gemini-skill-install 0.1.4

Bootstrap installer for the gemini Claude Code skill

gemini-skill

gemini-skill is a Gemini API front end that works in two modes:

• as a Claude Code skill via /gemini ...
• as a direct CLI via python3 scripts/gemini_run.py ...

It exposes the same command surface in both modes: text, multimodal analysis, structured output, embeddings, Files API, image/video/music generation, file search, deep research, and iterative plan review.

Quick Start

Install for Claude Code

Recommended:

uvx gemini-skill-install

Fallback:

uvx --python 3.13 gemini-skill-install

Or from a clone:

git clone https://github.com/reshinto/gemini-skill.git
cd gemini-skill
python3 setup/install.py

The installer copies the runtime payload into ~/.claude/skills/gemini/, creates or reuses ~/.claude/skills/gemini/.venv, installs the pinned google-genai SDK, and writes the canonical env block into ~/.claude/settings.json.

Configure credentials

The launcher resolves canonical Gemini env keys from the current working directory first, then Claude settings files, then existing process env:

1. ./.env
2. ./.claude/settings.local.json
3. ./.claude/settings.json
4. ~/.claude/settings.json
5. existing process env

Supported keys:

GEMINI_API_KEY
GEMINI_IS_SDK_PRIORITY
GEMINI_IS_RAWHTTP_PRIORITY
GEMINI_LIVE_TESTS

For repo-local CLI use:

cp .env.example .env

GEMINI_API_KEY is the only supported secret name. GOOGLE_API_KEY is ignored.

Use it from Claude Code

The skill manifest sets disable-model-invocation: true, so Claude Code will not start this skill on its own. Invoke it explicitly first.

/gemini text "Explain quantum computing in three sentences"
/gemini plan_review "Review this implementation plan for gaps and rollout risks"

Use it as a direct CLI

From a checkout:

python3 scripts/gemini_run.py text "hi"
python3 scripts/gemini_run.py plan_review "Review this migration plan"
python3 scripts/gemini_run.py plan_review

The last command starts the interactive plan_review REPL when stdin is a TTY.

Core Workflows

• text for one-shot prompts and multi-turn sessions
• plan_review for iterative plan review, either one-turn or REPL
• multimodal for PDFs, images, audio, video, and URLs
• structured for schema-constrained JSON output
• embed, token_count, files, cache, batch, and file_search
• image_gen, imagen, video_gen, and music_gen
• search, maps, computer_use, deep_research, and live

Documentation

• docs/install.md — install paths, env precedence, troubleshooting
• docs/usage.md — Claude Code skill usage and direct CLI usage
• docs/usage-tour.md — end-to-end examples
• docs/commands.md — command index by capability family
• reference/index.md — per-command reference pages
• docs/security.md — secrets handling, local data storage, privacy notes
• docs/how-it-works.md — launcher, dispatch, transport, and output flow
• docs/README.md — documentation hub

Backend Routing

The launcher and adapters are backend-agnostic. The transport coordinator chooses the SDK or raw HTTP backend from the two routing flags:

• GEMINI_IS_SDK_PRIORITY=true
• GEMINI_IS_RAWHTTP_PRIORITY=false

Both backends produce the same normalized response shape for callers.

License

MIT