Gemini CLI

Google’s open-source CLI for the Gemini model family. Significantly more capable than early wiki assessments suggested — has native equivalents for most CC f…

Gemini CLI

Google’s open-source CLI for the Gemini model family. Significantly more capable than early wiki assessments suggested — has native equivalents for most CC features.

Install: npm install -g @google/gemini-cli

---

Feature Parity with Claude Code

CC Feature	Gemini Equivalent	Parity
CLAUDE.md rules	GEMINI.md hierarchical discovery	High
@file.md imports	@file.md imports (same syntax)	High
Skills	activate_skill tool + .gemini/skills/	High
Custom commands	TOML files in .gemini/commands/	High
Subagents	Experimental subagents	Medium
Hooks	Hooks support	Medium
Headless mode	Headless mode (--headless)	High
Sandboxing	Sandboxing (containerized)	Medium
MCP servers	MCP servers (HTTP + stdio)	High
Plan mode	enter_plan_mode tool	High

---

Context System: GEMINI.md

Hierarchical instruction loading:

1. ~/.gemini/GEMINI.md — global, all sessions
2. Project-root GEMINI.md — workspace scope
3. Subdirectory GEMINI.md — JIT loaded when agent navigates into dir

Supports @file.md imports (relative paths). /memory add <text> appends to global file.

context.fileName in settings.json can be configured to also read ["AGENTS.md", "GEMINI.md"], making Gemini natively read AGENTS.md files.

---

Custom Commands: TOML Format

.toml files in .gemini/commands/ (project) or ~/.gemini/commands/ (global).

description = "Reviews a pull request."
prompt = """
Review PR: {{args}}
Run: !{git log --pretty=format:"%s" -n 5}
"""

Syntax features:

• {{args}} — user arguments injected into prompt (raw outside !{}, shell-escaped inside)
• !{command} — shell command executed, output injected; security confirmation required
• @{path/to/file} — file content injected into prompt (processed before !{})
• Namespacing: .gemini/commands/git/commit.toml → /git:commit
• MCP prompts also appear as slash commands

---

Skills: activate_skill Tool

Gemini has a native skill system via the activate_skill tool. Skills live in .gemini/skills/ directory. When triggered, Gemini loads the skill’s instructions into context (same progressive disclosure principle as CC).

/tools             # List all tools including activate_skill

Key difference from CC: Gemini uses activate_skill as a native tool; CC uses the Skill meta-tool with isMeta: true message injection.

---

Built-in Tool Set

Full list from official docs:

Tool	Category	Description
run_shell_command	Execute	Arbitrary shell; manual confirmation
glob, grep_search, list_directory	Search/Read	File navigation
read_file, read_many_files	Read	File content; @ shorthand
write_file, replace	Edit	File modification; confirmation
ask_user, write_todos	Interaction	Clarification, task tracking
save_memory	Memory	Persists to GEMINI.md
activate_skill	Memory	Loads skill from .gemini/skills/
get_internal_docs	Memory	Accesses Gemini CLI own docs
enter_plan_mode, exit_plan_mode	Planning	Read-only planning phase
google_web_search, web_fetch	Web	Search + URL fetch
list_mcp_resources, read_mcp_resource	MCP	MCP server integration
Tracker tools (tracker_*)	Experimental	Task graph tracking

Shell shorthand: !command runs run_shell_command. @path triggers read_many_files.

---

Migration Mapping (CC → Gemini)

CC asset	Gemini equivalent
~/.claude/CLAUDE.md	~/.gemini/GEMINI.md
claude-setup/rules/*.md	Same files, referenced via @rules/*.md imports
claude-setup/skills/*/SKILL.md	.gemini/skills/*/SKILL.md + activate_skill
.claude/commands/*.toml	.gemini/commands/*.toml (same format)
claude-setup/agents/*.md	Experimental subagents (not production-ready)
templates/AGENTS.md	Not natively read; use GEMINI.md or configure context.fileName

Related Pages

• Claude Code → Cross-Platform Migration Matrix — full migration matrix
• AGENTS.md (format) — AGENTS.md; Gemini can be configured to read it
• Agent Skills — skill architecture comparison (CC vs Gemini)