Agent Stack Guide

Agent Stack Guide is a practical guide for teams that want to run multiple AI coding agents in parallel without branch collisions or unsafe Git operations.

The guide focuses on:

• Isolating each agent in its own Git worktree
• Enforcing branch/worktree safety guardrails
• Standardizing instructions for Codex, Claude Code, and GitHub Copilot
• Keeping human review as the final merge/discard control point

What You Get

• A full workflow guide: GUIDE.md
• A navigation index: INDEX.md
• Agent/file mapping and setup model: AGENT_CONFIGURATION_GUIDE.md
• Commands-only quickstart: QUICKSTART_COMMANDS.md
• GitHub PR variant with branch protection: GITHUB_PR_WORKFLOW.md
• Human decision checklist: REVIEW_MERGE_OR_DISCARD.md
• Reusable templates and hooks in templates/
• Bootstrap scripts in scripts/ for Bash and PowerShell

Recommended Start

1. Open INDEX.md.
2. Follow GUIDE.md sections in order.
3. Install templates with:
   • scripts/setup-agent-guardrails.sh
   • scripts/setup-agent-guardrails.ps1

Installer behavior: if a repository already has hooks, the installer will back them up to .git/hooks/<hook>.orig and install a dispatcher that runs guardrails checks first and then the original hook.

4. Keep guardrails on by default; switch off only for local experiments only.

Scope

This repository is documentation and automation glue for process safety. It does not include project-specific CI, test commands, or language-specific build logic.