AI agent CLIs

Terminal coding agents (2026). Same job, different ecosystems — pick per project, or run more than one.

Claude CodeAnthropic — CLAUDE.md, slash commands, MCP, plan modeClaude Code advancedPermissions, -p / CI, allowlists, sandbox, dangerous flagsClaude Code pluginsMarketplaces, scopes, skills/hooks/MCP bundlesCodex CLIOpenAI — codex in terminal, ChatGPT or API authGemini CLIGoogle — open-source agent, npm install @google/gemini-cliGrok BuildxAI — grok TUI + headless, skills/plugins/MCPGrok Build advancedPermissions, -p / CI, worktrees, sandbox, always-approve

Shared good habits

1. Repo on Linux FS in WSL (~/code/...) — agents run lots of tools; /mnt/c is painful
2. Commit or worktree before big agent runs — easy rollback
3. Project memory files — CLAUDE.md, AGENTS.md, .gemini/ rules as applicable
4. Skills — reusable agent playbooks: apsolut-skills
5. Secrets — never paste keys into chat; use env / direnv

When to use which (rough)

Need	Try
Deep Anthropic / Claude workflow, MCP-heavy	Claude Code
ChatGPT subscription agent in terminal	Codex
Open-source Google stack, Gemini models	Gemini CLI
xAI Grok in terminal (TUI / SuperGrok path)	Grok Build

Skills repo

Agent skills are not documented fully here — they live in:

https://github.com/apsolut/apsolut-skills

See Skills page.