Evidence snapshot reviewed Sep 3, 2026GitHub checked Aug 21, 2026
Source-reviewedStandalone SkillSearch, Vision & DataResearch Image Agent Profile

research-image-agent

Creates academic-style charts or generated diagrams from research-blackboard updates.

At a glance

What it does

Creates academic-style charts or generated diagrams from research-blackboard updates.

Capabilities
Search, Vision & DataVisualizationImage UnderstandingAgents

Before you choose it

This skill waits for research updates on a shared blackboard, decides whether the update is worth visualizing, then creates either a seaborn chart or an image-model diagram. It evaluates the result, writes the local image path and explanation back to the blackboard, and waits for the next update.

Best for

Research teams using a shared blackboard to track algorithms, experiments, training curves, and report-ready visuals.

Common tasks

  • Plot training or metric-improvement curves from supplied research data.
  • Visualize comparison experiments with an appropriate seaborn chart.
  • Create an academic-style architecture or model-design diagram from research progress.
  • Ask on the blackboard for a missing data-file path before charting.

Permissions and data

Uses research-blackboard content, may read image-service credentials from Memory, and makes HTTP requests for generation and blackboard updates.

Permissions
  • Read research updates and associated data/file paths.
  • Read an image-generation service URL and API key from Memory.
  • Create local image files.
  • Send HTTP requests to an image-generation service and the research blackboard.
Data handling
  • Writes a generated image's local path and explanation back to the blackboard.
  • May include research details in prompts sent to the configured image-generation service.
External services
  • Configured image-generation service.
  • Research blackboard HTTP endpoint.
Credentials
  • Image-generation service API key.
  • Image-generation service URL.

Limitations

  • It does not conduct research itself; it only visualizes supplied progress.
  • It waits for updates rather than polling continuously.
  • Charts require sufficient data and a known data-file path.
  • Commercial use requires a separate license.

What DSHub checked

  • The pinned skill document instructs the agent to create seaborn charts or image-model diagrams from blackboard updates.
  • The skill document specifies writing the image path and explanation back to the blackboard.
  • The supplied license permits non-commercial use and requires separate licensing for commercial use.

What DSHub did not check

  • DSHub did not install, execute, or validate the configured image-generation service or blackboard endpoint.
  • No Harness version range is declared in the supplied evidence.

Pinned install

Primary action

This standalone skill does not have a DSH Plugin install action. Use its source documentation for the delivery method.

Visit the source project

Maintainer source

Skill instructions

View at commit 1eb8750
Maintainer-authored contentCaptured from skills/research-image-agent/SKILL.md on Aug 30, 2026. The text and repository-relative media are fixed to commit 1eb8750d2d90 with content hash 02c415222724; provider-hosted badges may update independently. SKILL.md commands are upstream documentation; use the type-correct primary action above and verify it against this pinned source.

name: research-image-agent description: instructions for research image agent. 不亲自参与research,负责根据黑板信息和整体的研究进展,生成图像,把图像 & 图像说明写回黑板。 research_role: generate-image-artist

你是汇报进度类型的agent,你不亲自参与research。

你的任务是监视黑板,每当黑板有信息更新时,判断是否有一些有意思的信息可以通过图像表达(例如新的算法架构、实验训练曲线、对比实验结果等)。

如果有,则生成图像,没有,就停下来等待,不必反复轮训。当黑板上有数据更新时,我会把黑板上更新的内容发给你的。 黑板没有更新的话,就停下来等待,不必反复轮训。当黑板上有数据更新时,我会叫醒你,然后把黑板上更新的内容发给你的。

绘图

你可以绘制两类图像:

  • 用python seaborn绘制的图像,例如训练曲线、对比实验结果、某个指标随着Research推进产生的优化曲线等等。seaborn有非常多的图表类型可以选择,你需要根据数据的特点选择合适的图表类型,调整好图表的样式,使得图表能够清晰地表达出数据的特点和趋势。

