feat: 图片识别技能
This commit is contained in:
parent
e1fe82a3a5
commit
1b0d157903
83
skills/image-recognition/SKILL.md
Normal file
83
skills/image-recognition/SKILL.md
Normal file
@ -0,0 +1,83 @@
|
||||
---
|
||||
name: image-recognition
|
||||
description: "AI 图像识别工具。当用户提供图片并希望识别、描述、提取文字、分析画面内容或回答图片相关问题时使用。"
|
||||
argument-hint: "需要 prompt(识别提示词)和 image_url(图片 URL)"
|
||||
---
|
||||
|
||||
# Image Recognition Skill
|
||||
|
||||
## 描述
|
||||
|
||||
这是一个 AI 图像识别技能,输入一张图片 URL 或本地图片路径和识别提示词,输出模型对图片的识别、描述或分析结果。
|
||||
|
||||
该技能从数据库读取当前会话的聊天 AI 配置:`chat_base_url`、`chat_api_key` 和 `image_recognition_model`,并调用 OpenAI 兼容的多模态 Chat Completions 接口完成图片理解。
|
||||
|
||||
这个仓库里额外提供了一个可执行脚本 `scripts/image_recognition.py`,方便宿主机器人直接调用。
|
||||
|
||||
## 触发条件
|
||||
|
||||
- 用户发送图片并要求描述图片内容。
|
||||
- 用户说「识别这张图」「看看图片里有什么」「分析一下这张图片」。
|
||||
- 用户要求从图片中提取文字、物体、场景、人物动作、票据内容等信息。
|
||||
- 用户基于图片提问,例如「这是什么」「图片里的文字是什么」「这张图哪里不对」。
|
||||
|
||||
## 参数说明(JSON Schema)
|
||||
|
||||
调用脚本时,需要通过 shell 风格参数传入,参数结构如下:
|
||||
|
||||
```json
|
||||
{
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"prompt": {
|
||||
"type": "string",
|
||||
"description": "图像识别提示词,用户想从图片中得到什么信息。"
|
||||
},
|
||||
"image_url": {
|
||||
"type": "string",
|
||||
"description": "图片的 URL 地址、本地图片路径,或 file:// 本地图片地址。"
|
||||
}
|
||||
},
|
||||
"required": ["prompt", "image_url"],
|
||||
"additionalProperties": false
|
||||
}
|
||||
```
|
||||
|
||||
对应的命令行参数为:
|
||||
|
||||
- `--prompt <识别提示词>` 必填
|
||||
- `--image_url <图片 URL 或本地图片路径>` 必填
|
||||
|
||||
## 依赖安装
|
||||
|
||||
- 脚本首次运行时会自动创建虚拟环境并安装依赖,无需手动执行。
|
||||
- 如需手动重新安装,可执行:`python3 scripts/bootstrap.py`
|
||||
|
||||
## 执行步骤
|
||||
|
||||
1. 当用户提供图片并要求识别、描述、分析或提取信息时触发该技能。
|
||||
2. 从用户输入中提取 `prompt` 和 `image_url`,不要改写用户真正想问图片的问题。
|
||||
3. 在仓库根目录执行脚本,例如:
|
||||
|
||||
```bash
|
||||
python3 scripts/image_recognition.py --prompt '请描述这张图片' --image_url 'https://example.com/image.jpg'
|
||||
```
|
||||
|
||||
本地图片示例:
|
||||
|
||||
```bash
|
||||
python3 scripts/image_recognition.py --prompt '请提取图片里的文字' --image_url '/tmp/example.jpg'
|
||||
```
|
||||
|
||||
4. 脚本将图片和提示词一起发送给 OpenAI 兼容的多模态模型。远程图片会直接传 URL;本地图片会转成 `data:image/...;base64,...` 后传入 `image_url.url`。
|
||||
5. 成功时,脚本输出图像识别结果文本,宿主机器人可直接作为消息回复给用户。
|
||||
|
||||
## 校验规则
|
||||
|
||||
- `prompt` 不能为空。
|
||||
- `image_url` 不能为空,支持 `http://`、`https://`、`file://` 和本地图片路径。
|
||||
|
||||
## 回复要求
|
||||
|
||||
- 成功时,脚本输出图片识别结果。
|
||||
- 失败时,返回脚本输出的具体错误信息。
|
||||
110
skills/image-recognition/scripts/bootstrap.py
Normal file
110
skills/image-recognition/scripts/bootstrap.py
Normal file
@ -0,0 +1,110 @@
|
||||
#!/usr/bin/env python3
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import hashlib
|
||||
import subprocess
|
||||
import sys
|
||||
import traceback
|
||||
from pathlib import Path
|
||||
|
||||
sys.stderr = sys.stdout
|
||||
|
||||
|
||||
def _skill_root_from(script_dir: Path) -> Path:
|
||||
return script_dir.parent
|
||||
|
||||
|
||||
def _venv_dir(script_dir: Path) -> Path:
