Heir Sync Management

Safely synchronize cognitive architecture from Master to Heirs without contamination.

Scope: Master-only skill. Covers sync pipelines, PII protection, drift detection, skill promotion, and clean-slate distribution.

Core Concepts

Concept	Definition
Heir	Platform-specific deployment inheriting Master's DNA (VS Code, M365, GitHub Copilot Web)
Deployment Channel	Delivery mechanism for an heir (Marketplace, Teams Package, .github/ commit)
Integration	Cross-heir communication (OneDrive Sync, GitHub Cloud)
Translation Heir	Heir requiring format/schema conversion (e.g., M365 — export pipeline)
Deployment Heir	Heir needing only configuration, no code translation (e.g., GitHub Copilot Web — .github/ files only)
Contamination	Master-specific data leaking into heir packages
Drift	Heir diverging from Master's architecture over time
Promotion	Elevating heir-developed capabilities back to Master

Rule: Never confuse delivery mechanism with inheritance relationship — the "what" (identity/DNA) stays constant, only the "how" (delivery) varies.

Sync Architecture

What Gets Synced

The sync script (sync-architecture.js) copies these folders from Master .github/ to Heir .github/:

Folder	Content
instructions/	Procedural memory
prompts/	Episodic memory
config/	Configuration (with exclusions)
agents/	Agent definitions
muscles/	Execution scripts
skills/	Skills (filtered by inheritance)

What Must NEVER Sync

Item	Why
user-profile.json (real)	Contains personal name, email, preferences
episodic/ memories	Session-specific to Master
Master-only skills	Only useful for managing the Master repo
API keys, PATs, secrets	Environment-specific credentials
Working memory with populated P5-P7	Gives new users pre-filled slots instead of clean defaults

3-Layer PII Protection

Every sync pipeline must implement three independent defense layers:

Layer 1: Exclusion List

Files that are never copied, period:

const EXCLUDED_CONFIG_FILES = [
  'user-profile.json',      // Personal data
  'MASTER-ALEX-PROTECTED.json', // Kill switch marker
  'goals.json',             // Session-specific
];

Layer 2: Source File Sanitization

Scan all files being copied for hardcoded personal data:

Pattern	Action
Real names in source headers	Replace with team/org name
Email addresses in code	Replace with placeholder
Personal names in package.json	Use organization name
Populated P5-P7 working memory slots	Reset to *(available)*

Rule: Personal identity belongs ONLY in user-profile.json. All other files use team/org names.

Layer 3: Pipeline Validation Gate

Post-copy regex scan that blocks packaging on violations:

Check	Regex Example	On Match
Real name in files	/\bFirstName\s+LastName\b/g	EXIT 1
Email addresses	/[\w.-]+@[\w.-]+\.\w+/g	EXIT 1 (except templates)
API keys	/[A-Za-z0-9]{32,}/ in non-code files	WARNING
Populated P5-P7	Check copilot-instructions Memory Stores	EXIT 1

Anti-pattern: Manual checklists. The copy function itself must be architecturally incapable of leaking.

Clean Slate Distribution

Template Generation

Simply excluding personal files leaves heirs without expected file structure. Generate fresh templates:

File	Master Version	Heir Template
user-profile.json	Real user data	Empty with defaults + setup instructions
copilot-instructions.md	Populated P5-P7	P5-P7 set to *(available)*
goals.json	Active session goals	Empty goals array

Post-Sync Reset Sequence

After copying files, apply these transformations:

1. Reset environment-specific values — P5-P7 slots, session state
2. Generate template files — Fresh starters with clear defaults
3. Remove broken synapse references — Master synapse IDs that don't exist in heir
4. Validate file structure — Ensure all expected files exist (even if empty templates)

Drift Detection

Pre-Release Checklist

Run these validations before every release:

Check	Method	Fail Condition
Skill count match	Count Master inheritable vs Heir skills	Mismatch
File hash comparison	SHA256 of synced files	Divergence without override
Inheritance field validation	All skills have inheritance in synapses	Missing field
Orphan reference detection	Grep for files referenced but not present	Broken references
Config drift	Compare heir config against Master template	Unexpected values

Heir Configuration Drift Signals

Signal	Indicates
Heir P5-P7 slots populated	Sync overwrote clean defaults
Heir has master-only skills	Exclusion filter not working
Heir synapse IDs don't resolve	Broken references from Master copy
Heir package.json has personal name	Sanitization missed

Leakage Prevention Patterns

Master-Only Content Categories

Content that must NEVER sync to heirs falls into three categories:

Category	Examples	Filter Method
Skills	heir-curation, master-alex-audit	inheritance: "master-only" in synapses.json
Files	ROADMAP-UNIFIED.md, alex_docs/architecture/	Path-based filtering in sync script
Metadata	Complete trifecta count (9 vs 8), safety imperatives I1-I4	Content transformation rules

Dual Filtering Strategy (2026-02-14 Fix)

Problem: cleanBrokenSynapseReferences() only filtered master-only skills, not master-only files.

