Recommended pattern for running gstack with Codex in Conductor when existing setup is Claude-based (global + project history)

Question: What is recommend/best practice suggestion for running gstack with Codex alongside existing Claude-based setup in Conductor

Label: question

---

## Context

I’m using Conductor (conductor.build) as my IDE with a multi-model workflow:

- Claude Code → primary host for gstack (historically)
- Codex → used in parallel for build, test, QA

Repo:
- `/Users/xadmin/conductor/workspaces/agent-skill-manager`
- Mature codebase with existing agent structure (AGENTS.md, .gstack, .context, etc.)
- Significant prior usage of gstack via Claude

---

## Current Situation

Due to recent Claude usage limit changes, I’m hitting limits frequently during normal development workflows.

- Claude usage is governed by session and token-based limits across all surfaces (Claude app, Claude Code, etc.) :contentReference[oaicite:0]{index=0}
- Limits can be reached faster depending on prompt size, conversation length, and tool usage :contentReference[oaicite:1]{index=1}
- Recently, Anthropic has also tightened effective usage during peak hours, causing sessions to deplete faster :contentReference[oaicite:2]{index=2}

Because of this, I’ve started using **Codex in between Claude sessions** to continue building without waiting for limits to reset.

---

## Environment Details

### Global installs
- Claude gstack exists and is actively used
- Codex gstack appears to be an alias/symlink pointing into Claude-generated gstack artifacts:
- `~/.codex/skills/gstack → ~/.claude/skills/gstack/.agents/skills/gstack`

### Project-level
- No `.agents/skills/gstack` currently exists in the repo
- Existing project already has its own agent conventions

---

## Goal

Temporarily run gstack via Codex inside Conductor:

- without breaking or modifying existing Claude setup
- without cross-contaminating Claude and Codex environments
- with minimal setup overhead
- easy to revert back to Claude later
- aligned with intended gstack architecture (not a hack)

---

## Questions

1. What is the **recommended approach** for running gstack across multiple hosts (Claude + Codex) in the same repo?

2. Is the intended pattern:
- global install per host, or
- project-level `.agents/skills/gstack` per repo?

3. Is it safe or recommended to:
- rely on a Codex global install that aliases into Claude-generated gstack artifacts?
- or should Codex always have its own independent install?

4. For my use case, would you recommend:
- fresh clone into `.agents/skills/gstack` + `./setup --host codex`
- copying from existing Claude gstack
- something else entirely?

5. How does gstack expect:
- host-specific setup (`--host codex` vs `--host claude`)
- generated `.agents` trees
- and skill discovery
to coexist without conflicts?

6. Does Conductor’s project-level `.agents/skills` override global installs in the intended design?

---

## What I’m trying to avoid

- Breaking my existing Claude-based gstack workflows
- Creating conflicting or duplicated agent trees
- Subtle bugs from mixing host-generated artifacts
- Overengineering (new repo/workspace) if not necessary

---

## Ask

What is the **cleanest, intended, and future-proof way** to run gstack with Codex in this scenario?

A concrete recommendation (with reasoning) would be very helpful.