feat: add agnes image provider (agnes-image-2.1-flash)

README.md

@@ -124,28 +124,47 @@ python3 toolkit/cli.py themes
 **Claude Code**：
 
 ```bash
+# macOS/Linux
 git clone --depth 1 https://github.com/oaker-io/wewrite.git ~/.claude/skills/wewrite
 cd ~/.claude/skills/wewrite && bash install.sh
+
+# Windows (PowerShell)
+git clone --depth 1 https://github.com/oaker-io/wewrite.git "$env:USERPROFILE\.claude\skills\wewrite"
+cd "$env:USERPROFILE\.claude\skills\wewrite"
+powershell -ExecutionPolicy Bypass -File install.ps1
 ```
 
 **OpenClaw**：
 
 ```bash
+# macOS/Linux
 git clone --depth 1 https://github.com/oaker-io/wewrite.git ~/.openclaw/skills/wewrite
 cd ~/.openclaw/skills/wewrite && bash install.sh
+
+# Windows (PowerShell)
+git clone --depth 1 https://github.com/oaker-io/wewrite.git "$env:USERPROFILE\.openclaw\skills\wewrite"
+cd "$env:USERPROFILE\.openclaw\skills\wewrite"
+powershell -ExecutionPolicy Bypass -File install.ps1
 ```
 
 **Codex**（OpenAI Codex CLI）：
 
 ```bash
+# macOS/Linux
 git clone --depth 1 https://github.com/oaker-io/wewrite.git ~/.codex/skills/wewrite
 cd ~/.codex/skills/wewrite && bash install.sh
 python3 scripts/build_codex.py --install   # 装自定义 prompt 到 ~/.codex/prompts/
+
+# Windows (PowerShell)
+git clone --depth 1 https://github.com/oaker-io/wewrite.git "$env:USERPROFILE\.codex\skills\wewrite"
+cd "$env:USERPROFILE\.codex\skills\wewrite"
+powershell -ExecutionPolicy Bypass -File install.ps1
+python scripts/build_codex.py --install   # 装自定义 prompt 到 ~/.codex/prompts/
 ```
 
 之后在 Codex 里用 `/wewrite 写一篇关于 X 的文章` 触发完整流程。Codex 没有 SKILL.md 自动触发机制，所以通过自定义 prompt 承载；源 `SKILL.md` 更新后重跑 `build_codex.py --install` 同步。详见 [`dist/codex/README.md`](dist/codex/README.md)。
 
-`install.sh` 会在 `.venv` 里创建隔离环境并安装依赖，自动绕过 macOS Homebrew Python 的 PEP 668 限制。skill 运行时会自动使用该 venv，无需手动 `activate`。
+`install.sh`（macOS/Linux）或 `install.ps1`（Windows）会在 `.venv` 里创建隔离环境并安装依赖，自动绕过 macOS Homebrew Python 的 PEP 668 限制。skill 运行时会自动使用该 venv，无需手动 `activate`。
 
 > 若你的环境没有 PEP 668 限制，也可以照旧直接 `pip install -r requirements.txt`。
 

SKILL.md

@@ -43,7 +43,11 @@ allowed-tools:
 
 **读取/检查约定**：本文档中 `读取: <路径>` / `检查: <路径>` = **用你环境的文件读取工具真实打开该文件、读完其全部内容，然后再继续本步**。这不是描述性注释——未读取前不得执行依赖该文件的步骤；不同 harness 的文件读取工具名不同，按你环境的对应工具执行。
 
-**Python 解释器约定**：本文档所有 `python3` 命令优先解析为 `{skill_dir}/.venv/bin/python3`（若该文件存在），否则回退系统 `python3`。venv 由 `install.sh` 创建，用于隔离依赖并绕过 macOS Homebrew Python 的 PEP 668 限制。
+**Python 解释器约定**：本文档所有 `python3` 命令优先解析为：
+- **macOS/Linux**：`{skill_dir}/.venv/bin/python3`（若该文件存在）
+- **Windows**：`{skill_dir}\.venv\Scripts\python.exe`（若该文件存在）
+
+否则回退系统 `python3`（或 Windows 上的 `python`）。venv 由 `install.sh`（macOS/Linux）或 `install.ps1`（Windows）创建，用于隔离依赖并绕过 macOS Homebrew Python 的 PEP 668 限制。
 
