Marketplace for trusted AI agent skills

Skill Detail

Plan, generate and review Gemini images in Claude Code with Banana Claude

Turn a visual brief into an approved Gemini image or edit from Claude Code, then inspect the saved pixels against explicit acceptance checks.

Image & Creative AutomationClaude Code
Image & Creative Automation Claude Code Security Reviewed
⭐ 1.1k GitHub stars
COPY SKILL INSTRUCTIONS (OPTIONAL)
npx skills add agentskillexchange/skills --skill plan-generate-and-review-gemini-images-with-banana-claude Copy
Uses the third-party skills CLI, not an ASE-owned installer. Check your agent’s compatibility. This copies instructions; complete the upstream tool setup below separately.
View source Documentation
At a glance
Tools required
Claude Code 2.1.199+; Python 3.11+ (python3); bundled Banana Claude MCP process; billing-enabled Google AI project and Gemini API key
Install & setup
Requirements: Claude Code 2.1.199+, Python 3.11+ with python3 on PATH, and a Gemini API key for a billing-enabled Google AI project. In Claude Code, run: /plugin marketplace add AgriciDaniel/banana-claude /plugin install banana-claude@banana-claude-marketplace /plugin enable banana-claude@banana-claude-marketplace /reload-plugins Supply google_ai_api_key through the sensitive plugin configuration prompt; do not put the key in chat or a command. Start with /banana-claude:banana generate followed by the visual brief. The plugin installs disabled and requires explicit enabling. Review the offline plan before approving any paid provider attempt. Existing 1.4.1 or 2.1.0 users must follow the upstream upgrade guide before enabling version 3.
Author
AgriciDaniel
Publisher
Community
Last updated
Oct 19, 2026
Quick brief

Use Banana Claude when Claude Code should turn a visual brief into a planned, approved and reviewed Gemini image generation or editing job. The operator specifies the intended asset, required copy, references, output dimensions and visible acceptance checks. The skill freezes those requirements, compiles the prompt, plans a provider request, obtains approval, saves the result and inspects the actual pixels against the brief.

How it works

What this skill actually does

Invoke it for repository-adjacent creative work where prompt review, reference constraints, output paths and a visible quality decision matter: for example, a hero image with a protected copy area or a product scene with an explicit preservation edit. It is useful when the agent should manage the whole brief-to-reviewed-artifact loop rather than leave the operator to copy prompts and inspect files across separate tools.

Installation and first invocation

Requires Claude Code 2.1.199 or newer, Python 3.11 or newer available as python3, and a Gemini API key for a billing-enabled Google AI project. Run these slash commands inside Claude Code:

/plugin marketplace add AgriciDaniel/banana-claude
/plugin install banana-claude@banana-claude-marketplace
/plugin enable banana-claude@banana-claude-marketplace
/reload-plugins

The plugin installs disabled. Supply google_ai_api_key through the sensitive plugin configuration prompt; never paste it into chat, source control or a command. The bundled local Python MCP process receives the configured key. Start with:

/banana-claude:banana generate a restrained 16:9 hero image for a ceramics studio with clear left-side copy space

Users upgrading from public 1.4.1 or 2.1.0 installations should follow the upstream upgrade guide before enabling version 3, including its review-first legacy configuration checks.

Bounded operator workflow

  1. Freeze the asset brief: goal, exact copy, protected details, allowed changes, references, destination and review criteria. For uploaded media, obtain the upstream-required authority statement covering rights, likeness, private media, intended use and transmission to Google.
  2. Inspect current model capabilities using the bundled model reference or model tool. Plan without a paid provider call. Review the exact prompt, model, dimensions, reference disclosures, storage settings, destination and nominal estimate.
  3. Obtain explicit approval for that exact request. One approval allows one provider attempt; changed plans, fixes and retries need fresh approval. A nominal estimate is not an invoice cap.
  4. Execute the unchanged approved request through the plugin, collect the image and metadata sidecar, and inspect every returned image against the frozen brief. Report Pass, Targeted fix, Regenerate or Blocked; successful transport alone does not establish visual quality.
  5. Return the output paths and review result. Stop at the reviewed local artifact. Publishing, campaign deployment and automatic retries are outside this workflow.

The upstream skill also supports bounded comparisons of up to three prompts across three model routes, with no more than nine provider attempts and explicit portfolio approval. Use this only when comparing alternatives serves the brief. Exact text, identity, geometry and factual accuracy still require pixel review.

The scope is a Claude Code visual-production skill with a brief, an approved request and inspectable artifacts. This entry does not represent a general Gemini SDK, hosted design platform or arbitrary MCP server directory. Gemini is the provider; Claude Code is the invocation framework.

Canonical skill instructions · User guide · Upstream security and privacy documentation. Metadata and documentation were verified; paid generation and installation were not executed, and no security-reviewed status is claimed.