|
||||
return _skill_root_from(script_dir) / ".venv"
|
||||
|
||||
|
||||
def _venv_python(venv_dir: Path) -> Path:
|
||||
if sys.platform == "win32":
|
||||
return venv_dir / "Scripts" / "python.exe"
|
||||
return venv_dir / "bin" / "python"
|
||||
|
||||
|
||||
def _stamp_file(venv_dir: Path) -> Path:
|
||||
return venv_dir / ".req_hash"
|
||||
|
||||
|
||||
def _file_hash(path: Path) -> str:
|
||||
return hashlib.sha256(path.read_bytes()).hexdigest()
|
||||
|
||||
|
||||
def _deps_up_to_date(requirements_file: Path, venv_dir: Path) -> bool:
|
||||
stamp = _stamp_file(venv_dir)
|
||||
if not stamp.is_file():
|
||||
return False
|
||||
return stamp.read_text().strip() == _file_hash(requirements_file)
|
||||
|
||||
|
||||
def _write_stamp(requirements_file: Path, venv_dir: Path) -> None:
|
||||
_stamp_file(venv_dir).write_text(_file_hash(requirements_file))
|
||||
|
||||
|
||||
def _ensure_venv(venv_dir: Path, venv_python: Path) -> int:
|
||||
if venv_python.is_file():
|
||||
return 0
|
||||
|
||||
sys.stdout.write(f"未检测到技能虚拟环境,正在创建: {venv_dir}\n")
|
||||
import shutil
|
||||
py = sys.executable or next(
|
||||
(shutil.which(candidate) for candidate in ("python3", "python") if shutil.which(candidate)), None
|
||||
)
|
||||
if not py:
|
||||
raise RuntimeError("无法找到 Python 解释器路径")
|
||||
|
||||
try:
|
||||
subprocess.run([py, "-m", "venv", str(venv_dir)], check=True, stdout=sys.stdout, stderr=sys.stdout)
|
||||
except subprocess.CalledProcessError as exc:
|
||||
sys.stdout.write(f"创建虚拟环境失败,退出码: {exc.returncode}\n")
|
||||
return exc.returncode or 1
|
||||
|
||||
return 0
|
||||
|
||||
|
||||
def main() -> int:
|
||||
script_dir = Path(__file__).resolve().parent
|
||||
requirements_file = script_dir / "requirements.txt"
|
||||
venv_dir = _venv_dir(script_dir)
|
||||
venv_python = _venv_python(venv_dir)
|
||||
|
||||
if not requirements_file.is_file():
|
||||
sys.stdout.write(f"未找到依赖文件: {requirements_file}\n")
|
||||
return 1
|
||||
|
||||
ensure_result = _ensure_venv(venv_dir, venv_python)
|
||||
if ensure_result != 0:
|
||||
return ensure_result
|
||||
|
||||
if _deps_up_to_date(requirements_file, venv_dir):
|
||||
sys.stdout.write("依赖已是最新,跳过安装\n")
|
||||
return 0
|
||||
|
||||
try:
|
||||
subprocess.run([str(venv_python), "-m", "pip", "install", "--upgrade", "pip"], check=True, stdout=sys.stdout, stderr=sys.stdout)
|
||||
except subprocess.CalledProcessError as exc:
|
||||
sys.stdout.write(f"升级 pip 失败,退出码: {exc.returncode}\n")
|
||||
return exc.returncode or 1
|
||||
|
||||
try:
|
||||
subprocess.run([str(venv_python), "-m", "pip", "install", "-r", str(requirements_file)], check=True, stdout=sys.stdout, stderr=sys.stdout)
|
||||
except subprocess.CalledProcessError as exc:
|
||||
sys.stdout.write(f"安装依赖失败,退出码: {exc.returncode}\n")
|
||||
return exc.returncode or 1
|
||||
|
||||
_write_stamp(requirements_file, venv_dir)
|
||||
sys.stdout.write(f"依赖安装完成,当前技能虚拟环境: {venv_dir}\n")
|
||||
return 0