Solution: Enhanced function with dual target detection:

// Master-only skills (from inheritance metadata)
skippedMasterOnly.some(removed => target.includes(removed))

// Master-only files (hardcoded registry)
masterOnlyFiles.some(file => target.includes(file))

Master-Only Files Registry:

• ROADMAP-UNIFIED.md — Master development roadmap
• alex_docs/architecture/ — Architecture documentation not deployed to heirs
• MASTER-ALEX-PROTECTED.json — Kill-switch marker

Content Transformation Rules

Some content needs adjustment not removal:

Content	Master Value	Heir Value	Transform Rule
Complete trifectas count	9	8	Remove heir-curation from list
Safety imperatives	I1-I7	I5-I6 only	Strip I1-I4, I7 (master workspace protection)
Active Context Persona	Current session	Developer (85%)	Reset to default
Last Assessed	Date	never	Clear assessment timestamp

Implementation: applyHeirTransformations() in sync-architecture.js

Validation Cascade (4-Stage Quality Gate)

1. Sync: Copy files with skill/config exclusions
2. Transform: Apply heir-specific content adjustments
3. Validate: PII scan, master-only file detection
4. Audit: Count verification, synapse connection integrity

Fail-Fast: Any contamination detection halts the build (exit 1).

Heir → Master Promotion

6-Step Promotion Workflow

Step	Action	Output
1. Discover	Review heir DK/skill files for portable knowledge	Candidate list
2. Create Skill	Write SKILL.md in Master's .github/skills/	New skill file
3. Compare Gaps	Diff heir knowledge against Master's existing coverage	Gap analysis
4. Implement	Port patterns, translate code (Python→TS if needed)	Working code
5. Test	Validate in Master context	Passing tests
6. Document	CHANGELOG entry, ROADMAP update	Release-ready

Consolidation During Promotion

Heirs naturally create granular one-capability-per-skill files during experimentation. During promotion:

1. Identify clusters — Group related heir skills by domain
2. Choose anchor skill — Pick the broadest skill in the cluster
3. Merge content — Absorb related skills into the anchor
4. Deduplicate — Remove redundancy from the merge
5. Mark inheritance — Set the promoted skill as inheritable

Anti-pattern: Promoting every heir skill as-is without consolidation review causes skill sprawl.

Code Translation Patterns (Heir → Master)

When porting from Python heirs to TypeScript Master:

Python	TypeScript
dataclass	interface
raise Exception	throw new Error
**kwargs	Optional config object
async def	async function
try/except	try/catch

Skill Inheritance Classification

Curation Rule

Ask: "Is this skill ONLY useful for managing the Alex repo itself?"

Answer	Classification	Example
Yes, master-repo only	master-only	release-preflight, heir-curation
No, any developer benefits	inheritable	deep-thinking, meditation, security-review

Truly Master-Only Skills (small list)

Only a few skills are genuinely master-only:

• heir-curation — Managing what heirs receive
• heir-sync-management — This skill
• master-alex-audit — Master workspace auditing
• release-preflight — Marketplace publishing
• release-process — Release pipeline

Heir Type Comparison

Heir	Type	Translation	Deploy Mechanism	Maintenance Cost
VS Code Extension	Source	Compile only	npx vsce publish	Low
M365 Copilot Agent	Translation	Full export/schema mapping	Teams Developer Portal	High
GitHub Copilot Web	Deployment	None (.github/ files only)	git commit + push	Very Low

Everything else should be inheritable unless it references Master-specific file paths or workflows.

Heir-Specific Positioning

Each platform heir must position against its native competitor, not a generic category:

Heir	Compares Against	Not Against
VS Code Extension	GitHub Copilot (native)	"AI assistants" generically
M365 Agent	Microsoft 365 Copilot	"AI assistants" generically
GitHub Copilot Web	GitHub Copilot (untuned)	"AI assistants" generically

Store descriptions, README headers, and comparison tables must use platform-specific language and keywords.

Release Pipeline Integration

The release script must enforce sync before packaging:

1. Run sync-architecture.js (copies Master → Heir)
2. Apply post-sync transformations (clean slate)
3. Run PII validation gate (blocks on contamination)
4. Check BUILD-MANIFEST.json timestamp (prevents stale packaging)
5. Package and publish

Rule: It must be impossible to publish stale content through the official release process.