dsh-tool-vision 0.10.4 → 0.10.6

package/README.en.md

@@ -0,0 +1,48 @@
+# DSH Tool Vision
+
+[简体中文](README.md) | English
+
+Connect a separate OpenAI-compatible vision model to [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness), with an image bridge and image tools. A text-only conversation model can ask the vision model to inspect an image and use its description.
+
+## Features
+
+- **Vision calls:** `inspect_image` sends a local or HTTP(S) image to the configured endpoint, optionally with a question. The model must support compatible image input.
+- **Image bridge:** for models that do not receive images directly, converts images into durable text instructions so the Agent can call a vision tool as needed; conversation thumbnails and previews are supported.
+- **Detection and overrides:** combines route declarations, probe results and user lists to choose between direct image input and bridging; allowlists and blocklists can correct inaccurate declarations.
+- **Image tools:** includes image Q&A, grounding, OCR, crops, pixel differences, colors, vectorization, HTML screenshots and attachment materialization. Some operations require optional dependencies; desktop screenshots are disabled until explicitly enabled.
+
+Bridging gives a text model the vision model's description. The result depends on that model and does not turn the text model into a native multimodal model.
+
+## Compatibility and installation
+
+Supports DSH Web / Desktop. The repository declares compatibility with `0.2.0-rc.2`, `0.2.1-rc.1` / `rc.2` / `rc.3` / `rc.4` / `rc.5` and versions matching `>=0.2.0-rc.2 <0.3.0`; see [package.json](package.json) for the complete prerelease range. Node.js `>=20.18` is required.
+
+For Desktop, install its `dsh` command through **Application → Manage dsh Command** first:
+
+```sh
+dsh plugin --profile desktop add dsh-tool-vision
+```
+
+Replace `desktop` with `web` for Web. Restart DSH and open **Plugins → Installed → dsh-tool-vision**. You can also install it through the [Enhancement Suite](https://github.com/Scorp1o117/dsh-enhancement-suite).
+
+## Quick setup
+
+1. Enter the vision service's **API address, key and model ID** in the plugin page. A compatible local service can also be used; supply a key if the service requires one.
+2. Choose **Automatic images and previews** for bridging and previews, or **Vision tools only** for explicit calls. Existing custom combinations are preserved.
+3. If image capability detection is inaccurate, adjust the model list, list mode or automatic detection under Advanced. Save to apply.
+4. Send DSH an image and a question, or ask the Agent to inspect a workspace image with `inspect_image`.
+
+The form writes edited fields only and retains an unchanged key. The interface follows the DSH language setting.
+
+## Usage and limits
+
+- Models with image support can receive images directly. Bridged images enter the conversation as instructions and are not automatically restored to image blocks when switching models later.
+- `vision_probe_model` can test image capability. An explicit user list takes precedence over probe results, which in turn take precedence over route declarations.
+- `vision_screenshot` requires enabling `desktopScreenshot`. `sharp`, `potrace` and `puppeteer-core` are optional dependencies; affected tools report missing dependencies without disabling other tools.
+- Vision requests send images to the configured service. Choose a trusted local or remote endpoint for the content. Input-size and provider limits apply.
+
+## Development, attribution and feedback
+
+- `npm test` runs the project tests. The [historical detailed documentation](https://github.com/Scorp1o117/dsh-tool-vision/blob/00d818b47bd91c330b91207d2be0e50caf8cd87a/README.md) retains configuration fields, tool parameters and release notes; its install commands and compatibility information may be outdated.
+- Pixel-level tools were ported from [dsh-vision-router](https://github.com/ysr666/dsh-vision-router) (MIT).
+- [npm](https://www.npmjs.com/package/dsh-tool-vision) · [Issues](https://github.com/Scorp1o117/dsh-tool-vision/issues) · [MIT License](LICENSE)

package/README.md

@@ -1,550 +1,48 @@
-# dsh-tool-vision
+# DSH Tool Vision
 
-## DSH 0.2.1-rc.3
+简体中文 | [English](README.en.md)
 
-Supports DSH 0.2.1-rc.3, including explicit prerelease engines.dsh and peer ranges. This host update adds More navigation menus, keeps unsent drafts on New Session, and follows the official /models listing on the DeepSeek API-key route; the tool, settings and pi-ai interfaces used here remain unchanged, and no removed host API (`ctx.layout.entryVisibility`, `agentPresetNavigation`) was in use.
+为 [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) 接入独立的 OpenAI 兼容视觉模型，并提供图片桥接和图像工具。即使当前对话模型只能处理文本，也可以通过视觉模型查看图片并把描述带回对话。
 
-## DSH 0.2.1-rc.2
+## 功能
 
-Supports DSH 0.2.1-rc.2, including explicit prerelease engines.dsh and peer ranges. This host update adds DeepSeek V4.1 Flash 1017 and Search/Fetch APIs; the plugin APIs used here remain unchanged. Existing reasoning settings and the opt-in augmentation policy are preserved.
