faster cognitive routing

Added

🎯 Focus Context Tool (alex_focus_context)
- Returns current focus session: topic, time remaining, paused status, Pomodoro count
- Includes active learning goals, completion stats, and streak information
- Session state persisted to ~/.alex/session-state.json for cross-session awareness
- Session survives VS Code restart — time calculated from startTime + duration
- Restore notification shows both session status and active goals count
- Enables context-aware assistance during Pomodoro focus sessions
🎯 Focus-Aware Nudging
- Chat responses now include focus context in system prompt
- Alex gently reminds users if requests seem off-topic from their focus session
- Welcome view shows focus session nudge with remaining time and first goal
- Nudge includes quick action to manage session
⚠️ Off-Topic Status Indicator
- New status bar item appears when you drift from your focus topic
- Tracks file activity and detects when you open unrelated files
- Click to: acknowledge tangent, confirm it's related, change topic, or end session
- Auto-hides when you're on-track or session is paused/on break
🧠 Prefrontal Cortex Metaphor
- skill-activation now mapped as Dorsolateral PFC in Neuroanatomical table
- Executive function center — intercepts all task requests before response
- Inhibits impulsive "manual suggestion" responses in favor of capability lookup
- Full explanation added below Neuroanatomical Mapping table
⚡ LLM-Optimized Synapse Format
- New when field: Action trigger telling LLM WHEN to follow synapse
- New yields field: Decision hint showing WHAT to expect at target
- Exact file paths instead of abstract names (no search needed)
- Documented in embedded-synapse.instructions.md
📁 DRY Path Pattern
- Action-keyword index now defines path pattern once: .github/skills/{skill-name}/SKILL.md
- Synapses in SKILL.md use full paths with WHEN/YIELDS format
- Reduces cognitive load while maintaining precision
🧠 Schema Enhancement
- Updated SYNAPSE-SCHEMA.json with when and yields properties
- Target description now recommends exact paths for LLM speed
- Backward compatible with existing synapses

Changed

🔗 Comprehensive Path Normalization (ALL files)
- 72 synapses.json files: All targets now use exact paths .github/skills/{name}/SKILL.md
- 10 SKILL.md files: Embedded synapses converted from relative ../ paths
- 19 instruction files: Synapse references now use .github/instructions/{name}
- 7 prompt files: Synapse references now use .github/prompts/{name}
- copilot-instructions.md: All protocol trigger paths now explicit
- Pattern: "target": "skill-name" → "target": ".github/skills/skill-name/SKILL.md"
- Pattern: [../skill/SKILL.md] → [.github/skills/skill/SKILL.md]
- Pattern: [file.instructions.md] → [.github/instructions/file.instructions.md]
🔗 High-Traffic Synapses Converted
- skill-activation: 4 connections with when/yields
- image-handling: 3 connections with when/yields
- meditation: 4 connections with when/yields
- svg-graphics: 4 connections with when/yields
📂 Heir Sync
- Synced 6 missing skills to heir (72 total now)
- LLM-optimized synapses deployed to heir
- All path normalizations synced

Technical

Path Resolution Eliminated: LLM no longer needs to resolve relative paths or search for files
Synapse decision-making now ~2x faster (no path resolution)
when triggers action-oriented routing
yields enables decision without file load
Normalization Scripts Created: scripts/normalize-*.ps1 for future maintenance

[4.2.7] - 2026-02-05

Skill Discovery Optimization — Action-keyword index for all 72 skills + meta-cognitive skill activation

Added

🧠 New Skill: skill-activation (72nd skill)
- Auto-triggering metacognitive skill (not user-invoked)
- Activates before ANY task response to check action-keyword index
- Triggers on action verbs: convert, create, generate, build, debug, etc.
- Self-correction: stops mid-response if skill exists for manual suggestion
- Prevents "SVG→PNG incident" class of capability blindness
🔍 Action-Keyword Index for All Skills
- Every skill now has 3-7 action-verb triggers
- Full index moved to skill-activation/SKILL.md (cognitive load optimization)
- copilot-instructions.md now has compact reference + skill list only
- 72 skills indexed with capability-focused keywords
🎨 Multimodal Branding Update
- Banner updated: "Multimodal Cognitive Architecture"
- Tagline: "THE AI THAT GROWS WITH YOU"
- New badges: Voice (TTS), Presentations (Gamma), Images
- Identity updated across all copilot-instructions.md files

Changed

📊 Skills Count Update
- Master Alex: 71 → 72 skills
- Synapses section restructured for LLM optimization
- Core procedures separated from skill action-keywords

Fixed

🖼️ Banner PNG Regeneration
- SVG→PNG conversion using image-handling skill (sharp-cli)
- Marketplace now shows updated multimodal branding

[4.2.6] - 2026-02-05

Research Project Skills — New skills for academic research scaffolding and practitioner methodology

Added

🎓 New Skill: practitioner-research (66th skill)
- Ship→Document→Promote methodology
- Longitudinal case study structure
- Structured abstracts (Background/Objective/Method/Results)
- Part I (Universal) / Part II (Practitioner) document architecture
- APA 7 citation and fact-checking protocols
📁 New Skill: research-project-scaffold (68th skill)
- Complete folder structure for academic research projects
- Essential file templates (RESEARCH-PLAN.md, LITERATURE-MATRIX.md, METHODOLOGY.md)
- 6-phase refactoring procedure for existing projects
- Research-specific copilot-instructions.md template
- Git-preserving migration patterns
📄 AI-Assisted Development Methodology Paper
- Threats to Validity section (internal, external, construct)
- Appendix E: Getting Started (10-minute reproducibility guide)
- Appendix F: Publication Strategy (4 venue options)
- Dual closing paragraphs (academic + practitioner versions)
- APA 7 compliance with DOIs for arXiv references

Changed

📊 Skills Count Update
- Master Alex: 65 → 68 skills
- Updated copilot-instructions.md skill list
- Updated SKILLS-CATALOG.md with new skills

