← 返回 Skills 市场
allen-jmc

JMCAI Comfypet

作者 JMCAI · GitHub ↗ · v1.2.5 · MIT-0
win32linuxdarwin ✓ 安全检测通过
218
总下载
0
收藏
1
当前安装
8
版本数
在 OpenClaw 中安装
/install comfypet-jmcai-skill
功能描述
调用 JMCAI Comfypet 桌面应用中已经配置好的 ComfyUI workflow。本技能必须配合 JMCAI 桌面端运行。用于查询可用 workflow、读取暴露参数、提交运行并轮询结果。适用于用户要求生成图片、生成视频,或执行带图片、遮罩、音频、视频、文件输入资产的 workflow 场景;不适用于...
使用说明 (SKILL.md)

JMCAI Comfypet Skill

[!IMPORTANT] Dependency Requirement: This skill MUST be used with the JMCAI Comfypet Desktop Application. Download the desktop app here: https://github.com/allen-Jmc/comfypet-jmcai-Dist

先运行:

python jmcai_skill.py doctor
python jmcai_skill.py registry --agent

然后按以下顺序执行:

  1. 根据 summarytagsinput_modalitiesoutput_modalitiesexample_prompts 选择最合适的 workflow。
  2. 只填写 alias schema 中暴露的参数。
  3. 提交运行:
python jmcai_skill.py run --workflow \x3Cworkflow_id> --args '{"prompt_1":"a cinematic scene"}'
  1. 读取 run_id 后,单次查询状态:
python jmcai_skill.py status --run-id \x3Crun_id>
  1. 如需回看,再读取:
python jmcai_skill.py history --workflow \x3Cworkflow_id> --limit 5

Parameter Rules

  • 缺少 required: true 的参数时,向用户追问或在上下文足够明确时补齐
  • 只能使用 registry --agent 返回的 alias 参数名
  • image | mask | video | audio | file 类型参数对本机 bridge 仍可直接传本机绝对路径
  • bridge_url 指向局域网另一台桌面端时,skill 会自动把本机资产上传成 upload:\x3Cid>,不用手工改写参数
  • file 类型远程自动上传只覆盖常见 workflow 资产格式,例如图片、音视频、txt/csv/pdf/srt/vtt/ass
  • 不要修改 schema、workflow 或 target 配置

Privacy & Security

  • Secure Upload: This skill only supports uploading workflow asset files needed by exposed schema fields (image, mask, video, audio, file).
  • Access Control: Generic file uploads stay restricted to a narrow whitelist (.txt, .csv, .pdf, subtitles, and common media assets) instead of arbitrary local files.
  • Local First: By default, it communicates with 127.0.0.1. If using a remote bridge, ensure you trust the endpoint.

Failure Handling

  • doctor 失败:提示用户启动或升级 JMCAI 桌面应用
  • run 返回错误:把错误原文反馈给用户,不要伪造成功
  • status 返回 queuedrunning:告诉用户仍在生成,并在下一次独立调用中继续查询
  • status 返回 success:优先返回 outputs 中已经自动下载到当前机器的本地文件路径
  • statushistory 返回 warnings:告诉用户任务已成功,但自动下载输出到当前机器时出现了网络或本地写入问题

需要 bridge 契约细节时,读取 references/bridge.md
需要图片 / 视频 / 资产输入工作流示例时,读取 references/usage.md

