zlh-grind/CLAUDE_GPT_COORDINATION.md

5.0 KiB

Claude ↔ GPT Coordination Guide

For GPT (Ceàrd) to read when context feels stale or unclear.

Last Updated: April 10, 2026


The Role Split

AI Repo What it owns
Claude jester/knowledge-base Architecture, strategy, canonical decisions, owner profile
GPT/Ceàrd jester/zlh-grind Execution, implementation, session handovers, debug notes

Both AIs share the same Gitea MCP (https://git.zerolaghub.com/mcp).


What Claude Does to Repos

Claude writes to knowledge-base

Claude is the sole author of knowledge-base. It writes and maintains:

  • ZeroLagHub_Master_Bootstrap_Apr2026.md — platform state, architecture, infrastructure
  • OWNER_PROFILE.md — James's background, working style, solo operator context
  • GHOST_SHELL_VISION.md — long-term platform strategy
  • CLAUDE_PROJECT_INSTRUCTIONS.md — Claude's own session start instructions
  • README.md — index and session start URLs

GPT should not write to knowledge-base. If GPT needs an architecture decision recorded, it should flag it to James who will bring it to Claude.

Claude writes to zlh-grind only when asked

Claude does not routinely write to zlh-grind — that is GPT's workspace. Claude may write to zlh-grind when:

  • James asks Claude to create or update a specific file
  • Claude is archiving resolved SCRATCH files (housekeeping)
  • Claude is writing a coordination note like this one

When Claude writes to zlh-grind, it will note what it changed so GPT is not surprised.


What GPT Should Own in zlh-grind

GPT is responsible for keeping these files current:

File When to update
OPEN_THREADS.md After completing work, discovering new issues, or changing priorities
SCRATCH/handover-*.md At the end of every session — write a new handover or update the current one
INFRASTRUCTURE.md If IPs or VM inventory changes
PROJECT_CONTEXT.md If stack or naming conventions change

GPT must not mark work as complete unless it has been verified.


SCRATCH/ Directory Rules

SCRATCH/ is for active working notes — session docs, debug findings, specs.

SCRATCH/archive/ is for resolved work. Files get moved there (not deleted) when the problem they describe is fully resolved.

Who archives: Either Claude or GPT can archive SCRATCH files. Claude did a bulk archive on April 10, 2026 covering all Denver migration and FQDN migration notes. Going forward GPT should archive its own resolved SCRATCH files as part of normal session hygiene.

Currently active SCRATCH files (as of Apr 10, 2026):

  • handover-apr-2026.md — primary handover, current state
  • session-stabilization-fabric-findings.md — active bug, agent fix needed
  • velocity-plugin.md — reference doc
  • service-discovery.md — reference doc
  • minecraft-velocity-forwarding.md — reference doc

Infrastructure Reality (Critical)

All active VM IDs are in the 9000s range on GTHost Detroit.

VM IDs in the 100s, 300s, 2000s range are legacy/unused — old Pterodactyl-era VMs kept for reference only. Do not treat them as active.

Denver host was fully decommissioned April 2, 2026.

Authoritative VM/IP list: INFRASTRUCTURE.md


Knowledge Base — What's There Now

Claude rebuilt knowledge-base on April 10, 2026. The following files are current and accurate:

File Purpose
README.md Index, session start URLs for both AIs
ZeroLagHub_Master_Bootstrap_Apr2026.md Full platform state, architecture rules, pre-launch blockers
OWNER_PROFILE.md James's background, AI workflow role, solo operator context
GHOST_SHELL_VISION.md Ghost Shell / ZLH platform strategy
CLAUDE_PROJECT_INSTRUCTIONS.md Claude's project-level instructions

Older files (ZeroLagHub_Master_Bootstrap_Dec2025.md, etc.) are still in the repo for historical reference but are superseded.


Session Start Checklist for GPT

At the start of every session, read:

  1. SCRATCH/handover-apr-2026.md — current platform state
  2. OPEN_THREADS.md — active and outstanding work
  3. This file if anything feels out of sync

Before making architectural decisions, read:

  • knowledge-base/ZeroLagHub_Master_Bootstrap_Apr2026.md

If unsure whether a decision is architectural, stop and ask James to bring it to Claude.


Pre-Launch Blockers (Current Priority Order)

  1. Billing / Stripe — scaffolded, DB writes missing, launch blocker
  2. Game server world backup / restore — not started, trust-critical
  3. User onboarding flow — not started
  4. Fabric readiness gating — agent fix, supervisor.go + autostart.go
  5. Password reset flow — verify wired up
  6. Usage limits / quota enforcement — not started
  7. Email notifications — not started
  8. Velocity resync endpoint — route missing, data model ready
  9. Upload testing, stress testing, OPNsense audit, provisioning validation

See OPEN_THREADS.md for full detail.