baidu qianfan agent chat

千帆AI应用对话接口调用技能。用于调用百度千帆平台的对话API进行AI对话交互。支持流式和非流式响应、Function Call工具调用、文件上传等功能。触发场景：(1) 用户需要调用千帆对话API；(2) 用户提到"千帆"、"qianfan"、"百度AI对话"；(3) 需要与千帆应用进行对话交互。

Security Scan

The skill's code, instructions, and requested environment variable line up with its stated purpose (calling Baidu 千帆/Qianfan conversation API); nothing appears to request unrelated credentials or perform unexpected data exfiltration.

Purpose & Capability

Name/description state the skill calls Baidu 千帆 conversation API and the included script posts to https://qianfan.baidubce.com/v2/app/conversation/runs using an API key; required binary (python3) and required env var (QIANFAN_API_KEY) are consistent with this purpose.

Instruction Scope

SKILL.md and scripts/chat.py stay within the claimed scope: they only read QIANFAN_API_KEY, accept user-provided arguments (query, file ids, tools file, etc.), call the Qianfan API, and save a local session_id to a state file in the skill folder. There are no instructions to read unrelated system files or additional environment variables.

Install Mechanism

No install spec (instruction-only) — low install risk. One usability note: the script imports the third-party 'requests' library but the skill metadata does not declare installing it; the runtime environment must already have requests installed or the script will fail. This is an operational omission, not an overt security issue.

Credentials

Only a single credential (QIANFAN_API_KEY) is required, which is appropriate for calling the documented API. The skill does write a small session file under the skill directory (state/session.json) to remember conversation_id — this is expected behavior for multi-turn sessions.

Persistence & Privilege

always:false and no special privileges requested. The skill persists only its own session state file under the skill directory and does not modify other skills or system-wide agent settings.

Assessment

This skill appears to do exactly what it claims: call Baidu 千帆 conversation API using QIANFAN_API_KEY. Before installing, consider: (1) only provide a QIANFAN_API_KEY you trust to this skill — the key is sent to qianfan.baidubce.com; (2) ensure the runtime has python3 and the 'requests' Python package installed (the script imports requests but does not install it); (3) the script will create a local file state/session.json in the skill directory to store conversation_id — if you need no local state, run with --new-session or remove the file; (4) the skill performs network calls to the Baidu endpoint, so run it in an environment where outbound network access is acceptable; (5) if you want extra assurance, inspect or run the included scripts/chat.py manually before enabling autonomous invocation. Overall the components are internally consistent and proportionate to the described purpose.

千帆对话技能

调用百度千帆平台的对话API，与千帆AI应用进行对话交互。

初始化配置

⚠️ 必须设置环境变量 QIANFAN_API_KEY：

export QIANFAN_API_KEY="your-api-key-here"

API Key 可从 千帆平台 获取。

默认配置

配置项	默认值	说明
app_id	e52a2419-4327-48e8-b9dc-9bf037199fc2	应用ID，可在调用时通过 --app-id 覆盖
stream	false	流式返回

---

接口详情

详细的请求/响应参数、错误码、curl 示例等，请参阅 API 参考文档。

---

基本用法

发起对话

使用 scripts/chat.py 脚本调用：

python3 scripts/chat.py --query "你好，请介绍一下自己"

参数说明

参数	说明	默认值
--query	用户提问内容（必选）	-
--app-id	应用ID	e52a2419-4327-48e8-b9dc-9bf037199fc2
--stream	是否流式返回	true
--conversation-id	会话ID（多轮对话时传入）	-
--file-ids	文件ID列表，逗号分隔	-

多轮对话

会话状态自动管理：

• 首次调用不传 conversation_id，API返回后会自动保存
• 后续调用自动使用已保存的 conversation_id
• 使用 --new-session 开始新会话

# 首次对话（自动保存 conversation_id）
python3 scripts/chat.py --query "你好"
# 输出: [conversation_id: xxx-xxx-xxx]

# 后续对话（自动使用已保存的 conversation_id）
python3 scripts/chat.py --query "刚才我们聊了什么"
# 输出: [使用已保存的会话: xxx-xxx-xxx]

# 开始新会话
python3 scripts/chat.py --query "新话题" --new-session

手动指定会话ID：

python3 scripts/chat.py --query "继续" --conversation-id "xxx-xxx-xxx"

非流式响应

python3 scripts/chat.py --query "你好" --stream false

高级功能

Function Call

定义工具并上报结果：

python3 scripts/chat.py --query "今天北京天气" --tools-file tools/weather.json

注意事项

1. 确保 QIANFAN_API_KEY 环境变量已设置
2. 首次对话不需要 conversation_id，后续多轮对话需传入
3. 流式模式下，响应以 data: 开头，以 data: [DONE] 结束