+- **视觉模型调用**：`inspect_image` 将本地图片或 HTTP(S) 图片发送到配置的视觉端点，可附带问题；模型需要支持兼容的图片输入。
+- **图片桥**：对不直接接收图片的模型，将图片转换为可持久化的文本指引，由 Agent 按需调用视觉工具；支持对话气泡缩略图和预览。
+- **模型识别与手动覆盖**：结合路由声明、实测结果和用户名单决定直发图片还是桥接；可用白名单 / 黑名单修正错误声明。
+- **图像工具**：包含看图问答、目标定位、OCR、裁剪、像素差异、取色、矢量化、HTML 截图和附件落盘等。部分操作需要可选依赖；桌面截图默认关闭，需在设置中主动启用。
 
-## DSH 0.2.1-rc.1
+图片桥让文本模型获取的是视觉模型返回的描述，效果取决于所选视觉模型，不能把文本模型变成原生多模态模型。
 
-Supports DSH 0.2.1-rc.1. Compatibility is declared through `engines.dsh` and matching peer ranges, including this exact prerelease. The existing 0.2.0-rc.2 support is retained.
+## 兼容性与安装
 
-The plugin follows the DSH language setting (Chinese and English in DSH 0.2.0-rc.2), including configuration, status messages and plugin-list metadata. Language-pack locales use the host fallback chain. Switching languages preserves unsaved settings; there is no separate plugin language selector.
+支持 DSH Web / Desktop；仓库声明兼容 `0.2.0-rc.2`、`0.2.1-rc.1` / `rc.2` / `rc.3` / `rc.4` / `rc.5` 和符合 `>=0.2.0-rc.2 <0.3.0` 的版本，预发布版本范围见 [package.json](package.json)。要求 Node.js `>=20.18`。
 
-## Configuration page (DSH 0.2.0-rc.2 and later)
+Desktop 先在 **应用 → 管理 dsh 命令** 安装桌面版 `dsh`：
 
-Open **Plugins → Installed → dsh-tool-vision** from the homepage sidebar to configure and save this plugin. The page uses the official `plugins.bundle.config` interface, without a duplicate entry in global Settings. Web and Desktop share the page. This version requires DSH 0.2.0-rc.2 or a later 0.2.x host; existing configuration is retained.
-
-Choose **Automatic images and previews** to enable image bridging, automatic admission and thumbnails together, or **Vision tools only** for explicit tool calls. Enter the endpoint, key and model on the main page; model overrides and request tuning live under **Advanced**. Existing mixed switch values appear as **Custom combination** and are retained until you change them. Selecting a mode edits a draft; click Save to apply it. Only edited fields are written, blank keys are preserved, and rejected writes keep the draft.
-
-[![中文文档](https://img.shields.io/badge/%E4%B8%AD%E6%96%87%E6%96%87%E6%A1%A3-blue)](README.zh.md)
-
-**GitHub**: [Scorp1o117/dsh-tool-vision](https://github.com/Scorp1o117/dsh-tool-vision) · **npm**: [dsh-tool-vision](https://www.npmjs.com/package/dsh-tool-vision)
-
-[![Enhancement Suite](https://img.shields.io/badge/part%20of-Enhancement%20Suite-3964fe)](https://github.com/Scorp1o117/dsh-enhancement-suite) [![npm](https://img.shields.io/npm/v/dsh-enhancement-suite)](https://www.npmjs.com/package/dsh-enhancement-suite)
-
-Part of the [DeepSeek Harness Enhancement Suite](https://github.com/Scorp1o117/dsh-enhancement-suite) — Vision · Soul/Persona · Long-term Memory · Plugin Marketplace.
-
-External vision model for [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness).
-
-DSH 0.1.1 adds native image input for DeepSeek's vision catalog. This plugin
-remains useful when you want a separate OpenAI-compatible vision endpoint,
-pixel-level image tools, screenshots, or a text-model bridge. The harness
-derives every model request strictly from the session log (`llm/stream`
-requests must equal the durable derivation — the agent-loop invariant), so the
-bridge keeps its conversion inside that durable path:
-
-1. **`inspect_image` tool** — sends an image (local file, or http(s) URL) to
-   **any OpenAI-compatible** `/chat/completions` endpoint that supports
-   `image_url` content parts, and returns the vision model's textual answer
-   into the agent loop.
-2. **Image bridge (v0.2.1)** — pasted images are turned into `inspect_image`
-   hints *before they enter the durable log*, on the `agent/pre-step`
-   waterfall (the one seam where the harness lets a plugin replace the
-   messages of a proposed step). Images already logged by an older version
-   are repaired lazily with a surface `replace` on the session's first
-   pre-step. Only models listed in `multimodalModels` receive image blocks
-   directly; a model's declared `inputModalities` are never consulted,
-   because profiles routinely declare `input: [text, image]` on text-only
-   models just to pass the harness's prompt-admission check.
-
-- Zero dependencies beyond the dsh SDK — works with any compatible endpoint:
-  OpenAI GPT-4o, Qwen-VL (DashScope), GLM-4V (Zhipu), Moonshot, Gemini
-  compatible endpoints, local Ollama, etc.
-- Registered on the **global tools layer**: every agent in the process can
-  call `inspect_image`.
-- **Web UI settings section (v0.3.0)**: Settings → 视觉模型 edits the
-  `tool-vision` namespace (API endpoint, write-only key, model, bridge
-  options) in the active Profile patch; changes hot-apply without a restart. The API
-  key lives in that patch. Mount by package name
-  (`name: 'dsh-tool-vision'`) so the web client bundle is discovered.
-
-## Compatibility (v0.9.7)
-
-Verified with DSH `0.1.7-rc.2` (Web) and `0.2.0-rc.2` (Desktop runtime) in isolated profiles. The Desktop app uses its own `desktop` profile. Other DSH prereleases remain unverified.
