This repository helps you set up a practical AI coding workspace with:
- reusable skills
- repeatable workflows
- safety rules
- MCP server integrations
- local archive tools for project memory and search
This guide is written for non-technical and low-technical users. Copy and paste the commands exactly as shown.
- OpenAI catalog under
Agent/OpenAI/ - Anthropic catalog under
Agent/Anthropic/ - Google catalog under
Agent/Google/ - MCP catalog under
MCP-Servers/mcp-docker-stack/docker-mcp-catalog.yaml - Local helper scripts under vendor-specific
Scripts/folders (root scripts are deprecated) - Local archive storage under
Agent-Context/(kept local, not versioned)
Install these first:
- Git
- Docker Desktop (with Docker MCP support available)
- PowerShell (Windows 10/11 is fine)
- Your AI client:
- OpenAI Codex and/or ChatGPT desktop app
- Anthropic Claude Code and/or Claude Desktop
- Google Antigravity
From a terminal:
git clone https://github.com/TheThirdRail/coding-agent-setup.git
cd coding-agent-setup
Copy-Item example.env .envOpen .env and fill in keys you actually use. Start with:
GITHUB_PERSONAL_ACCESS_TOKENBRAVE_API_KEY
Use secrets-acquisition-guide.md if you need help getting keys.
Use this path if you want Codex-style skills, rules, and automations.
cd Agent\OpenAI\Scripts
.\validate-codex-openai.ps1
.\install-rules.ps1 -DryRun
.\install-rules.ps1What this installs:
~/.codex/AGENTS.md~/.codex/rules/default.rules~/.codex/skills/*~/.codex/automations/*.automation.md
Optional mirror to ~/.agents/* too:
.\install-rules.ps1 -Target both..\..\..\MCP-Servers\scripts\install-mcp-servers.ps1 -Vendor openai -ServerNames "filesystem", "github", "desktop-commander", "context7"
..\..\..\MCP-Servers\scripts\set-mcp-secrets.ps1
..\..\..\MCP-Servers\scripts\setup_lazy_load.ps1 -ClientName "OpenAI ChatGPT/Codex"Then restart Codex/ChatGPT so it reloads config.
Use this path if you want slash-command workflows in Antigravity.
cd Agent\Google\Scripts
.\deprecation-checker.ps1 -DryRun
.\deprecation-checker.ps1
.\install-rules.ps1
.\install-skills.ps1
.\install-workflows.ps1What this installs:
~/.gemini/GEMINI.md~/.gemini/antigravity/skills/*~/.gemini/antigravity/global_workflows/*.md
..\..\..\MCP-Servers\scripts\install-mcp-servers.ps1 -Vendor google -ServerNames "filesystem", "github", "desktop-commander", "context7"
..\..\..\MCP-Servers\scripts\set-mcp-secrets.ps1
..\..\..\MCP-Servers\scripts\setup_lazy_load.ps1 -ClientName "Google Antigravity"Then restart Antigravity.
Use this path if you want Anthropic subagents with canonical CLAUDE.md policy routing.
cd Agent\Anthropic\Scripts
.\validate-rule-length.ps1
.\validate-subagents.ps1
.\deprecation-checker.ps1 -DryRun
.\install-rules.ps1
.\install-skills.ps1
.\install-subagents.ps1
.\install-settings.ps1What this installs:
~/.claude/CLAUDE.md~/.claude/rules/*.md~/.claude/skills/*~/.claude/agents/*.md~/.claude/settings.json(merged baseline)
..\..\..\MCP-Servers\scripts\install-mcp-servers.ps1 -Vendor anthropic -ServerNames "filesystem", "github", "desktop-commander", "context7"
..\..\..\MCP-Servers\scripts\set-mcp-secrets.ps1
..\..\..\MCP-Servers\scripts\setup_lazy_load.ps1 -ClientName "Anthropic Claude Code"Then restart Claude Code/Claude Desktop.
The project catalog currently defines 46 MCP servers in categories like:
- development and command execution
- search and research
- security scanning
- data and vector databases
- frontend and utility tools
Main file:
MCP-Servers/mcp-docker-stack/docker-mcp-catalog.yaml
This repo includes archive skills that use local data stores:
- semantic docs archive (Chroma)
- code structure graph (SQLite)
- durable memory notes (SQLite)
- git-history and code search helpers
Local archive path:
Agent-Context/Archives/
Important:
Agent-Context/is local-only and ignored by Git- do not store secrets in archive data
.
├── Agent/
│ ├── OpenAI/
│ │ ├── AGENTS.md
│ │ ├── default.rules
│ │ ├── Automations/
│ │ ├── Skills/
│ │ └── Scripts/
│ ├── Anthropic/
│ │ ├── Rules/
│ │ ├── Skills/
│ │ ├── Subagents/
│ │ └── Scripts/
│ └── Google/
│ ├── Rules/
│ ├── Skills/
│ ├── Workflows/
│ └── Scripts/
├── MCP-Servers/
│ ├── mcp-docker-stack/
│ └── scripts/
├── deprecated-Scripts/
├── Agent-Context/
├── example.env
├── secrets-acquisition-guide.md
└── README.md
If you use OpenAI Codex:
Agent/OpenAI/AGENTS.mdAgent/OpenAI/Scripts/script-commands.mdcodex-sample-workflow.md
If you use Google Antigravity:
Agent/Google/Rules/GEMINI.mdAgent/Google/Scripts/script-commands.mdantigravity-sample-workflow.md
If you use Anthropic Claude Code:
Agent/Anthropic/Rules/CLAUDE.mdAgent/Anthropic/Scripts/script-commands.mdclaude-code-sample-workflow.md
Use:
Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser- make sure Docker Desktop is running
- make sure your
.envfile exists and has needed keys - rerun
MCP-Servers/scripts/set-mcp-secrets.ps1
- restart the client after install scripts run
OpenAI:
cd Agent\OpenAI\Scripts
.\install-rules.ps1Google:
cd Agent\Google\Scripts
.\install-rules.ps1
.\install-skills.ps1
.\install-workflows.ps1Anthropic:
cd Agent\Anthropic\Scripts
.\install-rules.ps1
.\install-skills.ps1
.\install-subagents.ps1
.\install-settings.ps1MIT