Gemini CLI — setup guide

The canonical context file is AGENTS.md at the repo root. Gemini CLI reads it via .gemini/settings.json (context.fileName). This guide covers Gemini-specific setup only.

Install

gemini extensions install https://github.com/wshobson/agents
cd ~/.gemini/extensions/claude-code-workflows
make generate HARNESS=gemini
# restart Gemini CLI

What you get

• 156 skills at skills/<plugin>__<skill>/SKILL.md — described in AGENTS.md. Describe a task to activate.
• 192 subagents at agents/<plugin>__<agent>.md — invoke with @<agent>.
• 102 slash commands at /<plugin>:<command> — use /help to list.

Companion Memory Extension

Pensyve is maintained upstream as a separate Gemini CLI extension with MCP-backed memory and context injection:

gemini extensions install https://github.com/major7apps/pensyve

Gemini-specific differences

Capability	Claude Code	Gemini CLI
Plugin installation	/plugin install	gemini extensions install <url>
Context file	reads CLAUDE.md (a symlink to AGENTS.md)	reads via .gemini/settings.json redirect to AGENTS.md
Per-agent tool allowlist	tools: (always)	tools: (honored — remapped to Gemini-native names)
Skill / agent discovery	native	native (skills/, agents/ at extension root)
Model assignment	per-agent	session-level (override via model: frontmatter)
TodoWrite tool	yes	no equivalent

Regenerating

make generate HARNESS=gemini                            # all plugins
make generate HARNESS=gemini PLUGIN=javascript-typescript   # one plugin
make clean-generated HARNESS=gemini                     # remove output

See also

• AGENTS.md — canonical context (cross-harness conventions)
• docs/harnesses.md — full capability matrix
• docs/round-trip-results.md — Gemini round-trip verification recipe