-
-## Desktop install
-
-Use the Desktop-installed `dsh` command (Application → Manage dsh Command), or the app’s Plugins page. Then install into the Desktop profile:
-
-```powershell
-dsh plugin --profile desktop add dsh-tool-vision@0.9.7
+```sh
+dsh plugin --profile desktop add dsh-tool-vision
 ```
 
-Restart the Desktop app to load the client bundle. Desktop keeps its profile under `$DSH_HOME/profiles/desktop`.
-
-
-## v0.9.4: Windows bridge export filenames
-
-- Export filenames now derive from a stable hash of the complete attachment ID. IDs such as `sha256:...` no longer create NTFS alternate data streams and zero-byte visible files.
-- The export directory is created before writing; the in-process cache now respects changes to `bridgeExportDir`.
-- Passed install, Web boot, homepage and client-bundle HTTP checks, and uninstall in a disposable DSH `0.1.5-rc.3` Profile.
-
-## Install
-
-Mount in a profile patch (`$DSH_HOME/profiles/<name>/cordis.patch.yml`):
-
-```yaml
-- insert:
-    - id: tool-vision
-      name: 'dsh-tool-vision'     # after: pnpm add dsh-tool-vision in the profile
-      config:
-        baseURL: 'https://api.openai.com/v1'
-        apiKeyEnv: 'VISION_API_KEY'
-        model: 'gpt-4o-mini'
-```
-
-Or load it from a local path without npm:
-
-```yaml
-    - id: tool-vision
-      name: './plugins/dsh-tool-vision/index.js'
-```
-
-## Config
-
-| Field | Default | Meaning |
-|---|---|---|
-| `enabled` | `true` | **Master switch (v0.8.0).** Off unregisters everything this plugin contributes — `inspect_image`, the 14 `vision_*` tools, the image bridge, the preview route and the image-capability declaration. The settings section stays mounted so the switch can turn it back on. Hot-applies; no dsh restart. |
-| `baseURL` | `https://api.openai.com/v1` | OpenAI-compatible API base URL. |
-| `apiKey` | `''` | API key (takes precedence over env). |
-| `apiKeyEnv` | `VISION_API_KEY` | Env var holding the key. |
-| `model` | `gpt-4o-mini` | Vision model id. |
-| `maxTokens` | `1024` | Max output tokens. |
-| `timeoutMs` | `60000` | Per-request timeout. |
-| `maxImageBytes` | `10MB` | Largest accepted local image. |
-| `description` | default | Tool description shown to the model. |
-| `bridgeTextOnly` | `true` | Bridge pasted images to text hints on models that cannot see images. |
-| `bridgeExportDir` | temp | Export dir for bridged images (`os.tmpdir()/dsh-vision-bridge`). Filenames use a portable hash of the attachment ID. |
-| `multimodalModels` | `[]` | Model list (comma-separated). Each entry is matched case-insensitively against the full id, its bare id after the last `/`, and `provider/id`, with `*` / `?` globs (`*vl*`, `deepseek/*`). What the list *means* is set by the mode below. |
-| `multimodalListMode` | `whitelist` | **List mode (v0.9.0).** `whitelist`: listed models receive image blocks directly (the historical behaviour). `blacklist`: listed models are forced through the bridge — the correction layer for a model that claims image support it does not have. `off`: the list is ignored. An unknown value falls back to `whitelist`. |
-| `autoDetectMultimodal` | `true` | **Auto-detect (v0.9.0).** Decide from the current route's own declared `inputModalities`, then combine with the list (whitelist unions, blacklist subtracts). On by default: a text-only route is bridged, a multimodal one is treated like a whitelist member and gets images directly. The declaration is always read *before* this plugin's admission wrap, so `bridgeAutoImage` can never feed its own claim back in as evidence. Set false for the hand-maintained "list only" behaviour. |
-| `probeResults` | `{}` | **Measured verdicts (v0.9.0):** `"provider/model" → "yes"`/`"no"`, written by the `vision_probe_model` tool — do not edit by hand. A measurement outranks a **declaration** (a real request beats a claim) but not `multimodalModels` (explicit human intent has the last word). |
-| `bridgePreview` | `true` | Inline preview for bridged images: thumbnail above the hint text in the user bubble (click to zoom). |
-| `bridgePreviewScanIntervalMs` | `2000` | Fallback scan interval for the preview scanner (ms); `0` disables the fallback. |
-| `bridgePreviewHideHint` | `true` | Hide the bridged hint text once the preview image has loaded (kept on failure — safe degradation). |
-| `bridgeAutoImage` | `true` | While the bridge is on, report image input capability for **every** model to the host admission gate, so pasted images are accepted on text-only models without hand-editing provider configs. |
-| `sendSessionHeader` | `true` | Send a stable session-id header on vision requests. OpenCode Go and similar gateways require `x-opencode-session` (one stable id per conversation); requests without it may error from 2026-09-06. |
-| `sessionHeaderName` | `x-opencode-session` | Header name carrying the session id. |
-| `sessionId` | `''` | Fixed session id for calls without a dsh session context; empty = auto (current dsh session id, else a stable per-process random id). |
-
-## Image bridge setup
-
-1. (Optional, usually not needed) If `bridgeAutoImage` is disabled, declare
-   image input on the models you paste images onto, so the harness admits
-   image messages (pi-ai style):
-   ```yaml
-   llm-pi-ai:
-     providers:
-       your-provider:
-         models:
-           - id: deepseek-v4-flash
-             input: [text, image]
-   ```
-2. List genuinely multimodal models in the plugin config so they receive
-   image blocks untouched (see the next section for the list modes):
-   ```yaml
-   - id: tool-vision
-     name: 'dsh-tool-vision'
-     config:
-       multimodalListMode: whitelist    # default: listed models get images directly
-       multimodalModels: ['mimo-v2.5', 'grok-4.5']
-   ```
-
-Then pasting an image while on a text-only model stores a hint like
-`[User sent an image, exported to: <path>. Inspect it with the inspect_image tool...]`
-in the transcript (the pasted image no longer renders as pixels in that
-message), and the agent inspects it through the configured vision endpoint.