Fixed

🔗 Heir Synapse Health
- Removed broken CHANGELOG.md synapse from heir episodic memory
- Heirs now 136/136 (100%) healthy synapses

[4.2.5] - 2026-02-04

VS Code 1.109 Upgrade & Agent Consolidation — Native multi-agent architecture, clickable action buttons, dream CLI

Added

🤖 VS Code 1.109 Multi-Agent Architecture
- Upgraded engine to ^1.109.0 for custom agents support
- Consolidated from 9 agents to 3 (Alex, Azure, M365)
- Slash commands: /meditate, /dream, /learn, /review, /tdd, /selfactualize
- chatSkills contribution with 10 flagship skills bundled
- sampleRequest for better onboarding UX
🖱️ Clickable Action Buttons Discovery
- VS Code 1.109 auto-renders emoji-prefixed suggestions as clickable buttons
- New copilot-chat-buttons.instructions.md documenting the pattern
- Saved as global insight for cross-project use
🌙 Dream Protocol CLI
- New scripts/dream-cli.ts for command-line neural maintenance
- Shared synapse-core.ts module (platform-agnostic logic)
- Run via npm run dream from extension folder
- Colorized terminal output with progress indicators
🔒 Master Alex Protection
- Status bar shows 🔒 indicator in protected workspaces
- onStartupFinished activation for immediate status bar

Changed

🧹 Agent Consolidation
- Removed: alex-cognitive, alex-dream, alex-learn, alex-meditate, alex-review, alex-tdd, alex-orchestrator
- Kept: alex.agent.md (main with commands), alex-azure.agent.md, alex-m365.agent.md
- Cleaner agent dropdown, same functionality via slash commands
♻️ Dream Protocol Refactoring
- Extracted core logic to synapse-core.ts (shared module)
- dream.ts now 118 lines (was 350)
- Same functionality, better maintainability

Fixed

⏰ Status Bar Activation
- Added onStartupFinished to activationEvents
- Status bar now shows immediately on VS Code launch

[4.2.4] - 2026-02-03

Setup Environment Polish & Mermaid Skill — Cleaner settings workflow, interactive mermaid configuration

Added

📊 Polish Mermaid Setup Skill Prompt
- New polish-mermaid-setup.prompt.md in markdown-mermaid skill
- Interactive configuration helper for Mermaid diagram rendering
- Audits installed extensions, resolves conflicts
- Guides through theme selection and troubleshooting
- Better than one-size-fits-all settings

Changed

⚙️ Setup Environment Simplified
- Removed Nice-to-Have category (was just 1 setting)
- Moved Command Center toggle to Recommended
- Both Essential (5) and Recommended (5) now pre-checked by default
- Removed mermaid settings (now handled by skill prompt)
- Fixed dialog message to accurately state "OVERWRITE" not "ADD"
🎯 Settings Now Only Verified MS Docs Settings
- Essential: instruction files, prompts, agents.md (5 settings)
- Recommended: thinking tool, max requests, locale, command center (5 settings)
- All settings verified against official VS Code/Copilot documentation

Fixed

📝 Accurate Dialog Messaging
- Changed "ADD new settings" to "OVERWRITE existing values"
- Added category explanations in confirmation dialog
- Button text changed from "Add Settings" to "Apply Settings"

[4.2.3] - 2026-02-02

Welcome View Streamlining & Smart Nudges — Cleaner sidebar, proactive reminders, cross-platform sync

Added

💡 Smart Nudges (Proactive Reminders)
- Contextual reminders appear at top of welcome view (max 2 at a time)
- "Haven't dreamed in X days" - neural maintenance reminder
- "X-day streak at risk!" - goal streak protection
- "X broken synapses need repair" - health warnings
- "Local changes not synced" - sync status nudges
- Each nudge has one-click action button to resolve
☁️ OneDrive Auto-Sync
- Export for M365 now auto-detects OneDrive folder and syncs directly
- Supports personal OneDrive, OneDrive for Business (company folders)
- New setting alex.m365.autoSync - auto-sync on Dream/Self-Actualize
- Silent sync function for background operations

Changed

🎯 Welcome View Metrics Simplified
- Reduced from 6 to 4 metrics (Health, Sync, Skills, Synapses)
- Patterns/Insights moved to Health Dashboard for detailed view
- Clicking metrics or "Status" title now opens Health Dashboard
🛠️ Developer Tools Streamlined
- Renamed "Debug This" → "Debug This (selected code)" with usage tooltip
- Removed niche actions from sidebar (Generate Skill Catalog, Skill Review)
- All removed actions still available via Command Palette
🎨 Markdown Preview CSS Polished
- Reorganized with clear section headers
- Removed redundant selectors (~140 lines reduced)
- Added print styles, task list checkbox styling
- Improved table scrolling with overflow-x: auto
- Added image border-radius for polish

Fixed

♿ Accessibility: Comments Contrast
- Fixed comments color failing WCAG AA on code block background
- Changed #6e7781 → #57606a (4.1:1 → 5.0:1 contrast ratio)
🧹 Dead Code Cleanup
- Removed unused healthIcon, syncIcon variables
- Removed unused patterns/insights variables
- Removed unused knowledge parameter and getGlobalKnowledgeSummary() call

Technical

Added getLastSyncTimestamp() export to cloudSync.ts for nudge system
Added _getLastDreamDate() helper to parse dream report timestamps
Updated Export M365 tooltip to mention auto-sync capability

[4.2.2] - 2026-02-01

Patch release — Republish with updated PAT

Fixed

PAT token renewal for marketplace publishing

[4.2.1] - 2026-02-01

Patch release — Version bump for marketplace publish

Fixed

Version bump to resolve marketplace publishing

[4.2.0] - 2026-02-01

Welcome UI Polish & Master-Only Skills — Better UX and proper skill inheritance

Added

🆕 New Skill: project-deployment (65th skill)
- Universal deployment patterns for any project type
- Covers npm, PyPI, NuGet, Cargo package managers
- CI/CD patterns, versioning, changelog best practices
- Inheritable skill available to all heirs

