global-knowledge

type	skill
lifecycle	stable
inheritance	inheritable
name	global-knowledge
description	Cross-project knowledge search, pattern recognition, and insight management via unified AI-Memory
tier	standard
applyTo	**/*knowledge*,**/*insight*,**/*ai-memory*,**/*pattern*
currency	2026-04-20T00:00:00.000Z

Global Knowledge Skill

Your past solved problems, accessible from any AI surface.

Architecture: AI-Memory

Global knowledge lives in a single shared file: AI-Memory/global-knowledge.md

This replaces the previous GitHub-based system (~/.alex/global-knowledge/, GK-/GI- files, index.json). The AI-Memory folder is accessible across all surfaces:

• M365 Copilot: Reads via OneDriveAndSharePoint capability
• VS Code: Reads via local OneDrive sync path. Falls back to ~/.alex/AI-Memory/ when OneDrive is unavailable
• Agent Builder agents: Reads via OneDriveAndSharePoint capability

AI-Memory Files

File	Purpose
profile.md	User identity, preferences, expertise
notes.md	Session notes, reminders, observations
learning-goals.md	Active/completed goals with progress tracking
global-knowledge.md	Cross-project insights, patterns, and learnings

Templates: platforms/m365-shared/onedrive-templates/AI-Memory/

Check Before You Create

Always search global-knowledge.md before creating new skills, instructions, or prompts. This prevents duplicate work and ensures institutional knowledge is reused.

Lookup Process

1. Search AI-Memory/global-knowledge.md for existing content
2. Also search local .github/skills/ (the answer may already be a skill)
3. Also search Copilot Memory (/memories/) for VS Code-specific notes
4. Evaluate the result:

Result	Action
Exact match	Adopt directly; don't recreate
Similar match	Adapt as starting point
No match	Create new, then add to global-knowledge.md

This check is mandatory before skill creation (Phase 1 of skill-building procedure).

Knowledge Organization in global-knowledge.md

Insights are organized by category with a consistent format:

### [Topic Title]

- **Source**: [Project or context where learned]
- **Insight**: [The key learning]
- **Date**: [When captured]

Categories emerge organically: PowerShell Patterns, Azure Patterns, Frontend Patterns, AI Image Generation, Design Patterns, Workflow Patterns, etc.

What Goes in global-knowledge.md

Content	Example
Cross-project patterns	"SWA EasyAuth handles login; domain gating via fetch"
Hard-won gotchas	"PowerShell -replace corrupts multi-byte content"
Architecture decisions	"Cosmos liveness probes cause container kill loops"
Integration patterns	"MS Graph MCP: call suggest_queries first"
Tool-specific learnings	"Flux 1.1 Pro max width 1440px"

What Does NOT Go in global-knowledge.md

Content	Where It Belongs
Personal preferences	AI-Memory/profile.md
Session reminders	AI-Memory/notes.md
Learning goals	AI-Memory/learning-goals.md
Project-specific config	Project's own .github/ or docs
Temporary workarounds	Project notes (will become stale)

Memory System Differentiation

Alex uses three complementary memory systems:

1. AI-Memory (Cloud Storage): Cross-Platform Persistent Memory

Shared across all AI surfaces via cloud storage sync. Use for:

• User identity and structured preferences (user-profile.json)
• Narrative profile and expertise (profile.md)
• Session notes and observations (notes.md)
• Learning goals with progress tracking (learning-goals.md)
• Cross-project insights and patterns (global-knowledge.md)

2. Copilot Memory (/memories/): VS Code Workflow Memory

GitHub-cloud-synced notes for VS Code workflows. Use for:

• VS Code-specific workflow tips (terse, actionable rules)
• Tool-specific workarounds (npm, CLI, Mermaid palette, etc.)
• Cross-workspace tool access patterns (MS Graph MCP, etc.)
• Session-scoped working state (/memories/session/)

3. Project Memory (.github/): Local Architecture

Project-specific cognitive architecture. Use for:

• Skills, instructions, muscles (the trifecta), plus prompts
• Episodic memories from meditation sessions
• Connections between skills (via applyTo patterns)
• Project persona (.github/config/project-persona.json)

Decision Matrix

Data Type	AI-Memory	/memories/	.github/
Name, role, identity	user-profile.json
Formality, detail level, humor	user-profile.json
Learning style, explanation style	user-profile.json
Technologies, expertise areas	user-profile.json
Writing style rules (em dash ban)		/memories/
Tool workarounds (npm, CLI)		/memories/
Mermaid palette, diagram prefs		/memories/
Cross-workspace access (MS Graph)		/memories/
Cross-project insight	global-knowledge.md
Learning goals with progress	learning-goals.md
Session notes	notes.md
Project persona (detected)			project-persona.json
Project architecture			.github/skills/
Meditation output			.github/episodic/

Conflict Prevention Rules

AI-Memory/user-profile.json is the source of truth for user identity and preferences. Copilot Memory (/memories/) supplements with VS Code-specific workflow tips. They do not overlap.