-
-> Why not `llm/stream`? The harness freezes every request and the agent-loop
-> invariant fails any request whose messages diverge from the session-log
-> derivation (`log-reconstruction desync`), and this cordis waterfall's
-> `next()` cannot replace request arguments. The `agent/pre-step` waterfall is
-> the supported seam: its decision messages *become* the durable log, so the
-> invariant stays satisfied.
-
-Key resolution order: `config.apiKey` → `process.env[apiKeyEnv]` →
-`process.env.OPENAI_API_KEY`.
-
-## Bridge image preview (v0.4.0)
-
-On text-only models, pasted images become `[User sent an image...]` hint
-text in the transcript. With `bridgePreview` enabled (default), the browser
-half renders those hints as inline thumbnails **in the display layer only**:
-
-- **Thumbnail + lightbox**: click to zoom full-screen; click anywhere or
-  press `Esc` to close;
-- **Immediate + fallback**: new messages are handled by a MutationObserver;
-  history is back-filled by a periodic scan (interval via
-  `bridgePreviewScanIntervalMs`);
-- **Hide the hint (P2)**: with `bridgePreviewHideHint` on, the hint text is
-  hidden once the image has loaded, leaving just the image; on load failure
-  the text stays (safe degradation — never "no image AND no text");
-- **Precise identification**: bridged hints carry an invisible prefix marker
-  (`\u200b[bridge]`), so ordinary user text that happens to contain
-  "exported to:" is never misidentified;
-- **Display-layer red line**: persisted messages, the transcript, the
-  model-facing text and the `inspect_image` chain are untouched.
-
-Preview images are served by the same-origin loopback route
-`/plugins/dsh-tool-vision/image`: read-only access to the bridge export
-directory, localhost-only Host, image extensions only, ≤ 20MB per file,
-path-traversal protected.
-
-## Tool: `inspect_image`
-
-| Arg | Required | Meaning |
-|---|---|---|
-| `path` | ✅ | Image path (absolute, or relative to the current workspace) or http(s) URL. |
-| `question` | – | Optional specific question about the image. |
-| `detail` | – | `auto` / `low` / `high` resolution hint. |
-
-Example endpoints (`baseURL`):
-
-- **OpenAI**: `https://api.openai.com/v1` — `gpt-4o`, `gpt-4o-mini`
-- **Alibaba DashScope (Qwen-VL)**: `https://dashscope.aliyuncs.com/compatible-mode/v1` — `qwen-vl-plus`, `qwen-vl-max`
-- **Zhipu (GLM-4V)**: `https://open.bigmodel.cn/api/paas/v4` — `glm-4v-flash` (free tier), `glm-4v-plus`
-- **Moonshot (Kimi)**: `https://api.moonshot.cn/v1` — `moonshot-v1-8k-vision-preview`
-- **Ollama local**: `http://localhost:11434/v1` — `llama3.2-vision` (no key)
-
-
-> **Note for users**
-> - This plugin is a standard **profile bundle** (`dsh.bundle.patch`):
->   `dsh plugin --profile web add dsh-tool-vision` installs and mounts it in
->   one step — no manual `cordis.patch.yml` edits needed.
-> - Settings changes hot-apply (no restart needed).
-> - Version 0.6.3 and newer require DSH `0.1.0-rc.7` or newer and are tested
->   against `0.1.0-rc.7`, `0.1.0-rc.8`, and `0.1.1-rc.1`.
-> - DSH `0.1.0-rc.6` users must pin `dsh-tool-vision@0.6.1`, the last release
->   carrying the legacy settings-allowlist compatibility patch.
-
-## Pixel-level vision tools (v0.6.0, ported from dsh-vision-router)
-
-14 `vision_*` tools driven by the **same** configured endpoint as
-`inspect_image` (baseURL/apiKey/model) — no provider chain, no local models,
-no extra settings:
-
-| Tool | Purpose |
-|---|---|
-| `vision_describe` | Image Q&A / multi-image comparison (optional structured JSON) |
-| `vision_ground` | Locate a target and return its ORIGINAL-pixel bounding box |
-| `vision_detect` | Enumerate elements (buttons, inputs, icons…) with numbered boxes |
-| `vision_crop` | Crop a pixel region to a PNG artifact |
-| `vision_pixel_diff` | Per-pixel comparison: ratio, worst regions, heatmap, report |
-| `vision_colors` | Dominant-color quantization for palette matching |
-| `vision_ocr` | Verbatim text transcription (letters only — not scene analysis) |
-| `vision_long_screenshot_ocr` | Chunked long-screenshot transcription into Markdown |
-| `vision_trace` | Potrace vectorization into colored SVG (worker-thread, safe) |
-| `vision_extract_foreground` | Solid-background removal → transparent PNG |
-| `vision_html_screenshot` | Headless render of a local .html (network blocked) |
-| `vision_screenshot` | Desktop capture (privacy-gated: enable `desktopScreenshot` in settings; Win: PowerShell / macOS: screencapture / Linux: import/scrot) |
-| `vision_present` | Publish a generated image to the user via the host attachment store |
-| `vision_materialize` | Copy an attachment/local image into the workspace as a real path |
-
-Quality & safety details:
-
-- **Content-hash cache** keyed by endpoint+model+image+question (no stale
-  answers across model switches, failures are never cached).