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
try:
|
||||
raise SystemExit(main())
|
||||
except SystemExit:
|
||||
raise
|
||||
except Exception:
|
||||
traceback.print_exc(file=sys.stdout)
|
||||
raise SystemExit(1)
|
||||
331
skills/image-recognition/scripts/image_recognition.py
Normal file
331
skills/image-recognition/scripts/image_recognition.py
Normal file
@ -0,0 +1,331 @@
|
||||
#!/usr/bin/env python3
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import argparse
|
||||
import base64
|
||||
import mimetypes
|
||||
import os
|
||||
import re
|
||||
import subprocess
|
||||
import sys
|
||||
import traceback
|
||||
from pathlib import Path
|
||||
from urllib.parse import urlparse, unquote
|
||||
|
||||
sys.stderr = sys.stdout
|
||||
|
||||
|
||||
def _skill_root() -> Path:
|
||||
return Path(__file__).resolve().parent.parent
|
||||
|
||||
|
||||
def _skill_venv_python() -> Path:
|
||||
venv_dir = _skill_root() / ".venv"
|
||||
if sys.platform == "win32":
|
||||
return venv_dir / "Scripts" / "python.exe"
|
||||
return venv_dir / "bin" / "python"
|
||||
|
||||
|
||||
def _get_python_executable() -> str:
|
||||
if sys.executable:
|
||||
return sys.executable
|
||||
import shutil
|
||||
for candidate in ("python3", "python"):
|
||||
found = shutil.which(candidate)
|
||||
if found:
|
||||
return found
|
||||
raise RuntimeError("无法找到 Python 解释器路径")
|
||||
|
||||
|
||||
def _run_bootstrap() -> None:
|
||||
bootstrap = Path(__file__).resolve().parent / "bootstrap.py"
|
||||
result = subprocess.run([_get_python_executable(), str(bootstrap)])
|
||||
if result.returncode != 0:
|
||||
raise SystemExit(result.returncode)
|
||||
|
||||
|
||||
def _ensure_skill_venv_python() -> None:
|
||||
venv_python = _skill_venv_python()
|
||||
if not venv_python.is_file():
|
||||
_run_bootstrap()
|
||||
venv_python = _skill_venv_python()
|
||||
if not venv_python.is_file():
|
||||
sys.stdout.write("bootstrap 后仍未找到虚拟环境\n")
|
||||
raise SystemExit(1)
|
||||
|
||||
venv_dir = _skill_root() / ".venv"
|
||||
if Path(sys.prefix) == venv_dir.resolve():
|
||||
return
|
||||
|
||||
os.execv(str(venv_python), [str(venv_python), str(Path(__file__).resolve()), *sys.argv[1:]])
|
||||
|
||||
|
||||
_ensure_skill_venv_python()
|
||||
|
||||
try:
|
||||
import pymysql # type: ignore # noqa: E402
|
||||
from openai import OpenAI # type: ignore # noqa: E402
|
||||
except ModuleNotFoundError:
|
||||
_run_bootstrap()
|
||||
_py = _get_python_executable()
|
||||
os.execv(_py, [_py, str(Path(__file__).resolve()), *sys.argv[1:]])
|
||||
|
||||
|
||||
def _mysql_connect():
|
||||
host = os.environ.get("MYSQL_HOST", "127.0.0.1")
|
||||
port = int(os.environ.get("MYSQL_PORT", "3306"))
|
||||
user = os.environ.get("MYSQL_USER", "root")
|
||||
password = os.environ.get("MYSQL_PASSWORD", "")
|
||||
database = os.environ.get("ROBOT_CODE", "")
|
||||
if not database:
|
||||
raise RuntimeError("环境变量 ROBOT_CODE 未配置")
|
||||
|
||||
return pymysql.connect(
|
