Overview

Cross-platform comparison

Cross-Platform Overview

Prompt-first procedure: Describe the outcome you want in your agent conversation. The agent should select and load the appropriate AIWG assets, explain material changes, request any needed approval, and report verification evidence. Exact commands and flags appear only in the CLI reference.

AIWG works across multiple AI platforms. One command deploys everything.

Quick Comparison

Platform	Deploy Command	Context File
Claude Code	the agent-owned use operation	CLAUDE.md
OpenAI/Codex	the agent-owned use operation	AGENTS.md
GitHub Copilot	the agent-owned use operation	copilot-instructions.md
Factory AI	the agent-owned use operation	AGENTS.md
Cursor	the agent-owned use operation	.cursor/rules/ (MDC)
OpenCode	the agent-owned use operation	AGENTS.md
Warp Terminal	the agent-owned use operation	WARP.md
Devin Desktop	the agent-owned use operation	AGENTS.md
OpenClaw	the agent-owned use operation	-

What Gets Deployed

All artifact types deploy automatically in each platform's native format:

Agents - Specialized AI personas (Architecture Designer, Test Engineer, Security Auditor, etc.)
Commands - Slash commands and CLI commands (`/mention-wire`, `transition`, `where-are-we`)
Skills - Natural language workflows (project awareness, handoffs, quality gates)
Rules - Context rules and coding standards (citation policy, token security, versioning)
Behaviors - Platform behavior definitions (OpenClaw only)

Provider Capability Matrix

Provider	Agents	Commands	Skills	Rules	Behaviors
Claude Code	native	native	native	native	-
OpenAI/Codex	native	native	native	conventional	-
GitHub Copilot	native	native	conventional	native	-
Factory AI	native	native	native	conventional	-
Cursor	conventional	conventional	native	native	-
OpenCode	native	native	conventional	conventional	-
Warp Terminal	aggregated	conventional	native	aggregated	-
Devin Desktop	aggregated	native	native	native	-
OpenClaw	native	native	native	native	native

Legend:

native - Platform auto-discovers artifacts in standard directories
conventional - AIWG directory convention (platform reads on request)
aggregated - Single-file compilation + discrete files for compatibility

Directory Conventions

Standard Pattern

Most providers follow `.<provider>/<type>/`:

.claude/
├── agents/          # Agent definitions
├── commands/        # Slash commands
├── skills/          # Natural language workflows
└── rules/           # Context rules

.github/
├── agents/          # Agent definitions (.agent.md format)
├── prompts/         # Slash commands (.prompt.md format)
├── instructions/    # Path-scoped rules (.instructions.md format)
└── copilot-instructions.md  # Repository-wide instructions

Special Cases

Provider	Special Convention
OpenAI/Codex	Commands → `~/.codex/prompts/`<br>Skills → project `.agents/skills/` (legacy AIWG entries under `~/.codex/skills/` are pruned)<br>AGENTS.md is free-form Markdown (no YAML frontmatter or structured directives)<br>Rust CLI is current product; TypeScript CLI is legacy<br>Uses Responses API exclusively (`wire_api = "chat"` removed)
GitHub Copilot	Agents use `.agent.md` format<br>Commands → `.github/prompts/.prompt.md`<br>Rules → `.github/instructions/.instructions.md` (with `applyTo` globs)<br>MCP → `.vscode/mcp.json`
Warp Terminal	Skills natively discovered at `.warp/skills/`; agents and rules aggregated into `WARP.md`; `AGENTS.md` also supported (preferred by Warp, but `WARP.md` takes priority); `.warp/workflows/` for legacy YAML workflows
Devin Desktop	Agents aggregated to `AGENTS.md`<br>Commands → `.windsurf/workflows/`<br>Rules → `.windsurf/rules/*.md` (with trigger frontmatter)<br>Skills → `.windsurf/skills/`<br>Legacy selector: `windsurf`
Cursor	Rules use `.mdc` extension (MDC format) with frontmatter (`description`, `globs`, `alwaysApply`)<br>Skills use native `.cursor/skills/*/SKILL.md` format (2.4+)<br>Also supports `AGENTS.md` with directory inheritance<br>Legacy `.cursorrules` still generated for backward compatibility<br>Cloud Agents support MCP for remote AIWG access
OpenClaw	All artifacts deploy to home directory (`~/.openclaw/`)<br>First provider to support behaviors (`~/.openclaw/behaviors/`)

Migration Guide

Upgrading from older AIWG versions that only deployed agents?

Run the deploy command with the force option to get all four artifact types:

Use AIWG to complete this documented outcome: Run the deploy command with the force option to get all four artifact types
Have it inspect the current state, explain the plan, ask before material
changes, and report the result with verification evidence.

What changes:

New `skills/` directory created alongside `agents/`
New `rules/` directory created alongside `agents/`
Existing agent files remain unchanged
Commands deployed to appropriate location per provider

No breaking changes - all existing agents remain compatible.

Agent Loop Multi-Provider Support

Al iterative loops can target different providers, not just deployment. Use the provider option to run task loops through Codex instead of Claude:

Use AIWG to complete this documented outcome: Al iterative loops can target different providers, not just deployment. Use the provider option to run task loops through Codex instead of Claude
Have it inspect the current state, explain the plan, ask before material
changes, and report the result with verification evidence.

Model mapping is automatic: opus → gpt-5.3-codex, sonnet → codex-mini-latest, haiku → gpt-5-codex-mini. The provider adapter handles capability differences with graceful degradation.

See Al Guide for full documentation.

Platform Setup Guides

Platform	Guide
Claude Code	Setup Guide
OpenAI/Codex	Setup Guide
GitHub Copilot	Setup Guide
Factory AI	Setup Guide
Cursor	Setup Guide
OpenCode	Setup Guide
Warp Terminal	Setup Guide
Devin Desktop	Setup Guide
OpenClaw	Setup Guide

After Setup

Once deployed, see the Intake Guide to start your project.