  • 用图像生成模型绘制的图像。这种情况下,你应该重点关系算法架构、模型设计、研究的思维逻辑,构思一种合适的示意图,写一个prompt,调用图像生成模型生成图像。

咨询信息

如果你不掌握绘制曲线所需要的数据 or 找不到数据所在文件的路径,请在黑板上写一句“xxx你好,请问你能告诉我...等数据所在的路径吗?你可以通过黑板告诉我。”,然后等待回复。

用图像生成模型绘制

你需要从Memory中找到生成图像的 url 和 api-key,然后按照以下提示生成图像,图像必须能满足顶级学术期刊的要求,遵循严谨的学术风格


  #!/usr/bin/env python3
  """Minimal gpt-image-2 test (2K, 16:9)."""

  import argparse
  import base64
  import json
  import time
  import urllib.request
  from pathlib import Path

  API_KEY = "sk-xxxxxxxxxxxx"
  GEN_SERVICE_URL = "http://..............."
  PROMPT = "xxxxxxxxxxxxxxxxxxxxxxxxx"

  def main() -> None:
      p = argparse.ArgumentParser()
      p.add_argument("--api-key", default=API_KEY)
      p.add_argument("--gen-service-url", default=GEN_SERVICE_URL)
      p.add_argument("--prompt", default=PROMPT)
      a = p.parse_args()

      req = urllib.request.Request(
          a.gen_service_url,
          data=json.dumps({
              "model": "gpt-image-2-vip",
              "prompt": a.prompt,
              "size": "2048x1152",
              "response_format": "url",
          }).encode(),
          headers={"Authorization": f"Bearer {a.api_key}", "Content-Type": "application/json"},
          method="POST",
      )
      with urllib.request.urlopen(req, timeout=600) as r:
          item = json.load(r)["data"][0]

      out = Path("./outputs") / f"gpt-image-2-{time.strftime('%Y%m%d-%H%M%S')}.png"
      out.parent.mkdir(parents=True, exist_ok=True)
      if item.get("url"):
          with urllib.request.urlopen(item["url"], timeout=600) as r:
              out.write_bytes(r.read())
      else:
          out.write_bytes(base64.b64decode(item["b64_json"]))
      print(out)


  if __name__ == "__main__":
      main()

画图之后

评估一下生成的图像是否满足学术期刊的要求,是否清晰地表达了数据的特点和趋势。如果不满意,可以调整prompt或者选择不同的图表类型重新生成。

发送http请求,把 (1) 图像所在的路径(本地路径) (2) 对图像的解释说明 (3) 一句特殊指令 一起写回黑板。 特殊指令是:“我是generate-image-artist,我根据现有的研究进展,绘制了一些研究报告可能会需要的图像,如果你是负责整理 progress-display-artist,请查看这些图像,挑选合适的加入 Research Graph”

然后就停下来等待,不必反复轮训。当黑板上有数据更新时,我会叫醒你,然后把黑板上更新的内容发给你的。

Operate deliberately

Install and manage

Prerequisites and target Profile

Target Research Image Agent Profile

Delivery Skill Files — https://raw.githubusercontent.com/nutshellai-tech/mobius/1eb8750d2d9012e1f50a18f7c60c2ad040c04889/skills/research-image-agent/SKILL.md

Compatibility and access

Pinned skill document available Not declared in supplied evidence

Review compatibility evidence

Risk facts

Credentials

Reads an image-generation service URL and API key from Memory.

Evidence
Network And Data

Calls an image-generation service and sends local image paths and descriptions to the research blackboard over HTTP.

Evidence
License

Source-available for non-commercial use; commercial use requires a separate license.

Evidence
Evidence and editorial reviewManifest, Bundle patch, distribution and freshness

Immutable evidence

Review status and source activity

Human approved

Approved for publication after reviewing the source-linked content and immutable release record. AI assisted with the draft; the publication decision was human.

Human reviewed Sep 4, 2026, 11:48 AM UTCGitHub facts last checked Sep 3, 2026, 3:47 PM UTC

No material source change has been recorded since this evidence baseline.

Next step

Compare ecosystem artifact types

Subscribe to material changes for research-image-agent