||||
host=host,
|
||||
port=port,
|
||||
user=user,
|
||||
password=password,
|
||||
database=database,
|
||||
charset="utf8mb4",
|
||||
connect_timeout=10,
|
||||
read_timeout=30,
|
||||
)
|
||||
|
||||
|
||||
def _query_one(conn, sql: str, params: tuple = ()) -> dict | None:
|
||||
cur = conn.cursor()
|
||||
cur.execute(sql, params)
|
||||
columns = [desc[0] for desc in cur.description] if cur.description else []
|
||||
row = cur.fetchone()
|
||||
cur.close()
|
||||
if row is None:
|
||||
return None
|
||||
return dict(zip(columns, row))
|
||||
|
||||
|
||||
def _clean_text(value: object) -> str:
|
||||
if isinstance(value, (bytes, bytearray)):
|
||||
value = value.decode("utf-8")
|
||||
if isinstance(value, str):
|
||||
return value.strip()
|
||||
return ""
|
||||
|
||||
|
||||
def _extract_model(record: dict | None) -> str:
|
||||
if record:
|
||||
model = _clean_text(record.get("image_recognition_model"))
|
||||
if model:
|
||||
return model
|
||||
return ""
|
||||
|
||||
|
||||
def _normalize_ai_base_url(base_url: str) -> str:
|
||||
normalized = base_url.rstrip("/")
|
||||
if normalized and not re.search(r"/v\d+$", normalized):
|
||||
normalized += "/v1"
|
||||
return normalized
|
||||
|
||||
|
||||
def load_image_recognition_config(conn, from_wx_id: str) -> dict:
|
||||
global_fields = "chat_base_url, chat_api_key, image_recognition_model"
|
||||
global_record = _query_one(conn, f"SELECT {global_fields} FROM global_settings LIMIT 1")
|
||||
|
||||
config = {"base_url": "", "api_key": "", "model": ""}
|
||||
if global_record:
|
||||
base_url = _clean_text(global_record.get("chat_base_url"))
|
||||
api_key = _clean_text(global_record.get("chat_api_key"))
|
||||
if base_url:
|
||||
config["base_url"] = base_url
|
||||
if api_key:
|
||||
config["api_key"] = api_key
|
||||
model = _extract_model(global_record)
|
||||
if model:
|
||||
config["model"] = model
|
||||
|
||||
if from_wx_id.endswith("@chatroom"):
|
||||
override_fields = "chat_base_url, chat_api_key, image_recognition_model"
|
||||
override = _query_one(
|
||||
conn,
|
||||
f"SELECT {override_fields} FROM chat_room_settings WHERE chat_room_id = %s LIMIT 1",
|
||||
(from_wx_id,),
|
||||
)
|
||||
else:
|
||||
override_fields = "chat_base_url, chat_api_key, image_recognition_model"
|
||||
override = _query_one(
|
||||
conn,
|
||||
f"SELECT {override_fields} FROM friend_settings WHERE wechat_id = %s LIMIT 1",
|
||||
(from_wx_id,),
|
||||
)
|
||||
|
||||
if override:
|
||||
base_url = _clean_text(override.get("chat_base_url"))
|
||||
api_key = _clean_text(override.get("chat_api_key"))
|
||||
if base_url:
|
||||
config["base_url"] = base_url
|
||||
if api_key:
|
||||
config["api_key"] = api_key
|
||||
model = _extract_model(override)
|
||||
if model:
|
||||
config["model"] = model
|
||||
|
||||
config["base_url"] = _normalize_ai_base_url(config["base_url"])
|
||||
return config
|
||||
|
||||
|
||||
def _local_image_path(value: str) -> Path:
|
||||