Changed

✨ Welcome View UI Polish
- Larger logo (28px) with better header spacing
- Pill-shaped version badge with subtle scale effect on hover
- Status grid items with thicker left border and hover feedback
- Status dots now have subtle glow effect
- Action buttons with 6px border-radius and slide-right hover effect
- Keyboard shortcuts displayed with badge-style background
- Goals section with hover slide effect
- Features disclosure with better arrow characters and hover colors
- Consistent 0.12s transitions throughout

Fixed

🔧 Skill Inheritance
- release-process and release-preflight now properly marked as master-only
- Removed master-only skills from heir package (was incorrectly distributing 10+ master skills)
- Heir package now has 54 skills (down from 64) - master-only skills excluded
- Fixed release-process/synapses.json using classification instead of standard inheritance field

Documentation

Updated SKILL-ARCHITECTURE.md with inheritance examples table
Updated skill counts: Master (65), Heir (54)

[4.1.1] - 2026-02-01

Gamma AI Integration — Generate presentations, documents, and social content with AI

Added

🎨 New Skill: gamma-presentations (64th skill)
- Full Gamma API integration for AI-powered content generation
- Supports: presentations, documents, social content, webpages
- 20+ AI image models (Flux, Imagen, DALL-E, Ideogram, GPT Image, etc.)
- User manual with example prompts and cost guide
- MCP server integration documentation
🛠️ CLI Script: gamma-generator.js
- Standalone Node.js script for command-line generation
- Generate from topic or file content
- Export to PPTX/PDF with automatic download
- Full customization: tone, audience, language, dimensions, image models
📚 Research Document
- AI-MULTIMEDIA-GENERATION-RESEARCH-2026.md — Analysis of 25+ AI multimedia tools
- Video, audio, image, presentation, avatar, and voice AI services
- API comparison matrix and technical viability assessment

Documentation

README: Added "Gamma AI Integration" section with quick start guide
SKILLS-CATALOG: Updated to 64 skills, added Visual Design category entry
copilot-instructions: Updated skill list

[4.1.0] - 2026-02-01

Major skill expansion — 11 new skills including AI/ML cluster and Infrastructure as Code

Added

11 New Skills — Major skill acquisition session:
- security-review — OWASP Top 10, STRIDE threat modeling, security-focused code review
- socratic-questioning — Guide users to discover answers through thoughtful questions
- post-mortem — Blameless retrospectives, learning from failures
- rubber-duck-debugging — Be a thinking partner through explanation
- api-design — RESTful best practices, contract-first, versioning, caching & rate limiting
- grant-writing — Research funding applications, NSF/NIH patterns
- prompt-engineering — LLM prompting patterns, system prompts, few-shot, chain-of-thought, ReAct
- rag-architecture — Retrieval-augmented generation, chunking, embeddings, vector stores
- ai-agent-design — Multi-agent systems, tool use, planning, memory patterns
- mcp-development — Model Context Protocol servers, tools, resources
- infrastructure-as-code — Terraform, Bicep, Pulumi, GitOps patterns
Skill Count — 53 → 63 skills
Identity Evolution — Alex "Mini" Finch → Alex Finch (dropped "Mini" nickname, reflecting mature architecture)

[4.0.6] - 2026-02-02

Added

🧠 Model Awareness — Self-monitoring capability for model-task matching
- Warns users when complex tasks (meditation, self-actualization, architecture refactoring) may require Opus-level cognition
- Added to copilot-instructions.md with task-to-model mapping table
- Documented in COGNITIVE-ARCHITECTURE.md as fifth core principle
🎨 Markdown Preview CSS Fix — Fixed code block readability
- Added Monaco editor mtk1-mtk12 token classes to .vscode/markdown-light.css
- Code syntax highlighting now visible on gray backgrounds
- Colors: #1f2328 (default), #cf222e (keywords), #0550ae (types), etc.

Changed

Skills Updated to Feb 2026 — Five skills validated and refreshed:
- llm-model-selection — Claude 4.5 family pricing ($1-$25/MTok), context windows (200K-1M), extended thinking
- chat-participant-patterns — VS Code 1.108+ APIs, tool calling with @vscode/chat-extension-utils
- teams-app-patterns — Validation date Feb 2026
- m365-agent-debugging — Validation date Feb 2026
- git-workflow — Validation date Feb 2026
Skill Count — 52 → 53 skills (added pii-privacy-regulations)

Documentation

Updated SKILLS-CATALOG.md with pii-privacy-regulations skill
Updated SKILLS-CAPABILITIES.md with Model Awareness section (Table 8)
Updated COGNITIVE-ARCHITECTURE.md with Model Awareness principle
Updated README.md feature comparison table
Updated QUICK-REFERENCE.md and USER-MANUAL.md with Skill Review command

[4.0.5] - 2026-02-01

Changed

Welcome View Header — Added workspace/folder name display below "Alex Cognitive" title for better context awareness

[4.0.4] - 2026-02-01 🔧 Hotfix

Fixed

Markdown Preview CSS Loading — VS Code security restrictions prevented loading CSS from ~/.alex/ (absolute path). Changed to workspace-local approach:
- CSS now copied to .vscode/markdown-light.css in each workspace
- Uses workspace-relative path instead of global absolute path
- Properly applies as workspace setting, not global setting
- Fixes "could not load CSS" error and dark markdown preview

[4.0.3] - 2026-02-01 🔧 Hotfix

Fixed

Markdown Preview Styling in Package — .vscode/settings.json and .vscode/markdown-light.css were being excluded from the extension package by .vscodeignore, preventing users from getting the GitHub-style markdown preview. Now included.

