← 返回 Skills 市场
Nano banana 2
作者
mango_fish
· GitHub ↗
· v1.0.6
· MIT-0
316
总下载
0
收藏
1
当前安装
7
版本数
在 OpenClaw 中安装
/install nano-banana-image-gen
功能描述
Generate and edit images using Pixwith API's Nano Banana 2 model. Supports text-to-image and image-to-image (up to 4 reference images). Use when the user ask...
安全使用建议
Before installing: (1) Confirm the publisher and the skill's homepage (https://pixwith.ai) are legitimate. (2) Ask the publisher/registry to fix the metadata: the registry currently says no env vars required, but SKILL.md requires PIXWITH_API_KEY — that mismatch is suspicious. (3) Prefer not to store API keys in persistent agent config; instead supply PIXWITH_API_KEY at runtime or via a secure vault. If you must store it, be aware the key will live in ~/.openclaw/openclaw.json. (4) Verify the presigned upload flow and that upload URLs come from api.pixwith.ai / uploads.pixwith.ai; the wildcard *.amazonaws.com is explainable for S3 but widens network scope. (5) If you have sensitive images, understand uploads go to a CDN/S3 and will be accessible via the returned URLs. (6) If anything about the metadata or the instructions is unclear, request an updated skill package with consistent metadata and explicit guidance on secret handling before enabling it.
功能分析
Type: OpenClaw Skill
Name: nano-banana-image-gen
Version: 1.0.6
The skill is a standard integration for the Pixwith AI image generation service. It provides clear instructions for the agent to interact with the Pixwith API via curl, including credit checks, task creation, and image uploads to AWS S3. The requested permissions (network access to pixwith.ai and amazonaws.com, and access to user-provided files) are appropriate for its stated purpose, and there is no evidence of data exfiltration, malicious execution, or prompt injection attacks.
能力评估
Purpose & Capability
The SKILL.md describes an image-generation skill that legitimately needs a PIXWITH_API_KEY and network access to api.pixwith.ai and presigned S3 uploads; that capability matches the name and description. However the registry metadata listed earlier (Requirements) reported "Required env vars: none" and "Required config paths: none", which contradicts the SKILL.md that declares PIXWITH_API_KEY required and instructs adding it to ~/.openclaw/openclaw.json. This metadata/instruction mismatch is unexplained and suspicious.
Instruction Scope
Runtime instructions are narrowly scoped to creating tasks, polling results, and uploading user-provided images via presigned S3 URLs. The instructions explicitly limit filesystem access to user-provided image files (jpg/png) which is appropriate. However the SKILL.md also instructs the user to add their API key into ~/.openclaw/openclaw.json — an explicit instruction to persist a secret in the agent config that goes beyond only reading user-supplied images.
Install Mechanism
No install spec or code files are present; this is instruction-only and therefore writes nothing to disk at install time. That is low-risk from an installation perspective.
Credentials
The only credential the skill needs (per SKILL.md) is PIXWITH_API_KEY, which is reasonable for a Pixwith integration. But the registry metadata claiming no required env vars conflicts with the SKILL.md. Additionally, SKILL.md instructs persisting the API key into ~/.openclaw/openclaw.json (agent config), which increases the blast radius of that secret. The network permission wildcard "*.amazonaws.com" is broad but explainable (presigned S3 uploads); still worth noting.
Persistence & Privilege
Declared persistence settings say modify_agent_config: false and write_files: false, but the instructions explicitly tell the user how to add the API key to ~/.openclaw/openclaw.json (modifying agent config). That mismatch between declared persistence privileges and the runtime instructions is a notable inconsistency that should be resolved before trusting the skill.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install nano-banana-image-gen - 安装完成后，直接呼叫该 Skill 的名称或使用
/nano-banana-image-gen触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.0.6
- Added a new section emphasizing that all API response values (such as task_id, result_urls, image_url, upload_url, and fields) must be used exactly as returned, without any modifications.
- Updated instructions to highlight the importance of preserving the integrity of opaque API tokens throughout all workflows and responses.
- Clarified that result_urls must be presented to the user exactly as received from the API.
v1.0.5
- Added a new section emphasizing that all API response values (such as task_id, result_urls, image_url, upload_url, and fields) must be used exactly as returned, without any modifications.
- Updated instructions to highlight the importance of preserving the integrity of opaque API tokens throughout all workflows and responses.
- Clarified that result_urls must be presented to the user exactly as received from the API.
v1.0.4
- Added full metadata to SKILL.md, including version, publisher, homepage, categories, tags, environment variables, permissions, and capabilities.
- No functionality changes; documentation and metadata enhancements only.
- Removed obsolete metadata file (_meta.json).
v1.0.3
- Added _meta.json metadata file to the skill package.
- No functional or workflow changes to the skill logic.
v1.0.2
**This version removes detailed internal documentation and example/reference files, simplifying user guidance to essential setup and usage.**
- Removed example workflows and API references from the skill package.
- Revised setup instructions to focus on obtaining and configuring the Pixwith API key via environment variables.
- Trimmed scope to only Nano Banana 2 (model_id: 0-41); removed mention of other models or generic model discovery.
- Updated guidance for common workflows: text-to-image and image-to-image, including image upload steps and API endpoints.
- Added a pricing table, concise parameter descriptions, error handling summary, and default option values all in one place.
v1.0.1
- Minor update with no functional or behavioral changes.
- Updated setup instructions to use a bulleted list for clearer guidance.
v1.0.0
nano-banana-image-gen 1.0.0 — Initial release
- Generate and edit AI images with Nano Banana 2 on Pixwith using text prompts and reference images.
- Includes support for schema-aware options and async task polling.
- Guides users through setup, MCP server connection, and API key entry before first use.
- Supports both image and video generation workflows, model discovery, credit checks, and robust error guidance.
元数据
常见问题
Nano banana 2 是什么？
Generate and edit images using Pixwith API's Nano Banana 2 model. Supports text-to-image and image-to-image (up to 4 reference images). Use when the user ask... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 316 次。
如何安装 Nano banana 2？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install nano-banana-image-gen」即可一键安装，无需额外配置。
Nano banana 2 是免费的吗？
是的，Nano banana 2 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Nano banana 2 支持哪些平台？
Nano banana 2 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Nano banana 2？
由 mango_fish（@tate-kt）开发并维护，当前版本 v1.0.6。
推荐 Skills