Install for Codex, Copilot, Gemini, and other agents¶

Hosts without a plugin system read skills from the repository. install/install.sh vendors the six skills into .agents/skills/, merges MCP configuration for the hosts you use, and adds instruction files — without overwriting anything you already have.

Install¶

From the root of the Mule project you want to configure:

curl -fsSL https://raw.githubusercontent.com/Avinava/mule-skills/main/install/install.sh | bash

Or, from a clone of this repository:

./install/install.sh --target /path/to/your-mule-project

By default the script detects which hosts you use and configures only those. Claude Code is deliberately excluded from detection — install the plugin instead of vendoring a second copy. Pass --hosts claude if you want the vendored copy anyway.

Always preview first:

./install/install.sh --target /path/to/project --dry-run

Options¶

Option	Effect
`--target DIR`	Project to install into (default: current directory)
`--hosts LIST`	`claude`, `codex`, `copilot`, `vscode`, `gemini`, or `auto` (default), `all`, `none`
`--dry-run`	Print every change and exit without writing
`--no-mcp`	Install skills and instruction files but no MCP configuration
`--force`	Overwrite existing instruction files (never `AGENTS.md`)
`--ref REF`	Git ref to fetch when the script bootstraps itself

The script is idempotent. Skills are replaced on every run; MCP entries, instruction files, and AGENTS.md are only ever added when missing.

What it writes¶

your-mule-project/
├── .agents/skills/                  # the eight mule-* skills
├── .mcp.json                        # Claude Code, Copilot CLI, Gemini
├── .vscode/mcp.json                 # VS Code and Copilot Chat
├── .codex/config.toml               # Codex
├── .github/copilot-instructions.md  # GitHub Copilot
├── AGENTS.md                        # shared project context
├── CLAUDE.md                        # only when the claude host is selected
└── GEMINI.md                        # only when the gemini host is selected

Existing MCP servers in those files are preserved — the script adds only the entries that are missing. If a config file is not plain JSON (VS Code allows comments and trailing commas, which the merger cannot read), the script skips that file, prints the server names to add by hand, and continues with the rest of the install rather than guessing or aborting halfway.

AGENTS.md is never overwritten, including with --force. It holds project context you wrote and cannot be regenerated from a template.

Host reference¶

Host	Instructions read	MCP configuration	Verify with
Codex CLI, desktop, IDE extension	`AGENTS.md`	`.codex/config.toml`	`codex mcp list`
GitHub Copilot in VS Code	`.github/copilot-instructions.md`, `AGENTS.md`	`.vscode/mcp.json`	Reload VS Code, check its MCP server list
GitHub Copilot CLI	`.github/copilot-instructions.md`, `AGENTS.md`	`.mcp.json`	`copilot mcp list`
Gemini coding agents	`GEMINI.md`, `AGENTS.md`	`.mcp.json` where supported	The host's MCP status view
Other compatible agents	`AGENTS.md` plus whatever the host supports	`.mcp.json` when the host accepts `mcpServers`	The host's documented MCP check

Codex discovers repository skills from .agents/skills/ in directories between the working directory and the repository root — see the Codex skills documentation and Codex MCP documentation.

Local MCP files do not configure GitHub-hosted Copilot agents or code review; configure hosted MCP access through repository settings.

Manual install¶

If you cannot run the script, clone this repository and copy the eight directories under skills/ into .agents/skills/, then merge the matching file from install/hosts/ into your host's MCP config and copy the instruction templates from install/templates/.

The templates carry two placeholders the script normally fills in. Replace them by hand:

Placeholder	Replace with
`<!-- SKILLS_LOCATION -->`	A line stating that the skills live under `.agents/skills/`
`<!-- PROJECT_NAME -->`	Your project's name

The agent-followable version of this procedure is in agent-install.md.

Next¶

Fill in AGENTS.md from this project's actual evidence — see project-setup.md.

For runtime logs, metrics, or deployment evidence, authenticate the one server that needs it: Anypoint access. Skills that use runtime or design-platform evidence check access themselves and offer alternatives, so this is optional.