This document defines core terminology used throughout the project.
A specialized subagent that handles delegated tasks within a constrained scope. Invoked using the @agent-name syntax in Gemini CLI.
- frontmatter: Defined with
name,description, andtoolsfields. - Tools: Uses Gemini tools like
read_file,run_shell_command,write_file, etc. - Location:
agents/*.md(in repo),~/.gemini/agents/(when installed).
A collection of workflow definitions and domain knowledge. Referenced by agents or commands.
- Curated Skills: Located in
skills/<skill-name>/SKILL.md(shipped in repo). - Learned Skills: Located in
~/.gemini/skills/learned/(not shipped). - Imported Skills: Located in
~/.gemini/skills/imported/(not shipped).
Slash commands for quick execution. Gemini CLI uses .toml format for commands.
description = "Command description"
prompt = '''
# Command Content
...
'''- Gemini CLI:
.tomlfiles, located in~/.gemini/commands/. - Antigravity:
.mdfiles, used as workflows.
Triggers that execute automatically in response to tool events. Defined in hooks/hooks.json.
Always-follow coding guidelines that the AI must adhere to. Located in GEMINI.md or ~/.gemini/rules/.
Gemini CLI and Claude Code use different tool names:
| Claude Code | Gemini CLI |
|---|---|
Read |
read_file |
Write |
write_file |
Edit |
replace |
Bash |
run_shell_command |
Grep |
search_file_content |
Glob |
list_directory |
WebSearch |
google_web_search |
| Component | Gemini CLI | Antigravity |
|---|---|---|
| Agents | ~/.gemini/agents/ |
~/.gemini/antigravity/global_agents/ |
| Skills | ~/.gemini/skills/ |
~/.gemini/antigravity/global_skills/ |
| Commands | ~/.gemini/commands/ |
(Used as workflows) |
| Rules | ~/.gemini/rules/ |
~/.gemini/antigravity/global_rules/ |
The environment or platform running the AI agents:
- Gemini CLI: Terminal-based.
- Antigravity: IDE-integrated (e.g., VS Code, Cursor).
Origin information for learned or imported skills. Recorded in .provenance.json.
Required fields:
source: Origin (URL, path, or identifier).created_at: ISO 8601 timestamp.confidence: 0–1 confidence score.author: Entity that produced the skill.