parsed = urlparse(value)
|
||||
if parsed.scheme == "file":
|
||||
if parsed.netloc not in ("", "localhost"):
|
||||
raise ValueError("不支持非本机 file URL")
|
||||
return Path(unquote(parsed.path)).expanduser()
|
||||
|
||||
path = Path(value).expanduser()
|
||||
if not path.is_absolute():
|
||||
path = Path.cwd() / path
|
||||
return path
|
||||
|
||||
|
||||
def _local_image_to_data_url(value: str) -> str:
|
||||
path = _local_image_path(value)
|
||||
if not path.is_file():
|
||||
raise ValueError(f"本地图片不存在: {path}")
|
||||
|
||||
mime_type, _ = mimetypes.guess_type(str(path))
|
||||
if not mime_type or not mime_type.startswith("image/"):
|
||||
raise ValueError(f"无法识别本地图片类型: {path}")
|
||||
|
||||
encoded = base64.b64encode(path.read_bytes()).decode("ascii")
|
||||
return f"data:{mime_type};base64,{encoded}"
|
||||
|
||||
|
||||
def _resolve_image_url(value: str) -> str:
|
||||
parsed = urlparse(value)
|
||||
if parsed.scheme in {"http", "https"} and parsed.netloc:
|
||||
return value
|
||||
if parsed.scheme == "data" and value.startswith("data:image/"):
|
||||
return value
|
||||
if parsed.scheme and parsed.scheme != "file":
|
||||
raise ValueError(f"不支持的图片地址协议: {parsed.scheme}")
|
||||
return _local_image_to_data_url(value)
|
||||
|
||||
|
||||
def _extract_response_text(response) -> str:
|
||||
if not response.choices:
|
||||
return ""
|
||||
|
||||
content = response.choices[0].message.content
|
||||
if isinstance(content, str):
|
||||
return content.strip()
|
||||
if isinstance(content, list):
|
||||
texts: list[str] = []
|
||||
for item in content:
|
||||
text = getattr(item, "text", None)
|
||||
if isinstance(text, str) and text.strip():
|
||||
texts.append(text.strip())
|
||||
elif isinstance(item, dict) and isinstance(item.get("text"), str) and item["text"].strip():
|
||||
texts.append(item["text"].strip())
|
||||
return "\n".join(texts)
|
||||
return ""
|
||||
|
||||
|
||||
def recognize_image(prompt: str, image_url: str, config: dict) -> str:
|
||||
api_key = config.get("api_key", "")
|
||||
base_url = config.get("base_url", "")
|
||||
model = config.get("model", "")
|
||||
if not api_key or not base_url or not model:
|
||||
raise RuntimeError("AI图片识别未配置,请联系管理员进行配置")
|
||||
|
||||
resolved_image_url = _resolve_image_url(image_url)
|
||||
client = OpenAI(api_key=api_key, base_url=base_url)
|
||||
response = client.chat.completions.create(
|
||||
model=model,
|
||||
messages=[
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{"type": "image_url", "image_url": {"url": resolved_image_url}},
|
||||
{"type": "text", "text": prompt},
|
||||
],
|
||||
}
|
||||
],
|
||||
stream=False,
|
||||
)
|
||||
content = _extract_response_text(response)
|
||||
if not content:
|
||||
raise RuntimeError("图片识别失败,返回了空内容")
|
||||
return content
|
||||
|
||||
|
||||
def _parse_cli_params(argv: list[str]) -> dict:
|
||||
parser = argparse.ArgumentParser(add_help=False)
|
||||
parser.add_argument("--prompt", default="")
|
||||
