AGENT SKILL

sandbox-tfjs

Classify and embed images, detect objects, and answer from a passage in a Code node or CodeAct action, with TensorFlow.js running on the host

使用场景图像与多媒体
License 许可
AGPL-3.0
最近更新
2026-11-15
Stars
★ 635
Forks
91

01 前言 Front Matter

name
sandbox-tfjs
description
Classify and embed images, detect objects, and answer from a passage in a Code node or CodeAct action, with TensorFlow.js running on the host

02 资源文件 7

03 SKILL.md 正文

TensorFlow.js models in the sandbox

Specifier: @nodetool-ai/sandbox-tfjs. Import it at the top of the body.

The models are tens of megabytes of weights, fetched over the network and kept between calls. The guest has neither the heap for them nor a way to keep anything alive past a run, so this pack is a host module: the import resolves to a generated facade over NodeTool's own implementation, and each model loads once per process.

These models are small, local, and free — that is the whole reason to reach for them. For anything a description or a caption would answer better, use a vision model through nodetool.media instead.

classify — ImageNet labels

import { classify } from "@nodetool-ai/sandbox-tfjs";

const labels = await classify(await workspace.readBytes("photo.jpg"), { topK: 3 });
return { best: labels[0].className, probability: labels[0].probability };

MobileNet v2 over the 1000 ImageNet classes, most likely first. topK defaults to 5.

embed — a feature vector for an image

import { embed } from "@nodetool-ai/sandbox-tfjs";

const vector = await embed(bytes);   // 1280 numbers

The MobileNet penultimate layer. Compare two vectors by cosine similarity for near-duplicate detection or nearest-neighbour search.

detect — objects with boxes

import { detect } from "@nodetool-ai/sandbox-tfjs";

const objects = await detect(bytes, { maxBoxes: 10, minScore: 0.6 });
// [{ className: "person", score: 0.91, bbox: { x, y, width, height } }, …]

COCO-SSD over the 80 COCO classes. The box is in image pixels. maxBoxes defaults to 20 and minScore to 0.5.

answer — extractive question answering

import { answer } from "@nodetool-ai/sandbox-tfjs";

const spans = await answer("When was it founded?", inputs.article);
return { answer: spans[0]?.text ?? null };

BERT QnA quotes the passage — it never writes prose. Each span carries a score and its startIndex/endIndex in the passage.

Gotchas

  • Every export is async, and the first call to each model waits for its weights to download.
  • Images go in as encoded bytes (PNG/JPEG/WebP), decoded through the sandbox's own image backend — so the same pixel caps image.decode enforces apply, and 10 MB is the input limit.
  • The models see pixels, nothing else. classify on a screenshot of text answers with an ImageNet class, not with the text; that is @nodetool-ai/sandbox-ocr.

04 同仓库更多 Skills

nodetool-ai
nodetool-ai/nodetool

humanizer

Rewrite AI-sounding text so it reads like the writer without changing what it says. Use when editing or reviewing prose for AI tells: not-X-but-Y contrasts, one-line closers, staged openers, forced triads, dashes everywhere, inflated claims, sales language, stock AI words, bold labels, or filler.

★ 635
nodetool-ai
nodetool-ai/nodetool

nodetool

Use the NodeTool MCP server to build, validate, run, and debug visual AI workflows, and to work with assets, collections, and jobs. Use when the user mentions NodeTool or asks for a workflow graph, a node, or a media generation run on their NodeTool server.

开发工程师Skill与智能体开发
★ 635
nodetool-ai
nodetool-ai/nodetool

wan-2-6-prompting

Prompt Alibaba's Wan 2.6 video — text-to-video with a global style line plus timing-bracketed shots, and image-to-video that describes only what changes over time. Covers the shot-mode switch per provider, the prompt budget, negative prompts, motion intensity wording, audio input and prompt expansion. Use whenever the model id names a Wan 2.6 video route (alibaba/wan-2.6/text-to-video, alibaba/wan-2.6/image-to-video, wan-video/wan-2.6-t2v, wan-video/wan-2.6-i2v, wan-video/wan2.6-i2v-flash) on generate_video, animate_image or a TextToVideo node. Not for Wan 2.5 or 2.7, or the Wan 2.6 image routes.

图像与多媒体
★ 635
nodetool-ai
nodetool-ai/nodetool

video-workflow

Build a reusable NodeTool video workflow for batches or repeated inputs. For a single finished piece, use a storyboard skill.

图像与多媒体Skill与智能体开发
★ 635
nodetool-ai
nodetool-ai/nodetool

video-clone

Analyze a reference video and recreate its shot structure with a new product or cast in NodeTool.

图像与多媒体
★ 635
nodetool-ai
nodetool-ai/nodetool

video-audio-continuity

Keep sound continuous across a multi-scene piece cut from generated video — why one clip per scene hard-cuts the audio at every boundary, when to write all the scenes into a single generation instead, the sound-design vocabulary that survives a provider's copyright filter, and how to check the length, the aspect ratio and the mix of what came back. Use when a piece has more than one scene and the video model writes its own audio (Seedance 2, Veo 3, MiniMax H3, Kling 2.6 and later), or when a cut's sound drops out at the shot changes.

图像与多媒体
★ 635