安全使用建议
This skill appears to do what it claims: it talks to a JMCAI Comfypet Workflow Bridge (default localhost), lists workflows, submits runs, and uploads/downloads media assets. Before installing: 1) Confirm the config.json bridge_url is local (127.0.0.1) unless you explicitly trust a remote bridge — pointing it at an untrusted host will cause the skill to upload your local files. 2) Note the skill enforces a whitelist of allowed extensions (images, common media, .txt/.pdf/.csv/subtitle formats) — avoid passing sensitive documents even if their suffix is allowed. 3) Run the provided doctor/registry commands to verify the desktop application and bridge version. 4) If you have concerns, review the jmcai_skill.py source (included) or test with non-sensitive files first. If you expect the skill to never transmit files over the network, do not change bridge_url to a remote host.
功能分析
Type: OpenClaw Skill Name: comfypet-jmcai-skill Version: 1.2.5 The skill bundle provides a legitimate interface for interacting with the JMCAI Comfypet desktop application to execute ComfyUI workflows. The core logic in `jmcai_skill.py` is well-structured, uses only Python standard libraries, and implements proactive security measures such as a strict file extension whitelist (e.g., .jpg, .mp4, .txt, .pdf) to prevent arbitrary file exfiltration. The documentation in `SKILL.md` and the reference files accurately describe the tool's functionality without any signs of prompt injection or malicious intent.
能力标签
crypto
能力评估
Purpose & Capability
Name/description match the implementation: the package exposes registry/run/status/history/doctor commands that talk to a Workflow Bridge. The declared requirements (python binary, config.json) align with the CLI-mode Python implementation and SKILL.md.
Instruction Scope
SKILL.md instructs the agent to call the included Python CLI (registry/run/status/history/doctor), to pass only aliased parameters, and to provide absolute local paths for asset fields. It does not instruct reading arbitrary system files or secrets. However, it explicitly supports uploading local asset files to a remote bridge when bridge_url is a non-loopback host — this expands scope to network transfer of user files and is called out in the documentation.
Install Mechanism
No install script; this is instruction-only usage of the provided Python script. The code files are included in the skill bundle and invoked directly by python; there is no external download or extraction of third-party code during install.
Credentials
The skill requests no secrets or environment variables. It does require a config.json (defaulting to localhost bridge). The primary risk is file upload: if the user configures a remote bridge_url, the skill will upload asset files (images, video, audio, and a narrow whitelist of other file types). The whitelist is enforced in code, which reduces but does not eliminate the risk of unintentionally uploading sensitive .txt/.pdf files. Users should ensure bridge_url points to a trusted endpoint.
Persistence & Privilege
always is false and the skill does not request persistent elevated privileges or modify other skills. It operates as a CLI invoked by the agent and writes outputs/downloads to the local machine as part of normal operation.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install comfypet-jmcai-skill
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /comfypet-jmcai-skill 触发
  4. 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
v1.2.5
Release 1.2.5
v1.2.4
Release 1.2.4: Improve remote upload reliability, structured timeout errors, and multi-image regression coverage
v1.2.3
Release 1.2.3: Fix hardcoded IP in config example to satisfy ClawHub static analysis
v1.2.2
Release 1.2.2: Implement secure media upload whitelist
v1.2.1
Release 1.2.1
v1.2.0
Release 1.2.0
v1.1.0
Release 1.1.0
v1.0.0
Initial public release
元数据
Slug comfypet-jmcai-skill
版本 1.2.5
许可证 MIT-0
累计安装 1
当前安装数 1
历史版本数 8
常见问题

JMCAI Comfypet 是什么?

调用 JMCAI Comfypet 桌面应用中已经配置好的 ComfyUI workflow。本技能必须配合 JMCAI 桌面端运行。用于查询可用 workflow、读取暴露参数、提交运行并轮询结果。适用于用户要求生成图片、生成视频,或执行带图片、遮罩、音频、视频、文件输入资产的 workflow 场景;不适用于... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 218 次。

如何安装 JMCAI Comfypet?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install comfypet-jmcai-skill」即可一键安装,无需额外配置。

JMCAI Comfypet 是免费的吗?

是的,JMCAI Comfypet 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

JMCAI Comfypet 支持哪些平台?

JMCAI Comfypet 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(win32, linux, darwin)。

谁开发了 JMCAI Comfypet?

由 JMCAI(@allen-jmc)开发并维护,当前版本 v1.2.5。

💬 留言讨论