-- **Uniform 4MP downscale** before every model call; oversized inputs are
-  rejected with a clear error (stat pre-check, 20MB cap on both file and
-  attachment paths).
-- **Rate-limit / 5xx auto-retry** with Retry-After-aware backoff; endpoint
-  **content-safety rejections** are surfaced as `VISION_CONTENT_FILTERED`
-  instead of a generic backend error.
-- **Long-OCR bounds**: 120s total budget, 40-chunk cap, cancellation checks,
-  stop-on-first-backend-failure.
-- **Path containment** for relative inputs; artifacts land in
-  `<workspace>/.dsh-tool-vision/`.
-
-Requires `sharp` / `potrace` / `puppeteer-core` (declared as optional
-dependencies: a failed platform install never blocks the plugin; missing ones
-degrade lazily with an install hint and never break other tools).
-
-`vision_screenshot` is privacy-sensitive and therefore **not registered by
-default** — set `desktopScreenshot: true` in the tool-vision settings to
-enable desktop capture.
-
-## v0.9.0: list modes, and auto-detection
-
-The bridge answers one question: **can the current model see images directly?**
-v0.9.0 splits it into two independent inputs.
-
-```
-base = autoDetectMultimodal ? (route declares image) : {}
-off        → direct = base              the list takes no part
-whitelist  → direct = base ∪ list       the list only adds
-blacklist  → direct = base \ list       the list only subtracts
-```
-
-**A list hit always wins**: the list is explicit user intent, so it outranks the
-model's own declaration — which is what makes it a usable correction layer.
-
-**A blacklist never degrades into "everything unlisted is direct".** Its base set
-is the auto-detected one; with auto-detection off that base set is empty, so an
-unlisted model is still bridged. That is deliberate: the alternative lets one
-typo push images at a text-only endpoint.
-
-**Matching**: `mimo-v2.5`, `xiaomi/mimo-v2.5` and `commandcode/xiaomi/mimo-v2.5`
-all address the same route; `*` / `?` are globs; matching is case-insensitive.
-v0.8.1 compared ids literally, so this README's own `mimo-v2.5` example silently
-did nothing on a route spelled `xiaomi/mimo-v2.5` — fixed here, and the fix only
-ever *adds* models to the direct set (no entry that used to force a model direct
-stops doing so).
-
-**In the panel** (Plugins → dsh-tool-vision Model):
-
-- the list field is followed by a **clickable list of the models dsh actually has
-  configured** (`llm.listProviders()` + `llm.listModels()`), grouped by provider
-  and labelled with whether the route *declares* image input. Tick to add,
-  untick to remove; the text field above still takes globs by hand. Both edit the
-  *same* draft, persisted by Save — the picker head flags it as unsaved until then.
-- a **current route** readout shows `provider / model`, whether images go direct
-  or through the bridge, and why (list hit / auto-detect / default).
-
-> Why not just a `<datalist>`: a native datalist stays invisible until the user
-> focuses the field and types, which made v0.9.0's first cut look like a dead
-> panel. The list is now always visible, with the datalist kept as a typing aid.
->
-> Unticking removes **the entry that actually matched** — if `mimo-v2.5` in the
-> list is what covers `xiaomi/mimo-v2.5`, unticking drops `mimo-v2.5` rather than
-> inventing a full id. Which entry hit is computed server-side with the *same*
-> matcher the bridge uses (`matchedEntries` in the payload), so the panel can
-> never display a state that disagrees with the decision.
-
-**Read path (the easiest thing to get wrong here)**: `autoDetectMultimodal` MUST
-read the value from *before* `resolveModelInfo` was wrapped, or the "image
-support" that `bridgeAutoImage` stamps onto every model becomes evidence for
-itself. `unwrappedResolveModelInfo()` enforces that, with a dedicated regression
-test.
-
-Candidates come from the plugin's own loopback route:
-`GET /plugins/dsh-tool-vision/models` (loopback Host only, read-only, `no-store`).
-It returns provider/model ids and one declared-capability boolean — **no keys and
-no endpoint addresses**. It is registered on the plugin fiber rather than the
-master switch's child fiber, so the panel keeps working while the plugin is off.
-
-> ⚠️ `inputModalities` is a **declaration**, not a guarantee — profiles commonly
-> set `input: [text, image]` on text-only models just to pass the admission gate.
-> Upstream `dsh-llm-pi-ai` makes the same call for undeclared models, and its
-> source says why: the two wrong answers **do not cost the same**. Under-claiming
-> refuses the image before it is attached and names the model; over-claiming
-> admits one the provider rejects mid-turn, after the message is already durable.
->
-> That is why detection is on by default *with* three backstops: (1) the first
-> time a route is promoted purely by its own declaration, the log says so and
-> names the fix; (2) the panel always shows the current route, the decision and
-> the reason; (3) listing that model under `blacklist` mode forces the bridge
-> back on.
-
-## Tests
-
-```bash
-npm test             # server-side unit tests (no extra dependencies)
-npm run test:render  # panel render test (needs devDependencies)
-```
-
-`npm run test:render` loads the **real client bundle** in jsdom, drives the
-**real registration path** (`apply` → `slots.register` → the component), feeds it
-from the **real server route handler**, and asserts on the real DOM and the real
-settings writes.