Rule	Rationale
Never store name, role, or identity in /memories/	AI-Memory owns identity
Never store formality, detailLevel, explanationStyle in /memories/	AI-Memory owns structured preferences
Never store technologies or expertise in /memories/	AI-Memory owns the profile
VS Code workflow tips (em dash ban, palette, CLI workarounds) go in /memories/ only	These are terse, actionable, VS Code-specific
If a preference exists in BOTH systems, AI-Memory wins	Single source of truth

Integration Workflow

User learns something
  Is it about who they are?  -> AI-Memory/user-profile.json
  Is it a cross-project pattern? -> AI-Memory/global-knowledge.md
  Is it a VS Code workflow tip?  -> /memories/ (terse, 3-workspace test)
  Is it project-specific?  -> .github/skills/ or .github/episodic/

No duplication: Each piece of information lives in ONE system.

Saving Insights

From VS Code

During a session, when you learn something cross-project:

1. Open %OneDrive%/AI-Memory/global-knowledge.md (synced locally)
2. Add entry under the appropriate category
3. Use the standard format: Topic, Source, Insight, Date
4. Save the file (OneDrive syncs automatically)

Or say "consolidate" to trigger the cognitive protocols consolidation workflow, which prompts you to capture insights.

From M365 Copilot / Agent Builder

1. Say "consolidate" or "save this insight"
2. Alex generates the content update in a code block
3. You save it to OneDrive > AI-Memory > global-knowledge.md

Promotion from Project to Global

When a project-specific learning proves cross-project:

1. Identify the insight in .github/episodic/ or session notes
2. Abstract it (remove project-specific details)
3. Add to AI-Memory/global-knowledge.md under the right category
4. Verify it reads well without project context

Promotion Signals

Strong Promotion Signals

Pattern	Description
Cross-project applicability	Would 3+ other projects benefit?
Resolution pattern	Solved problem with concrete solution
Hard-won gotchas	Prevents repeat mistakes
Architecture with rationale	ADR-style decisions with "why"
Integration patterns	How systems connect

Anti-Promotion Signals

Signal	Why Skip
Project-specific config	Only works in one context
Temporary workarounds	Will be obsolete
Personal preferences	Goes in profile.md
Incomplete/draft content	Needs validation

Migration from Legacy GK System

The previous system used ~/.alex/global-knowledge/ with GK-/GI- files and index.json. All valuable content has been migrated to AI-Memory/global-knowledge.md. The legacy system is deprecated:

Legacy	Replacement
~/.alex/global-knowledge/	OneDrive/AI-Memory/
patterns/GK-*.md	Categories in global-knowledge.md
insights/GI-*.md	Entries in global-knowledge.md
index.json	Not needed (single file, searchable)
Alex-Global-Knowledge GitHub repo	OneDrive sync (automatic)
skill-registry.json	Master Alex .github/skills/ is canonical

Triggers

• "save insight", "promote to global", "consolidate"
• "search knowledge", "have I solved this before?"
• "global knowledge", "cross-project"

$patterns = (Get-ChildItem patterns\GK-*.md).Count
$insights = (Get-ChildItem insights\GI-*.md).Count

Sync During Dream

Automated sync during dream/meditation cycles: check for uncommitted changes, pull latest, regenerate KNOWLEDGE-INDEX.md, report sync status.

Skill Pull-Sync (For Heirs)

Discover and pull new skills from GK via skills/skill-registry.json. Commands: /knowledge checkskills (discover), /knowledge pullskill <id> (pull on demand).

Quality Gates

✅ Knowledge sync reports 0 orphaned, 0 unindexed ✅ Heir contributions documented as insights (if any) ✅ Counts in copilot-instructions.md match disk ✅ git status clean (or changes committed)

---