 **Onboard 例外**：Onboard 是交互式的（需要问用户问题），不受"全自动"约束。Onboard 完成后回到全自动管道。
 
@@ -70,14 +74,19 @@ allowed-tools:
 
 ```bash
 # 优先用 venv 解释器（PEP 668 环境下依赖装在 .venv 里）；后续所有 python3 调用同此规则
+# macOS/Linux:
 PY="{skill_dir}/.venv/bin/python3"; [ -x "$PY" ] || PY="python3"
 "$PY" -c "import markdown, bs4, cssutils, requests, yaml, pygments, PIL" 2>&1
+
+# Windows (PowerShell):
+# $venvPython = "{skill_dir}\.venv\Scripts\python.exe"
+# if (Test-Path $venvPython) { & $venvPython -c "import markdown, bs4, cssutils, requests, yaml, pygments, PIL" } else { python -c "import markdown, bs4, cssutils, requests, yaml, pygments, PIL" }
 ```
 
 | 检查项 | 通过 | 不通过 |
 |--------|------|--------|
 | `config.yaml` 存在 | 静默 | 引导创建，或设 `skip_publish = true` |
-| Python 依赖 | 静默 | 引导执行 `bash {skill_dir}/install.sh`（自动建 .venv 装依赖，解决 macOS PEP 668 报错）；若环境无此限制也可 `pip install -r requirements.txt` |
+| Python 依赖 | 静默 | 引导执行 `bash {skill_dir}/install.sh`（macOS/Linux）或 `powershell -ExecutionPolicy Bypass -File {skill_dir}\install.ps1`（Windows）（自动建 .venv 装依赖，解决 macOS PEP 668 报错）；若环境无此限制也可 `pip install -r requirements.txt` |
 | `wechat.appid` + `secret` | 静默 | 设 `skip_publish = true` |
 | `image.api_key` 或 `image.providers` 至少一项有效 | 静默 | 设 `skip_image_gen = true` |
 | `references/exemplars/index.yaml` | 静默 | 提示："范文库为空。如果你有已发布的文章（markdown），可以说**'导入范文'**建立风格库，写出来的文章会更像你。没有也不影响使用。" |

config.example.yaml

@@ -22,6 +22,7 @@ wechat:
 # │ gemini          │ https://aistudio.google.com/apikey             │ 免费额度较多       │
 # │ openrouter      │ https://openrouter.ai/settings/keys            │ 多模型代理         │
 # │ replicate       │ https://replicate.com/account/api-tokens       │ 开源模型丰富       │
+# │ agnes           │ https://agnes-ai.com/                          │ 新模型,免费额度   │
 # └─────────────────┴────────────────────────────────────────────────┴────────────────────┘
 #
 # 支持两种配置方式：
@@ -65,6 +66,9 @@ image:
 #     - provider: replicate
 #       api_key: "r8_..."
 #       # model: "google/nano-banana-pro"
+#     - provider: agnes
+#       api_key: "your_agnes_key"
+#       # model: "agnes-image-2.1-flash"
 
 # 默认排版主题
 theme: "professional-clean"

install.ps1

@@ -0,0 +1,59 @@
+# WeWrite 依赖安装脚本 (Windows PowerShell 版本)
+# 用法：powershell -ExecutionPolicy Bypass -File install.ps1
+
+$ErrorActionPreference = "Stop"
+Set-Location $PSScriptRoot
+
+$PYTHON = "python"
+if ($env:PYTHON) { $PYTHON = $env:PYTHON }
+
+if (-not (Get-Command $PYTHON -ErrorAction SilentlyContinue)) {
+    Write-Host "找不到 $PYTHON，请先安装 Python 3.11+" -ForegroundColor Red
+    exit 1
+}
+
+$pythonVersion = & $PYTHON --version 2>&1
+if ($LASTEXITCODE -ne 0) {
+    Write-Host "无法获取 Python 版本" -ForegroundColor Red
+    exit 1
+}
+
+$versionMatch = [regex]::Match($pythonVersion, "Python (\d+)\.(\d+)")
+if ($versionMatch.Success) {
+    $major = [int]$versionMatch.Groups[1].Value
+    $minor = [int]$versionMatch.Groups[2].Value
+    if ($major -lt 3 -or ($major -eq 3 -and $minor -lt 11)) {
+        Write-Host "Python 版本过低：$pythonVersion，需要 Python 3.11+" -ForegroundColor Red
+        exit 1
+    }
+}
+
+if (-not (Test-Path ".venv")) {
+    Write-Host "创建虚拟环境 .venv ..." -ForegroundColor Cyan
+    & $PYTHON -m venv .venv
+    if ($LASTEXITCODE -ne 0) {
+        Write-Host "创建虚拟环境失败" -ForegroundColor Red
+        exit 1
+    }
+} else {
+    Write-Host "复用已有的 .venv" -ForegroundColor Cyan
+}
+
+Write-Host "安装依赖到 .venv ..." -ForegroundColor Cyan
+$venvPython = ".venv\Scripts\python.exe"
+
+& $venvPython -m pip install --upgrade pip --quiet
+if ($LASTEXITCODE -ne 0) {
+    Write-Host "升级 pip 失败" -ForegroundColor Red
+    exit 1
+}
+
+& $venvPython -m pip install -r requirements.txt
+if ($LASTEXITCODE -ne 0) {
+    Write-Host "安装依赖失败" -ForegroundColor Red
+    exit 1
+}
+
+Write-Host ""
+Write-Host "完成！依赖已装入 $PWD\.venv" -ForegroundColor Green
+Write-Host "无需手动 activate，skill 运行时会自动使用 .venv\Scripts\python.exe" -ForegroundColor Green

