Poe Chat
使用 @gemini/@gpt/@claude 等触发词调用 Poe 模型(含 Gemini/GPT/claude/kimi/Deepseek等主流模型),自动选择 model_id 并说明使用了哪一个,支持文件上传。
MIT-0 · Free to use, modify, and redistribute. No attribution required.
⭐ 2 · 932 · 0 current installs · 0 all-time installs
bylongman@longmans
MIT-0
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
Name and description (call Poe models, select model IDs, support file upload) match the included scripts and SKILL.md. The code calls Poe's model-list endpoint and uses a Poe client library; requesting a POE_API_KEY is expected.
Instruction Scope
Runtime instructions are limited to installing requirements, setting POE_API_KEY (or passing --api-key), listing models, and invoking the client script. The scripts only read local files when explicitly passed via --file and upload them to Poe; they do not attempt to read unrelated system files or other env vars.
Install Mechanism
No install spec in registry; SKILL.md directs pip install -r scripts/requirements.txt (fastapi-poe, requests). This is a standard PyPI install and proportional to the Python scripts provided.
Credentials
Only POE_API_KEY (or CLI --api-key) is used. No unrelated credentials, config paths, or broad environment access are requested.
Persistence & Privilege
always is false and the skill does not modify other skills or system settings. It writes a models.json cache file only when the list_models script is run, which is within expected behavior.
Assessment
This skill appears to do what it says: it lists Poe models and sends messages (and optionally uploaded files) to Poe. Before installing, note: (1) you must provide a POE_API_KEY — do not share that key; (2) pip will install fastapi-poe from PyPI — verify you trust that package and your environment's package sources; (3) any file paths you pass with --file will be read locally and uploaded to Poe, so avoid uploading sensitive files; (4) the scripts contact api.poe.com (no other external endpoints observed). If you need higher assurance, audit the fastapi-poe package and run the scripts in an isolated environment.Like a lobster shell, security has layers — review code before you run it.
Current versionv1.0.0
Download ziplatest
License
MIT-0
Free to use, modify, and redistribute. No attribution required.
SKILL.md
Poe Chat Skill
适用场景
- 用户输入包含
@gemini、@gpt等 @触发词,希望用 Poe 上的具体模型回答问题。 - 需要自动匹配最合适的具体模型名称(例如 gemini-3-flash)。
- 支持把本地文件上传给模型分析。
使用方式
1) 安装依赖(只需一次)
pip install -r scripts/requirements.txt
如系统没有 python 命令,请改用 python3 执行下列命令。
2) 设置 Poe API Key(如未设置)
export POE_API_KEY="your_api_key"
也可以在调用时通过 --api-key 传入(优先生效)。
3) 查看可用模型(独立脚本)
python scripts/list_models.py
默认会在当前目录生成精简版 models.json(仅包含 id),并作为本地缓存(缓存时间 1 小时,基于文件修改时间)。
也可指定输出文件名:
python scripts/list_models.py --out models.json
如需查看完整模型信息(如 description 等),使用 --full:
python scripts/list_models.py --full --out models-full.json
4) 根据模型列表选择模型 ID
从 models.json 中选择要使用的模型 ID:
- 重点字段是
data[].id(这就是model_id) - 可结合
data[].description判断用途和能力
示例(手动选择):
{
"data": [
{"id": "gemini-3-pro", "description": "..."},
{"id": "claude-opus-4.6", "description": "..."}
]
}
当用户输入 @gemini ... 时,选择包含 Gemini 的模型,并遵循以下优先级:
- 默认优先非 Pro(例如优先
gemini-3-flash而不是gemini-3-pro) - 优先最新版本号(例如 2.5 优于 2.0)
- 只有用户明确要求 Pro/Ultra 时才选 Pro/Ultra(例如用户输入
@gemini-pro或明确说“用Pro”)
5) 调用脚本(直接传入模型 ID)
python scripts/poe_client.py \
--message "请解释量子计算" \
--model-id "gemini-3-flash" \
--api-key "your_api_key" \
--file "/path/to/document.pdf"
说明:
--message必填,内容中包含@xxx触发词即可(如 @gemini、@gpt)。--file可选,可重复多次上传多个文件。
行为准则
- 解析触发词:从用户消息中提取第一个
@xxx触发词。 - 模型选择:根据
models.json里的data[].id手动选择最相关的具体模型。 - 模型列表:通过
list_models.py获取模型列表,内存缓存 1 小时(不持久化)。 - API Key:若环境变量
POE_API_KEY未设置,poe_client.py会提示用户输入。 - 文件上传:如提供
--file,poe_client.py使用fastapi-poe上传并附加到请求。 - 响应输出:回答正文前必须标注 具体模型名(例如
Model used: gemini-3-flash)。
输出格式
Model used: <具体模型名>
<模型回答内容>
如存在模型返回的附件,列出附件信息(名称、类型、URL)。
Files
5 totalSelect a file
Select a file to preview.
Comments
Loading comments…
