Feedback: Personal Data vs Personal Skills

[mellanon]

Feedback: Personal Data vs Personal Skills

Summary

The "move out, link back" pattern works excellently for personal data but creates challenges for personal skills when actively developing PAI.

What Works Well

Symlinking these to a separate repo is ideal:

Directory	Why Separate Repo Works
MEMORY/	Session history, learnings - pure data, no versioning needs
USER/	Config overrides - stable preferences, rarely change
WORK/	Scratch space - ephemeral, just needs backup
CORE_USER/	TELOS, identity - personal config, not code

These are data and configuration - they don't need feature branches, don't coordinate with PAI releases, and benefit from being isolated from PAI updates.

What Doesn't Fit For Me: Personal Skills

The template suggests also symlinking skills/_ALLCAPS/ to the personal data repo. This creates problems for PAI developers:

The Coordination Problem

Scenario: Developing _SPECFIRST v2 with new quality gates

PAI worktree:                    pai-personal-data:
  feature/specfirst-v2             (linear history only)
  - New hook system                - _SPECFIRST skill
  - Framework changes              - Can't branch with PAI
  - Version 2.3                    - No version coordination

When skills need to change WITH PAI (new hooks, new framework features, breaking changes), having them in a separate linear-history repo breaks:

1. Branching - Can't have skill on feature branch with PAI changes
2. Testing - Can't test skill + PAI changes together atomically
3. Rollback - Can't revert skill + PAI together
4. Releases - Can't tag skill + PAI at same commit

The Key Distinction

Type	Examples	Versioning Need	Recommendation
Data	MEMORY, session logs	None	pai-personal-data ✓
Config	USER overrides, TELOS	Minimal	pai-personal-data ✓
Code	Skills, workflows, CLI	Full git workflow	PAI worktree

Personal skills are code, not data. The _ALLCAPS prefix means "private/personal", not "static/stable".

Recommendation

Update the template to clarify the use case:

## Who This Template Is For

**End Users** who:
- Install PAI and customize it
- Have stable personal skills that rarely change
- Don't actively develop PAI itself

**PAI Developers** should:
- Use this template for MEMORY, USER, WORK, CORE_USER only
- Keep personal skills in their PAI worktree/fork
- Skills need to version with PAI for coordinated changes

My Setup

I use pai-personal-data for:

• MEMORY/ - learnings, session history
• USER/ - config overrides
• WORK/ - scratch space
• CORE_USER/ - TELOS, identity, preferences

I keep in PAI worktree:

• skills/_SPECFIRST/ - actively developed
• skills/_CONTEXT/ - may need PAI-coordinated updates
• skills/_JIRA/ - work integration, evolves with PAI
• ...

This gives me version-controlled personal data backup while maintaining proper git workflow for skill development.

---

Feedback from a PAI developer using git worktrees for multi-version development