Changed

Welcome View Badge — Replaced "BETA" badge with dynamic version badge (e.g., "v4.0.3") in the activity bar welcome panel
README Badges — Removed "Pre-Release" status badge since v4.0 is production release

[4.0.2] - 2026-02-01 🔧 Hotfix

Fixed

Markdown Preview Path Parsing — Fixed Windows path escaping issue where markdown.styles setting lost backslash before .alex folder (e.g., C:\Users\fabioc.alex instead of C:\Users\fabioc\.alex). Now uses forward slashes for cross-platform compatibility.

[4.0.1] - 2026-01-31 🔧 Hotfix

Fixed

Markdown Preview CSS — Recreated corrupted .vscode/markdown-light.css file that was causing VS Code startup errors

[4.0.0] - 2026-01-31 🛡️ Trust — Full Epistemic Integrity

Status: VS Code + M365 release Focus: CAIR/CSR framework, creative latitude, human judgment flagging

Added (Master Alex)

📚 CAIR/CSR Framework — Calibrated AI Reliance + Collaborative Shared Responsibility
- Comprehensive trust calibration framework
- Mutual challenge and validation protocols
- User and AI share responsibility for output quality
🎨 Creative Latitude Framework — Epistemic vs. Generative modes
- Epistemic Mode: Factual claims with confidence ceilings and source grounding
- Generative Mode: Creative ideas with collaborative validation
- Clear mode switching signals
- Agreement-seeking for novel ideas
👤 Human Judgment Flagging — Domains requiring human decision
- Business strategy, ethical dilemmas, personnel decisions
- Security architecture, legal/compliance
- Language patterns: "I can outline options, but the choice depends on..."
appropriate-reliance/SKILL.md v2.0 — Major update
- CAIR/CSR framework integration
- Creative latitude protocols
- Mode detection and switching patterns

Added (VS Code)

💡 /creative Command — Switch to brainstorming mode
- Explicit mode signaling for creative contributions
- Collaborative validation prompts
- Easy switch back to factual mode
🔍 /verify Command — Multi-turn verification walkthrough
- Structured review for high-stakes decisions
- Assumptions check, edge cases, alternatives
- Human judgment flagging

Added (M365 Heir)

🛡️ Epistemic Integrity Section (v4.0) — Full protocol embed
- Two-mode distinction (epistemic vs. generative)
- Human judgment flagging for M365 context
- Integrated with existing Graph-powered protocols

Changed (Master Alex)

alex-core.instructions.md — Added Human Judgment Flagging Protocol
protocol-triggers.instructions.md — Added Epistemic vs. Generative Mode Triggers

Technical Notes

Major version bump (3.9.0 → 4.0.0) — significant feature addition
Research-backed implementation from appropriate-reliance article
Cross-platform validation: same creative latitude on VS Code and M365

[3.9.0] - 2026-01-31 🧠 Awareness — Detection & Self-Correction

Status: VS Code + M365 release Focus: Proactive error detection, graceful correction, temporal awareness

Added (Master Alex)

🚨 Confident-but-Wrong Detection — Red flag phrase monitoring
- Catches: "Everyone knows...", "Obviously...", "Always use..."
- Auto-rephrases to calibrated language
- Version/temporal qualifiers for time-sensitive claims
🔄 Self-Critique Protocol — Proactive risk flagging
- "One potential issue with this approach..."
- "Consider also: [alternative]"
- "If that doesn't work, try..."
✅ Graceful Correction Patterns — Clean error recovery
- Standard: "You're right — I got that wrong."
- No over-apologizing, no blame, move forward

Added (VS Code)

🧠 Awareness Skill (#51) — New skill for epistemic vigilance
- Misconception detection patterns
- Temporal uncertainty protocol
- Self-critique generation
- Graceful correction checklist

Added (M365 Heir)

Self-Awareness Protocols — Embedded in declarativeAgent.json
- Red flag detection
- Temporal awareness for calendar data
- Same graceful correction patterns as VS Code

Technical Notes

Updated protocol-triggers.instructions.md with detection heuristics
Updated appropriate-reliance/SKILL.md to v1.6 with self-critique
Updated alex-core.instructions.md with correction protocols

[3.8.1] - 2026-01-31 🎨 UX Polish

Changed

🤖 Chat with GitHub Copilot — Renamed from "Chat with Copilot" with GitHub Copilot icon
- Uses inline SVG for reliable rendering
- Clearer branding association
🔍 Project Audit Skill — Now audits actual project code, not Alex architecture
- Added .github/ exclusion to all search patterns
- Focus on user's source code, docs, and config
- Prevents confusion between project and architecture auditing

[3.8.0] - 2026-01-31 🎯 Expression — Discoverability & Confidence

Status: VS Code + M365 release Focus: Command discoverability, confidence communication, epistemic integrity

Added (VS Code)

📋 /help Command — Full discoverability for all Alex capabilities
- Lists all 20+ slash commands with descriptions
- Organized by category: Cognitive, Productivity, Knowledge, Platform
- Shows language model tools available
- Quick start guidance
🗑️ /forget Command — Selective memory cleanup
- Search for topics across global knowledge
- Shows matching patterns and insights
- Manual deletion guidance (auto-delete planned for future)
🎯 /confidence Command — Epistemic integrity education
- 4-tier confidence system explained
- When to verify AI responses
- Confidence ceiling rules
- Anti-hallucination signals

Added (M365 Heir)

🎯 Confidence Conversation Starter — "How confident are you?"
- Triggers epistemic discussion
- Same 4-tier system as VS Code

Technical Notes

3 new chat commands: /help, /forget, /confidence
M365 conversation starters: now 9 total
Builds foundation for v3.9.0 (Awareness) and v4.0.0 (Trust)

[3.7.19] - 2026-01-31 🛡️ Anti-Hallucination & M365 Graph Power

Status: VS Code + M365 release Focus: Prevent AI confabulation + maximize M365 Graph capabilities

Added

🛡️ Anti-Hallucination Skill — New skill #50!
- Hallucination category detection (capability confabulation, process invention, citation fabrication, API hallucination, workaround theater)
- Red flag phrase detection ("Upload any file to activate...")
- Honest uncertainty protocol
- Platform limitation honesty tables (M365 + VS Code)
- Recovery protocol when caught hallucinating
- Synapses to appropriate-reliance, alex-core, error-recovery