parser.add_argument("--image_url", default="")
|
||||
|
||||
namespace, unknown = parser.parse_known_args(argv)
|
||||
if unknown:
|
||||
raise ValueError(f"存在不支持的参数: {' '.join(unknown)}")
|
||||
|
||||
return {"prompt": namespace.prompt, "image_url": namespace.image_url}
|
||||
|
||||
|
||||
def main() -> int:
|
||||
if len(sys.argv) < 2:
|
||||
sys.stdout.write("缺少输入参数\n")
|
||||
return 1
|
||||
|
||||
try:
|
||||
params = _parse_cli_params(sys.argv[1:])
|
||||
except ValueError as exc:
|
||||
sys.stdout.write(f"参数格式错误: {exc}\n")
|
||||
return 1
|
||||
|
||||
prompt = params.get("prompt", "").strip()
|
||||
image_url = params.get("image_url", "").strip()
|
||||
if not prompt:
|
||||
sys.stdout.write("缺少图像识别提示词\n")
|
||||
return 1
|
||||
if not image_url:
|
||||
sys.stdout.write("缺少图片 URL\n")
|
||||
return 1
|
||||
|
||||
from_wx_id = os.environ.get("ROBOT_FROM_WX_ID", "").strip()
|
||||
if not from_wx_id:
|
||||
sys.stdout.write("环境变量 ROBOT_FROM_WX_ID 未配置\n")
|
||||
return 1
|
||||
|
||||
try:
|
||||
conn = _mysql_connect()
|
||||
except Exception as exc:
|
||||
sys.stdout.write(f"数据库连接失败: {exc}\n")
|
||||
return 1
|
||||
|
||||
try:
|
||||
config = load_image_recognition_config(conn, from_wx_id)
|
||||
except Exception as exc:
|
||||
sys.stdout.write(f"加载图像识别配置失败: {exc}\n")
|
||||
return 1
|
||||
finally:
|
||||
try:
|
||||
conn.close()
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
try:
|
||||
content = recognize_image(prompt, image_url, config)
|
||||
except Exception as exc:
|
||||
sys.stdout.write(f"图片识别失败: {exc}\n")
|
||||
return 1
|
||||
|
||||
sys.stdout.write(f"{content}\n")
|
||||
return 0
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
try:
|
||||
raise SystemExit(main())
|
||||
except SystemExit:
|
||||
raise
|
||||
except Exception:
|
||||
traceback.print_exc(file=sys.stdout)
|
||||
raise SystemExit(1)
|
||||
3
skills/image-recognition/scripts/requirements.txt
Normal file
3
skills/image-recognition/scripts/requirements.txt
Normal file
@ -0,0 +1,3 @@
|
||||
cryptography
|
||||
openai>=2.34.0
|
||||
pymysql>=1.1,<2
|
||||
@ -17,10 +17,11 @@ argument-hint: "无需参数,直接调用即可"
|
||||
## 执行步骤
|
||||
|
||||
1. 使用 `web-page` 这个网页内容读取、自动化交互和截图技能,访问 `https://quote.eastmoney.com/stockhotmap/` 这个网页。
|
||||
2. 网页上有这个类名(`topzs`)的 div 元素,展示了`上证指数`、`深证成指`、`创业板指`、`科创综指`、`北证50`的涨跌情况。你需要这些信息的时候,就从这个元素里提取出来。
|
||||
3. 网页上有这个类名(`stockmap`)的 div 元素,展示了各个板块的涨跌情况。你需要这些信息的时候,就从这个元素里提取出来。
|
||||
4. 这个网页是 ajax 动态渲染的,访问后等待 5 秒钟再提取内容。
|
||||
5. 如果需要截图,截图前先将带有这个类名(`em_widget em_show`)的元素删除
|
||||
2. 这个网页是 ajax 动态数据由 javascript 动态渲染的,访问后等待 5 秒钟再提取内容。
|
||||
3. 如果需要截图,截图前先将带有这个类名(`popwscps_d`)的元素删除
|
||||
4. 网页上有这个类名(`topzs`)的 div 元素,展示了`上证指数`、`深证成指`、`创业板指`、`科创综指`、`北证50`的涨跌情况。优先截图这个元素,然后使用图像识别技能识别具体内容。
|
||||
5. 网页上有这个 id(`stockmap`)的 div 元素,展示了各个板块的涨跌情况。如果用户想看板块涨跌情况,直接截图并使用发送图片技能发送图片。
|
||||
6. 如果没有获取到具体涨跌情况,则截图 id(`stockmap`) 的 div 元素并使用发送图片技能发送图片。
|
||||
|
||||
## 回复要求
|
||||
|
||||
|
||||
Loading…
Reference in New Issue
Block a user