-
-It is a separate command and deliberately **not part of `npm test`**: it needs
-`react` / `react-dom` / `jsdom`, and a DOM test that silently skips when a
-dependency is missing is a false comfort. Install with
-`npm i -D react@18 react-dom@18 jsdom`.
-
-Its reason to exist is specific: v0.9.0's first cut rendered the model list only
-into a native `<datalist>` — every server-side unit test passed while the panel
-looked completely dead. Nothing below the DOM can catch that class of bug.
-
-## Capability probe: `vision_probe_model` (v0.9.0)
-
-Each request defaults to a 60-second deadline and 2048 output tokens. Response
-bodies are limited to 1 MiB while streaming; an oversized body stops the read
-and yields `unknown`. Authentication, rate-limit and server errors remain
-`unknown` even if their messages mention image input. Only HTTP 400, 415 or
-422 image-rejection errors yield a negative verdict from an error response.
-
-Every other signal here rests on what a model **says** about itself. This tool
-sends a real image to the route and reports what it **does** — the only
-ground truth in the plugin.
-
-```
-base = autoDetect ? route declares image : {}
-a measured verdict (if any) overrides base    measurement > claim
-a list hit (if any) overrides everything      human intent > measurement
-```
-
-**Why one request is not enough** (each of these was learned the hard way):
-
-- **Models guess.** One may answer "blue" to an orange square, so a probe asks
-  about **two different colors** and only passes a model that reads both.
-- **Reasoning models return empty `content`** when the thinking budget eats
-  `max_tokens`; the default is 2048 and the reader falls back to
-  `reasoning_content`.
-- **A control group is mandatory.** A text-only request runs first to prove the
-  route works at all; without it a 401 or a timeout would be misread as "cannot
-  see images" and push a perfectly good multimodal route back onto the bridge
-  **permanently**.
-- **An accepting endpoint is not an image-reading model.** Measured against a
-  real gateway: `meituan/LongCat-2.0:free` returned HTTP 200 for the image
-  request and answered `"I can't see any image."` Only an endpoint that
-  *actively rejects* the image part is a conclusive negative.
-
-So a probe ends in one of three verdicts: `yes` (control passed, both colors
-correct), `no` (the endpoint rejected the image part, or answered without
-reading it), or `unknown` (network, auth or protocol trouble — **never** turned
-into a capability claim).
-
-**How it gets a route's endpoint and credential** (no new configuration):
-`llm.listConfigurableProviders()` names the provider's settings namespace and
-path → `settings.get(ns)` resolves `baseURL`/`apiKeyEnv`/`api` →
-`credentials.resolve(apiKeyEnv)` yields the secret (`.value`) — the same path
-`dsh-llm-pi-ai` uses for a real call. Read-only, and neither the endpoint nor
-the key ever appears in a probe result.
-
-**Linking**: a verdict is written to `probeResults` and takes effect in the
-bridge decision **immediately** — `yes` sends images directly, `no` forces the
-bridge back on — regardless of the list mode. (That is also why it does not
-silently write into `multimodalModels`: a `blacklist` list means the *opposite*,
-so an automatic entry there would produce exactly the wrong result.) Each row in
-the panel's picker carries a badge:
-
-- `measured: reads images` / `measured: no image reading` (blue / red), shown in
-  preference to the declaration;
-- a never-probed route falls back to the `declares image` badge;
-- the current-route readout reports `measured (a real image was read / NOT read)`.
-
-> A route on an unknown protocol (e.g. `openai-responses`) returns `unknown`
-> with a reason, rather than guessing confidently with `chat/completions`.
-
-## v0.9.3: adapter snapshot model admission (pi-ai guard)
-
-**The bug.** `v0.9.2`'s `installDispatchImageAdmission` satisfied DSH core's
-`LlmService.generate` by declaring `image` on `adapterCall.model.inputModalities`.
-However, `dsh-llm-pi-ai`'s adapter executes a second internal guard inside its
-streaming path (`streamWithSnapshot`):
-
-```js
-const model = this.modelOf(snapshot, options.provider, options.model);
-if (containsImage && !model.input.includes("image"))
-  throw new LlmError(`pi-ai model "${model.id}" does not support image input`, "UNSUPPORTED_CONTENT");
-```
-
-`this.modelOf` resolves from the adapter's own `snapshot.models` catalog, which
-defaults to `input: ["text"]` unless the user explicitly declared
-`input: [text, image]` in the Profile patch. When admitted by `v0.9.2`, the raw image
-reached `streamWithSnapshot` and triggered `UNSUPPORTED_CONTENT`. Since the image
-was already in the session's durable transcript, every subsequent turn failed.
-
-**The fix.** On direct routes, `installDispatchImageAdmission` now additionally:
-1. Wraps `adapter.modelOf` (when present) to include `"image"` in `resolved.input`;
-2. Enriches the live snapshot model (`snapshot.models.getModel(route, model).input`)
-   with `"image"`;
-3. Fully restores `adapter.modelOf` and reverts modified input arrays upon dispose.
-
-## v0.9.2: admission is not dispatch
-
-**The bug.** Every capability signal this plugin collects — `probeResults`,
-`multimodalModels`, `autoDetectMultimodal` — decided whether the *bridge*
-intercepts an image. None of them decided whether the model *receives* it.