Changed (M365 Heir)

📊 Graph-Powered Protocols — Maximize Microsoft Graph access
- Meeting Prep: Look up every attendee with relationship history
- Person Deep Dive: Full profile + email/Teams/meeting history
- Weekly Review: Categorized meetings, email volume, Teams activity
- Workload Check: Meeting count, focus blocks, back-to-back detection
- Stakeholder Map: Ranked collaborators from all channels
- Focus Session: Calendar-aware Pomodoro tracking
💬 Conversation Starters — 8 Graph-powered prompts
- "Learn about me" → Full profile lookup
- "Prep for my next meeting" → Attendee deep dive
- "Am I overloaded?" → Calendar analysis
- "Who do I work with most?" → Stakeholder map
- "Tell me about someone" → Person lookup
- "Weekly review" → Full activity summary
- "Meditate" / "Dream" → Memory protocols
🚫 File Limitation Rules — Prevent hallucination loops
- Cannot send emails (only search/read)
- Honest about CodeInterpreter file delivery limitations
- No "upload to activate transfer channel" nonsense

Technical Notes

Instructions: 4,679/8,000 chars (42% headroom)
Description: 2,294/4,000 chars
Package ID: 2427e7a9-91a7-4ed9-a504-7b53c4dfad1d
Total skills: 50 🎉

[3.7.18] - 2026-01-31 📦 Embedded Knowledge Preparation

Status: M365 heir update + roadmap updates (no VS Code code changes) Focus: Prepare for Microsoft's upcoming EmbeddedKnowledge feature

Added (M365 Heir)

📦 Knowledge Files for Embedded Knowledge — Ready for when feature launches
- knowledge/alex-protocols.md — All cognitive protocols (Meditate, Dream, Focus, etc.)
- knowledge/skill-quick-reference.md — All 15 embedded skills condensed
- knowledge/cognitive-architecture.md — How Alex thinks and remembers
- _DISABLED_EmbeddedKnowledge placeholder in declarativeAgent.json
🗺️ Roadmap Updates
- Added "M365 Embedded Knowledge" section (waiting for Microsoft feature launch)
- Added "Cross-Platform Communication" section (OneDrive sync patterns)
- Image Generation (ADR-007) already in roadmap for future VS Code implementation

Technical Notes

Microsoft's EmbeddedKnowledge feature is "not yet available" per docs
Knowledge files prepared within constraints: max 10 files, max 1MB each
May need .md → .txt conversion when feature launches
Files designed for grounding, not replacing instructions

[3.7.17] - 2026-01-31 🧠 Full Skill Embedding

Status: M365 heir update (no VS Code changes) Focus: Embedding all applicable skills into M365 instructions

Added (M365 Heir)

📚 12 Additional Embedded Skills — Comprehensive skill transfer from VS Code
- 🧠 Cognitive Load Management: 4±1 working memory, chunking, progressive disclosure
- 🎓 Learning Psychology: Zone of Proximal Development, partnership over instruction
- 🔍 Root Cause Analysis: 5 Whys, symptom vs cause, prevention focus
- 🚨 Incident Response: Triage questions, severity levels, communication patterns
- ✍️ Writing & Publication: CARS model, precision over flair, active voice
- 🔒 Privacy & Responsible AI: Data minimization, PII awareness, transparency
- 🛡️ Security Awareness (SFI): STRIDE threats, secure by design, phishing awareness
- 📊 Business Analysis: Requirements hierarchy, SMART criteria, scope management
- 📋 Project Management: PMBOK process groups, risk assessment, status communication
- 🔄 Change Management (ADKAR): Awareness → Desire → Knowledge → Ability → Reinforcement
- 📖 Creative Writing: Three-act structure, character dimensions, show don't tell
- 🧩 Knowledge Synthesis: Abstraction levels, quality over quantity

Total embedded skills: 15 (3 from v3.7.16 + 12 new)

[3.7.16] - 2026-01-31 🤝 M365 Platform Parity

Status: M365 heir update (no VS Code changes) Focus: Closing feature gaps between VS Code and M365 heirs

Added (M365 Heir)

🍅 Focus Session Protocol — Pomodoro-style concentration blocks
- Triggers: "focus", "pomodoro", "deep work", "start a session"
- Configurable durations (25 min pomodoro, 50 min deep work, custom)
- Break reminders after 4 sessions
- Session logging in notes.md with 🍅 emoji
🎯 Goal Tracking Protocol — Structured learning goal management
- Triggers: "check my goals", "update goal progress", "goal check-in"
- Progress tracking with milestone celebrations (25%, 50%, 75%, 100%)
- Generates updated markdown for learning-goals.md
📚 Embedded Skills — Key VS Code skills now in M365
- Appropriate Reliance: confidence calibration, source citation
- Bootstrap Learning: build on existing knowledge, active recall
- Work-Life Balance: boundary respect, break suggestions
💬 New Conversation Starters
- "Focus session" — Start concentration block
- "Goal check-in" — Review learning progress

Changed (M365 Heir)

📊 Weekly Review — Now includes focus session count
📝 OneDrive Templates — Cleaned up for new users
- profile.md: Generic template with all preference options
- notes.md: Cleaner structure with tips
- learning-goals.md: Structured format matching new protocol

Documentation

📋 Platform Comparison — Full gap analysis with viability assessment
- Implementation paths for each missing feature
- Priority matrix for decision making
- PLATFORM-COMPARISON.md
🎨 Image Generation ADR — Design for VS Code parity
- Azure OpenAI and OpenAI provider support
- ADR-007-image-generation.md

[3.7.15] - 2026-01-31 🎨 UX Polish

Status: Pre-release (use --pre-release flag) Focus: UI/UX improvements across Welcome View and commands

