Global Knowledge Skill

Your past solved problems, searchable from any project.

Knowledge Types

Type	Prefix	Purpose	Lifespan	Example
Pattern	GK-*	Reusable solution, proven in 2+ projects	Long-term	`GK-error-handling-patterns.md`
Insight	GI-*	Timestamped observation, may not generalize	Medium-term	`GI-react-hooks-gotcha-2026-01-24.md`

Key distinction: A pattern is abstracted and proven. An insight is specific and dated. Insights may be promoted to patterns after validation.

Commands

Command	Purpose
`/knowledge <query>`	Search patterns + insights
`/saveinsight`	Capture learning from current session
`/promote`	Promote local knowledge to global pattern
`/knowledgestatus`	Check GK health and stats

Memory System Differentiation (VS Code 1.109+)

Alex uses two complementary memory systems. Use the right one for the right data:

Copilot Memory (GitHub Cloud)

Cloud-synced preferences and personal context. Use for:

Data	Example	Why Cloud
Preferences	"Use 4 spaces, dark mode"	Same across all machines
Coding Style	"Prefer functional components"	Consistent patterns
Learning Goals	"Master K8s by March"	Personal growth tracking
Session Notes	"Finish auth tests tomorrow"	Cross-session reminders

Characteristics:

☁️ Syncs across all machines automatically
👤 Personal to GitHub account
🔒 Encrypted at rest
💬 Accessible via natural language in chat

Global Knowledge (~/.alex/)

Local domain knowledge and project learnings. Use for:

Data	Example	Why Local
Domain Expertise	"How OAuth2 works in our system"	Project-specific, detailed
*Patterns (GK-)**	"Rate limiting implementation"	Searchable, categorized
*Insights (GI-)**	"Fixed N+1 query with eager load"	Timestamped learnings
Session History	Episodic meditation records	Full context preserved

Characteristics:

💾 Local storage, you control the data
🔍 Full-text searchable via MCP tools
📁 Organized in patterns/ and insights/
🔗 Synaptic connections between items

Decision Matrix

Question	Copilot Memory	Global Knowledge
Is this personal preference?	✅
Is this project know-how?		✅
Should it sync to new machines?	✅
Does it need full-text search?		✅
Is it a learning goal?	✅
Is it a pattern/solution?		✅

Integration Workflow

User learns something → Is it personal? → Copilot Memory
                     → Is it shareable? → Global Knowledge (GI-*)
                     → Is it reusable?  → Global Knowledge (GK-*)

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

Promotion Candidate Patterns

When scanning project knowledge for promotion candidates, use these patterns to identify high-value content:

🎯 Strong Promotion Signals

Pattern	Description	Example
Cross-project applicability	Would 3+ other projects benefit?	Error handling strategy that works anywhere
Resolution pattern	Solved problem with concrete solution	"N+1 query fixed with eager loading"
Hard-won gotchas	Prevents repeat mistakes	"Jest timers require runAllTimers() after async"
Architecture with rationale	ADR-style decisions with "why"	"Chose event sourcing because..."
Pipeline/workflow patterns	Automatable processes	CI/CD template, release flow
Integration patterns	How systems connect	OAuth flow, API gateway setup
Migration patterns	Version/format transformations	"Upgrading from v3 to v4 schema"
Schema/format patterns	Data structures that work	Synapse JSON schema, config templates

❌ Anti-Promotion Signals

Signal	Why Skip	Example
Project-specific config	Only works in one context	"Our AWS account settings"
Temporary workarounds	Will be obsolete	"Hack until v2.0 releases"
Personal preferences	Not generalizable	"I like 4 spaces"
Incomplete/draft content	Needs validation	Work-in-progress notes
Already in GK	Avoid duplicates	Check `index.json` first
Too specific names/IDs	Not portable	Hardcoded team names, repo URLs

Automated Scoring (Score ≥5 = Candidate)

The extension evaluates skill files with this scoring:

Criteria	Points	Detection
Has synapses	+3	Synapse format in content
Well-structured (3+ H2 sections)	+2	`## Section` headings
Has tags defined	+1	`Tags: tag1, tag2`
Substantial content (>1KB)	+1	File size
Rich content (>5KB)	+2	File size
Has code examples	+2	Code blocks in content
General applicability	+1-3	Contains: pattern, best practice, guideline, framework, principle, architecture

Minimum for auto-promotion: Score ≥ 5

Quick Decision Tree

Is this knowledge?
├── Personal preference → Copilot Memory (not GK)
├── Project-specific config → Keep in project skills
├── Would I search for this in another project?
│   ├── No → Keep in project skills
│   └── Yes → Continue...
│       ├── Is it a reusable solution? → Pattern (GK-*)
│       └── Is it a timestamped learning? → Insight (GI-*)

Real Examples from Curation

These patterns were identified during cross-project knowledge curation:

Promoted as Patterns (GK-*):

GK-kill-switch-pattern-protecting-master-alex.md — Safety mechanism applicable to any critical system
GK-synapse-schema-2026-standard.md — Data format standard for the entire architecture
GK-domain-knowledge-azure-synapse-pipeline-.md — ETL patterns for data engineering

Promoted as Insights (GI-*):

GI-agent-consolidation-pattern-kiss-merge-2026-02-09.md — Lesson from merging duplicate agents
GI-schema-migration-discrimination-flag-vs-change-2026-02-09.md — Distinguishing migration types
GI-post-upgrade-verification-checklist-2026-02-09.md — Checklist developed after failed upgrade

Source Priority for Promotion

When curating from multiple projects:

Source Type	Priority	Rationale
Master Alex episodic memories	Highest	Core cognitive learnings
Production project skill files	High	Battle-tested knowledge
Platform heir skill files	Medium	May have implementation-specific details
Research/experimental projects	Lower	May not be validated
Archive content	Lowest	May be outdated

Promotion Workflow Commands

@alex /promote                  # Interactive promotion from project DK
@alex /saveinsight              # Quick insight capture
@alex /knowledgestatus          # Check current GK health
@alex /knowledge <query>        # Search before adding (avoid duplicates)

Synapses

Synapse: heir-skill-promotion.instructions.md (0.9, procedural, bidirectional) — "How skills move from heirs to Master" Synapse: global-knowledge-curation.instructions.md (0.85, procedural, bidirectional) — "Periodic cleanup and implementation" Synapse: unified-meditation-protocols.prompt.md (0.8, episodic, forward) — "Auto-promotion during meditation"