toolkit/image_gen.py

@@ -12,6 +12,7 @@
   - azure_openai — Azure-hosted DALL-E
   - openrouter — multi-model proxy
   - jimeng (ByteDance) — good for Chinese prompts
+  - agnes (Sapiens AI) — agnes-image-2.1-flash, free credits available
   - Custom providers via ImageProvider base class
 
 Usage as CLI:
@@ -74,21 +75,25 @@ def _load_config() -> dict:
         "doubao": "2952x1256", "openai": _DEFAULT, "gemini": _DEFAULT,
         "dashscope": _DEFAULT, "minimax": _DEFAULT, "replicate": _DEFAULT,
         "azure_openai": _DEFAULT, "openrouter": _DEFAULT, "jimeng": _DEFAULT,
+        "agnes": _DEFAULT,
     },
     "article": {
         "doubao": "2560x1440", "openai": _DEFAULT, "gemini": _DEFAULT,
         "dashscope": _DEFAULT, "minimax": _DEFAULT, "replicate": _DEFAULT,
         "azure_openai": _DEFAULT, "openrouter": _DEFAULT, "jimeng": _DEFAULT,
+        "agnes": _DEFAULT,
     },
     "vertical": {
         "doubao": "1088x2560", "openai": _DEFAULT_V, "gemini": _DEFAULT_V,
         "dashscope": _DEFAULT_V, "minimax": _DEFAULT_V, "replicate": _DEFAULT_V,
         "azure_openai": _DEFAULT_V, "openrouter": _DEFAULT_V, "jimeng": _DEFAULT_V,
+        "agnes": _DEFAULT_V,
     },
     "square": {
         "doubao": "2048x2048", "openai": _DEFAULT_SQ, "gemini": _DEFAULT_SQ,
         "dashscope": _DEFAULT_SQ, "minimax": _DEFAULT_SQ, "replicate": _DEFAULT_SQ,
         "azure_openai": _DEFAULT_SQ, "openrouter": _DEFAULT_SQ, "jimeng": _DEFAULT_SQ,
+        "agnes": _DEFAULT_SQ,
     },
 }
 
@@ -612,6 +617,42 @@ def generate(self, prompt: str, size: str) -> bytes:
         raise ValueError("Jimeng polling timeout")
 
 
+class AgnesProvider(ImageProvider):
+    """Agnes Image 2.1 Flash via Sapiens AI APIhub.
+
+    NOTE: Agnes requires `response_format` to be placed inside `extra_body`,
+    not at the top level of the request body.
+    """
+
+    provider_key = "agnes"
+
+    def __init__(self, api_key: str, model: str = "agnes-image-2.1-flash",
+                 base_url: str = "https://apihub.agnes-ai.com", **_kw):
+        self._api_key = api_key
+        self._model = model
+        self._base_url = base_url.rstrip("/")
+
+    def generate(self, prompt: str, size: str) -> bytes:
+        resp = requests.post(
+            f"{self._base_url}/v1/images/generations",
+            headers={"Content-Type": "application/json",
+                     "Authorization": f"Bearer {self._api_key}"},
+            json={"model": self._model, "prompt": prompt, "size": size,
+                  "extra_body": {"response_format": "url"}},
+            timeout=120,
+        )
+        data = resp.json()
+        if resp.status_code != 200:
+            err = data.get("error", {})
+            msg = err.get("message") if isinstance(err, dict) else str(err)
+            raise ValueError(f"Agnes error ({resp.status_code}): "
+                             f"{msg or str(data)}")
+        url = data.get("data", [{}])[0].get("url")
+        if not url:
+            raise ValueError(f"No image URL: {data}")
+        return _download_image(url)
+
+
 # --- Provider registry ---
 
 PROVIDERS = {
@@ -624,6 +665,7 @@ def generate(self, prompt: str, size: str) -> bytes:
     "azure_openai": AzureOpenAIProvider,
     "openrouter": OpenRouterProvider,
     "jimeng": JimengProvider,
+    "agnes": AgnesProvider,
 }