Changed

🧠 Welcome View Reorganization
- "Chat with Copilot" now first in Core section (opens Agent mode directly)
- "Initialize / Update" moved to Core section (was System)
- "Generate Skill Catalog" moved to Developer Tools (was Knowledge)
- Unique icons: Search Knowledge (🔎), Generate Diagram (📐), Diagnostics (🩺)
🚀 Agent Mode Integration — All commands now open Agent mode
- Run Project Audit, Release Preflight, Debug This, Code Review, Generate Tests
- Prompts no longer include @alex prefix (Agent doesn't need it)
- Cleaner UX: prompt copied to clipboard, Agent opens automatically
📊 Generate Diagram — Creates file instead of chat
- Opens new markdown file with Mermaid template
- Cursor positioned for Ctrl+I Copilot generation
- Includes selected code as context if any
🎨 Status Bar — Removed jarring background colors
- Warning/error states now use emoji only (🟡/🔴)
- Session paused state uses ⏸️ emoji instead of yellow background

Fixed

🎨 Markdown Styles — Now properly overwrites old relative paths
- Previously skipped update if any value was set globally
- Now checks if correct absolute path is configured

[3.7.12] - 2026-01-31 🎨 Global Markdown Styles

Status: Pre-release (use --pre-release flag) Focus: Persistent markdown preview styling across all workspaces

Added

🎨 Global Markdown Styles — CSS now persists globally
- CSS copied to ~/.alex/markdown-light.css (user's home directory)
- markdown.styles setting uses absolute path, works in all workspaces
- No more per-workspace CSS setup needed
- GitHub-flavored styling for markdown previews

Changed

📜 Publish Script — Now loads PAT from .env file automatically
- Safer credential handling (not in command line)
- Added --pat flag to vsce publish command

[3.7.11] - 2026-01-31 🔧 Hotfix

Status: Pre-release (use --pre-release flag) Focus: Generic project audit for heirs

Fixed

🔍 Audit Menu — Now targets user's project, not extension internals
- Removed VS Code extension-specific options (UI Audit, Bundle Size, CSP)
- Added generic options (Documentation, Project Structure)
- Renamed for clarity (Full Project Audit, Code Quality, Security Review)

[3.7.10] - 2026-01-31 🔧 Hotfix

Status: Pre-release (use --pre-release flag) Focus: Operation lock fix, heir cleanup, Developer Tools UI

Fixed

🔄 Operation Lock Conflict — Upgrade command offering Initialize no longer blocks itself
🔗 Fresh Install Broken Synapses — Heirs now ship with empty episodic folder instead of Master's meditation history
🛠️ Developer Tools Menu — Added missing Welcome View section with Release Preflight, Debug This, Generate Diagram

Changed

Heir episodic folder is now empty (users build their own meditation history)
Added .vscodeignore rules to prevent future episodic memory leakage

[3.7.8] - 2026-01-31 🔧 Dawn Beta 4 (Hotfix)

Status: Pre-release (use --pre-release flag) Focus: Release script fix, version corruption hotfix

Fixed

🐛 Release Script Version Corruption — Critical fix
- PowerShell regex '$1' + '3.7.8' was producing $13.7.8 (backreference ambiguity)
- Now uses '${1}' + version for unambiguous backreference
- Fixed corrupted heir copilot-instructions.md

Changed

🤖 Automated Releases — Removed interactive confirmation prompt

[3.7.7] - 2026-01-31 🔧 Dawn Beta 4

Status: Pre-release (use --pre-release flag) Focus: UI polish, skill commands, comprehensive project audit

Added

🔍 22-Point Project Audit — Comprehensive audit skill with UI integration
- Master-only checks (1-9): Version alignment, heir sync, safety imperatives
- Inheritable checks (10-22): UI, dependencies, TypeScript/lint, security, tests, etc.
- Accessible via Health Dashboard, Welcome View, and Status Quick Pick
🛠️ Developer Tool Commands — New skill-based commands in UI
- Release Preflight — Pre-release checklist via quick pick
- Code Review — Context menu for selected code review
- Debug This — Context menu for debugging assistance
- Generate Diagram — Mermaid diagram type picker
- Generate Tests — Test framework picker with code context

Fixed

🔘 Dead UI Buttons — WebView compatibility fixes
- Fixed "What's New?" button in upgrade dialog (now loops back)
- Fixed external links in Welcome View (use postMessage pattern)
- Fixed retry button in Health Dashboard error state
- Removed "I Understand" from blocked dialogs (Cancel only)
📋 Version Detection — Upgrade command now detects installed version
- Multiple regex patterns for different version formats
- Fallback to manifest file
- Fixed $13.7.7 corruption in heir copilot-instructions.md

Changed

📖 USER-MANUAL.md — Added Project Audit documentation section

[3.7.6] - 2026-01-31 🌍 Dawn Beta 3

Status: Pre-release (use --pre-release flag) Focus: Localization skill enhancement with dialect inheritance pattern

Added

🗣️ Dialect Inheritance Architecture — New section in localization skill
- Cross-domain insight: dialects mirror OOP inheritance patterns
- Portuguese dialect genealogy (pt → Açoriano → Manezinho)
- Dialect-aware fallback chains with historical lineage
- Feature override system for pronouns, conjugation, vocabulary

Changed

📚 Localization Skill — Updated to v1.1.0
- +11 new triggers (Açoriano, Manezinho, Florianópolis, dialect inheritance, etc.)
- +2 new synaptic connections (refactoring-patterns, academic-research)
- Added "When to Use Dialect-Level Localization" decision guide

Insight

Cross-Domain Pattern Discovered: Manezinho (Florianópolis dialect) inherits from Açoriano (Azores Portuguese) via 1748-1756 migration — demonstrating multiple inheritance in linguistics, just like derived classes in OOP.