-
-`LlmService.generate` resolves modalities from the adapter, not from the plugin:
-
-```js
-const adapterCall = await adapter.prepareCall(provider, model, signal);
-modelInfo = this.normalizeModelInfo(registration, model, adapterCall.model);
-if (modelInfo.inputModalities !== undefined
-    && !modelInfo.inputModalities.includes("image")
-    && projectedMessages.some((message) => contentHasImage(message.content)))
-  projectedMessages = projectImagesForTextModel(projectedMessages);
-```
-
-`projectImagesForTextModel` rewrites every image block into
-`[image omitted because this model accepts text only; attachment sha256:…]`
-**before the adapter is called**. So on a route the plugin had measured as
-image-capable — where the bridge therefore stepped aside and let the image
-through — the image was still destroyed, by a check that reads the adapter's
-declaration and nothing else. The `resolveModelInfo` wrap (`bridgeAutoImage`) is
-*admission*: it decides who may offer an image, and it never changes what the
-adapter streams. Nothing in the plugin reached the layer that does.
-
-The symptom is a route that probes `yes`, is listed in `multimodalModels`, and
-still answers `[image omitted …]`.
-
-**The fix.** `installDispatchImageAdmission` wraps `llm.registration` — the
-accessor the core itself calls — so every adapter, including one registered
-later, answers `prepareCall` through the same `routeDirectDecision` the bridge
-uses. One precedence rule, two seams, so the two can never disagree about a
-route. When the decision is "bridge", nothing changes: the bridge already turned
-the image into an `inspect_image` hint, and the core's projection stays as the
-correct fallback for any image that still reaches dispatch.
-
-Three properties worth stating, because each one is a test:
-
-- **Fail closed.** A decision that cannot be made (a throwing config) leaves the
-  call untouched, rather than sending pixels a route may reject.
-- **Reversible.** Dispose restores every adapter it patched *and* the accessor
-  itself; a re-install after dispose wraps again.
-- **Late providers count.** The wrap is on the accessor rather than on a snapshot
-  of adapters, so a provider registered after install is covered too.
-
-`routeDirectDecision` is now the single implementation of the precedence rule,
-shared by the bridge and the dispatch path — two copies of it would drift, and
-this one has three inputs.
-
-## v0.8.0: master switch, and the save-path fix
-
-**Master switch.** `enabled`, plus a one-click button at the top of the section
-(`Disable all` / `Re-enable`). Registrations are effects on the cordis fiber
-that makes them, so the plugin now puts every tool, the image bridge, the
-preview route and the image-capability declaration in a **child fiber**: turning
-the switch off disposes it, and all 15 tools leave the model's tool list
-together. The settings section stays on the parent fiber, so the switch can turn
-the plugin back on. No dsh restart.
-
-**Save-path fix.** The form used to submit its 18 fields as **parallel**
-`scope.set()/unset()` calls. Each write carries its own revision fence, a fence
-behind the Host document is refused with `settings/conflict`, and **a refused
-write still resolves** — the scope's contract is "settle after the write and any
-recovery read", not "throw on refusal". The section therefore reported "Saved"
-while the edits silently reverted, which reads as "settings cannot be saved at
-all".
+Web 将 `desktop` 改为 `web`。安装后重启 DSH，打开 **插件 → 已安装 → dsh-tool-vision**。也可通过 [Enhancement Suite](https://github.com/Scorp1o117/dsh-enhancement-suite) 安装。
 
-Writes are now **one atomic `mutate()`**, so the whole batch shares one fence and
-one persistence decision, and the section is inspected after the write settles:
-"Saved" only when the change is really there, otherwise "Write did not take
-effect" plus a reload of the form. Hosts without `mutate()` fall back to
-**sequential** writes (each waits for its predecessor, keeping the revision chain
-intact) — never parallel.
+## 快速配置
 
-Also removes three `if (typeof scope.load === "function") scope.load()` guards.
-The `SettingsScope` seam has never had `load()` — it is `getSnapshot` /
-`subscribe` / `mutate` / `set` / `unset`, and reads ride the shared describe
-mirror driven by the Host's `settings/document-updated`. Those guards were dead
-code that read like a refresh which never happened, and they made the missing
-write verification look intentional.
+1. 在插件页填写视觉服务的 **API 地址、密钥和模型 ID**。也可使用兼容图片输入的本地服务；密钥按服务要求填写。
+2. 选择**自动图片与预览**以启用桥接等功能，或选择**仅视觉工具**进行显式调用；已有自定义组合不会自动覆盖。
+3. 如模型的图片能力识别不准确，在高级设置中调整模型名单、名单模式或自动识别。保存后生效。
+4. 向 DSH 发送图片并提问，或让 Agent 使用 `inspect_image` 查看工作区图片。
 
-## Limitations
+设置表单只写入编辑过的字段；未修改的密钥保留。语言跟随 DSH。
 
-- A bridged image enters the conversation as a text hint (a transcript, not
-  pixels) — pixel-precise in-context reasoning is not available to text-only
-  models; the vision model's description comes back through `inspect_image`.
-- The bridge is a **one-way door**: an image pasted on a text-only model is
-  rewritten into the durable log at `agent/pre-step`, so switching to a
-  multimodal model later does not turn it back into an image block. (The other
-  direction — multimodal to text-only — is repaired automatically by
-  `repairLoggedImages`.)
-- Images are base64-transferred; mind privacy and size limits.
-- Independent of the dsh-llm routing/retry system; failures return clear
-  errors to the agent.