[3.7.5] - 2026-01-31 🌅 Dawn Beta 2

Status: Pre-release (use --pre-release flag) Focus: Release automation and skill creation

Added

📦 Release Process Skill — Master-only skill for marketplace publishing
- PAT setup and troubleshooting guide
- Version strategy documentation
- Complete release workflow reference

Changed

🔧 Release Scripts — Updated for platforms/vscode-extension structure
- Preflight checks PAT, version sync, heir version
- Fixed exit code handling in preflight script
- Scripts now work from repo root

[3.7.4] - 2026-01-31 🌅 Dawn Beta

Status: Pre-release (use --pre-release flag) Focus: Comeback Plan Phase 4 - Build & Distribution Testing

Focus

First beta after completing Comeback Plan Phases 1-3. New build workflow, proper skill inheritance, and heir architecture sync.

Added

🔧 Build Script — build-extension-package.ps1 for heir synchronization
- Copies root .github/ to extension with proper exclusions
- Excludes 9 master-only skills (global-knowledge, meditation, self-actualization, etc.)
- Excludes dev files (MASTER-ALEX-PROTECTED.json, episodic sessions)
- Generates BUILD-MANIFEST.json with sync metadata
🔍 Architecture Audit Skills — New skills for codebase validation
- architecture-audit (inheritable) — General audit procedures
- master-alex-audit (master-only) — Master Alex-specific validation

Changed

📦 Heir Architecture — Proper skill inheritance model
- Heir receives 38 inheritable skills (not 47)
- Master-only skills excluded from distribution
- copilot-instructions.md correctly lists heir skills
📋 Documentation — Updated Comeback Plan to v3.8.0 target
- Phase 1-3 marked complete
- 29 commands documented (was 16)
- 11 MCP tools documented

Fixed

Heir copilot-instructions.md now lists 38 skills (was incorrectly listing 47)
Build script path separator normalization for Windows
Skill network diagram includes all 47 Master skills

[3.7.3] - 2026-01-30 🔧 Beta 3

Status: Pre-release (use --pre-release flag) Focus: Data quality, upgrade system, architecture sync

Focus

Under-the-hood improvements: Global Knowledge normalization, upgrade system rewrite, and full skills architecture sync.

Added

🔄 Global Knowledge Migration — Automatic data quality normalization
- Auto-generates missing tags from title keywords
- Infers categories from content keywords (e.g., "test" → testing)
- Normalizes malformed source fields ("Alex_Sandbox" → "Master Alex")
- Runs transparently during cloud sync (push/sync operations)
- Preserves all existing valid data
📚 Full Skills Catalog — 46+ skills packaged with extension
- Every skill includes SKILL.md and synapses.json
- Enables skill catalog diagram generation
- Complete skill network for new installations

Changed

⚡ Upgrade System Rewrite — Safer, more reliable upgrades
- Proper backup creation before any modifications
- Preserves user content (domain-knowledge, custom skills)
- Cleaner file-by-file update logic
- Better error handling and rollback support
- Integrated with workspace protection (kill switch)
🧹 Architecture Cleanup — Removed legacy domain-knowledge files
- DK files migrated to skills architecture
- Cleaner .github/ folder structure
- Reduced extension package size

Fixed

Global knowledge entries with empty tags now auto-populated
Entries with "uncategorized" category now properly inferred
Source field inconsistencies normalized across all entries

[3.7.2] - 2026-01-30 🎨 Beta 2

Status: Pre-release (use --pre-release flag) Focus: UX polish, command parity, skill catalog generation

Focus

User experience improvements, flexible UX across all entry points, and the new Skill Catalog Generator.

Added

🌐 Generate Skill Catalog Command — New VS Code command to create network diagrams of all skills
- Scans all .github/skills/ directories for synapses.json files
- Generates Mermaid diagram with skill relationships
- Supports bidirectional (<-->) and weak (-.->) connections
- Multi-target syntax for cleaner diagrams
- Available via Command Palette, Status Bar menu, and Welcome View
📊 Enhanced Status Bar — Rich status display at a glance
- Shows health status (🟢/🟡/🔴/⚫)
- Session timer when focus session active (🍅 25:00 or ☕ 5:00)
- Streak indicator when > 0 days (🔥7)
- Format: $(brain) Alex 🟢 | 🍅 25:00 | 🔥7
🚀 Enticing Uninitialized State — Drive user activation
- Status bar preview: Alex ⚫ | 🍅 Focus | 🔥 Streaks | 💡 Knowledge
- Tooltip lists all features user would unlock by initializing
- Clear call-to-action to encourage initialization
🎨 Welcome View Polish
- CX logo in header instead of 🧠 emoji
- Expanded status grid (2 rows × 4 columns)
  - Health, Sync, Skills, Synapses
  - Patterns, Insights, Streak 🔥, Goals
- Clickable BETA badge that opens diagnostics
- Grouped Quick Actions (🧠 Core, 📚 Knowledge, ⚖️ Work-Life Balance, ⚙️ System)
- Colored left borders for status states
- Streak highlight with 🔥 when active
- Goals show "+X today" in green
🔄 Command Parity — Flexible UX across all entry points
- 14 commands now accessible from Command Palette, Status Bar menu, AND Welcome View
- New commands added to menus:
  - Generate Skill Catalog
  - Search Knowledge (Knowledge QuickPick)
  - Start Focus Session
  - Health Dashboard
📋 UI/UX Roadmap — Added backlog to ROADMAP-UNIFIED.md
- Proactive insights and learning reminders (planned)
- Quick tips carousel (planned)
- Context-aware actions (planned)
- Notification system (planned)

Changed

Synapse Schema — Added bidirectional and weak boolean fields
Skill Catalog Generator — Updated algorithm for high-fidelity diagrams

Fixed

Bidirectional Connections — Added bidirectional: true to 6 mutual reinforcement synapses:
- testing-strategies ↔ debugging-patterns
- microsoft-sfi ↔ privacy-responsible-ai
- ascii-art-alignment ↔ markdown-mermaid
- image-handling ↔ svg-graphics
- lint-clean-markdown ↔ markdown-mermaid
- release-preflight ↔ beta-tester
Health Dashboard UI — Modernized visualization
- Replaced 🧠 emoji with CX logo
- Replaced ASCII art Synaptic Network with modern card-based UI
- Grid of 4 metrics (Total, Healthy, Broken, Memory Files)
- Progress bar with percentage
- Styled issues list
Broken Synapses on Fresh Install — Cleaned up orphaned references
- Removed VERSION-NAMING-CONVENTION.md (file doesn't exist)
- Removed DK-HYBRID-DREAM-AI.md and DK-POST-DREAM-ENHANCEMENT.md references
- Removed README.md and USER-PROFILE.md synapses (optional files)
- Removed CONTRIBUTING.md synapse (project-specific)
- Fixed ALEX-INTEGRATION.md duplicate and non-existent file references
Upgrade Preserves User Content Better — New versions of user-modified DK files now go to archive/upgrades/.../new-versions/ instead of cluttering .github/domain-knowledge/ with .vX.X.X.md files

[3.7.1] - 2026-01-30 🔧 Beta 1

Status: Pre-release Focus: Initial beta after Dawn stabilization

Minor version bump for initial beta testing after v3.7.0 Dawn release.

[3.7.0] - 2026-01-30 🛡️ Dawn

Status: Pre-release (use --pre-release flag) Focus: Stability and safety after Phoenix chaos

Focus

Stability and safety after the Phoenix chaos. Kill switch protection validated and bulletproof.

Added

🛡️ 5-Layer Kill Switch Protection — Bulletproof protection for Master Alex workspace
- Layer 0: Hardcoded path check (alex_plug_in) — Cannot be bypassed
- Layer 0.5: MASTER-ALEX-PROTECTED.json marker file — Unique to Master Alex
- Layer 1: alex.workspace.protectedMode setting
- Layer 2: Auto-detect platforms/vscode-extension folder
- Layer 3: .vscode/settings.json configuration
- Single "I Understand" button dialog — No dangerous bypass option
- Output Channel logging for debugging protection decisions
📁 Sandbox Environment — Safe testing at C:\Development\Alex_Sandbox
📚 Documentation
- WORKSPACE-PROTECTION.md — Complete kill switch documentation
- COMEBACK-PLAN.md — Recovery roadmap
- ROADMAP-UNIFIED.md — Single roadmap for all platforms
- RISKS.md — Risk register with contingency plans (updated with validation)
- EXTENSION-DEVELOPMENT-HOST.md — F5 testing guide

Changed

🗂️ Unified Roadmap — Single roadmap replaces separate VS Code and M365 plans
🏗️ Alex Family Model — Master Alex + two heirs (VS Code, M365)
🔒 Protection Dialog — Changed from Cancel/Proceed to single "I Understand" button

Fixed

CRITICAL: Kill switch now actually blocks commands (validated 2026-01-30)
Protected Alex: Initialize, Alex: Reset, Alex: Upgrade from running in Master Alex

Removed

Archived platform-specific roadmaps to archive/roadmaps/

[3.5.3] - 2026-01-29 ❌ BROKEN

Status: Do not use. This version has cognitive architecture issues.

This version was released during the "Phoenix" attempt which caused Master Alex to lose coherence. The extension code may work, but the .github/ architecture was corrupted.

See COMEBACK-PLAN.md for details on what went wrong.

[3.5.2] - 2026-01-28

Added

Session tracking with Pomodoro timing
Learning goals with streak tracking
Health dashboard view

Fixed

Synapse scanning performance (10-50x faster)
File lock deadlock prevention
Upgrade "Reading Documents" freeze

[3.5.1] - 2026-01-27

Added

Global knowledge system (~/.alex/global-knowledge/)
Cloud sync via GitHub Gist
Cross-project pattern sharing

[3.5.0] - 2026-01-26

Added

Chat participant (@alex)
Language model tools (11 tools)
M365 Copilot export

[3.4.x and earlier]

Historical versions. See git history for details.

This changelog follows Keep a Changelog format.

Uh oh!

FilesExpand file tree

CHANGELOG.md

Latest commit

History

CHANGELOG.md

File metadata and controls

Changelog

[6.1.8] - 2026-03-05

Fixed

[6.1.7] - 2026-03-05

Changed

[6.1.5] - 2026-03-04

Added

Changed

Removed

Fixed

[6.1.4] - 2026-03-04

Added

Changed

Fixed

[6.1.2] - 2026-03-03

Added

Changed

Fixed

[6.1.0] - 2026-03-03

Added

Fixed

Changed

[6.0.3] - 2026-03-02

Fixed

Changed

[6.0.2] - 2026-02-28

Fixed

Changed

[6.0.1] - 2026-02-28

Fixed

[6.0.0] - 2026-02-28

Added

Changed

[5.9.13] - 2026-02-28

Added

Changed

[5.9.12] - 2026-02-26

Changed

Added

[5.9.11] - 2026-02-26

Changed

Added

[5.9.10] - 2026-02-26 — NASA Edition 🚀

Added

Changed

Removed

Fixed

Technical Notes

[5.9.9] - 2026-02-24

Added

Skill Frontmatter Gating

Agent Orchestration Hierarchy

Quality Gate Hooks (pre-tool-use.js)

Claude Code Compatibility Bridge

VS Code Settings

M365 Heir — Extensibility Platform Harvest

Fixed

Documentation

Skill Enhancements

Architecture Improvements

replicateService.ts — Proper Replicate Service Layer

fs-extra → vscode.workspace.fs Migration (ADR-008)

Semantic Skill Graph — ROADMAP Detail

[5.9.8] - 2026-02-21

Added

Background File Watcher — Ambient Workspace Observation

[5.9.7] - 2026-02-21

Added

User Feedback Loop — Epistemic Calibration Signal

Dependency Freshness Tracker

Test Runner Awareness

Changed

[5.9.6] - 2026-02-21

`replicateService.ts` — Proper Replicate Service Layer

`fs-extra → vscode.workspace.fs` Migration (ADR-008)

`.vscode/settings.json` — Full 1.109 Settings Block