+## 使用与限制
 
-## License
+- 支持图片的当前模型可直接接收图片；走桥接的图片以指引进入会话，之后切换模型不会自动还原原始图片块。
+- `vision_probe_model` 可实测模型图片能力；手动名单优先于实测，实测优先于路由声明。
+- `vision_screenshot` 需在设置中开启 `desktopScreenshot`。`sharp`、`potrace` 和 `puppeteer-core` 是可选依赖；缺失时受影响的工具会提示，其他工具仍可使用。
+- 视觉请求会将图片发送给配置的服务，请根据内容选择可信的本地或远程端点。输入大小与服务端能力均有限制。
 
-MIT — bridge preview & integration: xing666173. Pixel vision tools ported
-from [dsh-vision-router](https://github.com/ysr666/dsh-vision-router)
-(© ysr666, MIT) with gratitude.
+## 开发、来源与反馈
 
+- `npm test` 运行项目测试；[历史详细文档](https://github.com/Scorp1o117/dsh-tool-vision/blob/00d818b47bd91c330b91207d2be0e50caf8cd87a/README.zh.md)保留配置字段、工具参数和旧版本说明，其中安装命令与兼容性信息可能已过时。
+- 像素级视觉工具移植自 [dsh-vision-router](https://github.com/ysr666/dsh-vision-router)（MIT）。
+- [npm](https://www.npmjs.com/package/dsh-tool-vision) · [问题反馈](https://github.com/Scorp1o117/dsh-tool-vision/issues) · [MIT License](LICENSE)

package/README.zh.md

@@ -1,381 +1,5 @@
-# dsh-tool-vision
+# DSH Tool Vision（简体中文）
 
-## DSH 0.2.1-rc.3
+中文文档已迁移至仓库首页：[阅读中文版](README.md)。
 
-适配 DSH 0.2.1-rc.3，明确声明该预发布版本的 engines.dsh 与 peer 范围。宿主新增侧栏/会话「更多」导航菜单、修复新建会话丢失未发送草稿、DeepSeek API Key 路由改跟随官方 /models 模型列表；本插件使用的工具、settings 与 pi-ai 接口保持不变，也未使用任何被移除的宿主 API（`ctx.layout.entryVisibility`、`agentPresetNavigation`）。
-
-## DSH 0.2.1-rc.2
-
-适配 DSH 0.2.1-rc.2，明确声明该预发布版本的 engines.dsh 与 peer 范围。宿主新增 DeepSeek V4.1 Flash 1017 及 Search/Fetch API，本插件使用的接口保持不变。保留已有推理配置及自动补齐显式开启策略。
-
-## DSH 0.2.1-rc.1
-
-支持 DSH 0.2.1-rc.1，通过 `engines.dsh` 和一致的 peer 范围声明兼容，明确包含该预发布版本，并保留已有 0.2.0-rc.2 支持。
-
-插件跟随 DSH 的语言设置（DSH 0.2.0-rc.2 内置中文和 English），配置页面、状态提示和插件列表名称/简介同步切换。扩展语言使用宿主的回退链。切换语言保留未保存的设置，无需单独选择插件语言。
-
-## 配置入口（DSH 0.2.0-rc.2 起）
-
-在首页侧边栏打开 **插件 → 已安装 → dsh-tool-vision**，直接在插件详情页配置并保存。配置页注册到官方的 `plugins.bundle.config` 接口；全局设置页不再重复显示配置入口。Web 与桌面版使用相同界面，本版要求 DSH 0.2.0-rc.2 或更新的 0.2.x 版本。现有配置无需迁移。
-
-选择 **自动接图与预览**，一次开启图片桥接、自动接图和缩略图；选择 **仅使用视觉工具**，保留手动工具调用。主页面填写端点、密钥和模型，模型覆盖与请求参数在 **高级设置** 中。已有开关组合显示为 **自定义组合**，不会自动改写。选择模式只修改草稿，点击保存后才生效；只写入编辑过的字段，密钥留空保持原值，写入被拒绝时保留草稿。
-
-**GitHub**: [Scorp1o117/dsh-tool-vision](https://github.com/Scorp1o117/dsh-tool-vision) · **npm**: [dsh-tool-vision](https://www.npmjs.com/package/dsh-tool-vision) · [English](README.md)
-
-[![Enhancement Suite](https://img.shields.io/badge/part%20of-Enhancement%20Suite-3964fe)](https://github.com/Scorp1o117/dsh-enhancement-suite) [![npm](https://img.shields.io/npm/v/dsh-enhancement-suite)](https://www.npmjs.com/package/dsh-enhancement-suite)
-
-属于 [DeepSeek Harness Enhancement Suite](https://github.com/Scorp1o117/dsh-enhancement-suite) —— Vision · Soul/Persona · 长期记忆 · 插件市场。
-
-给 [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) 外接**视觉模型**的插件。
-
-DSH 0.1.1 已为 DeepSeek 视觉模型目录加入原生图片输入。本插件继续提供独立的 OpenAI 兼容视觉端点、像素级图片工具、截图，以及文本模型图片桥。Harness 的每次模型请求都**严格从会话日志推导**（`llm/stream` 请求必须与持久化推导一致，否则 agent-loop invariant 会报 `log-reconstruction desync`），因此图片桥仍在可持久重建的路径内完成转换：
-
-1. **`inspect_image` 工具** —— 把图片（本地文件或 http(s) URL）发给任意支持 `image_url` 内容块的 OpenAI 兼容 `/chat/completions` 端点，把视觉模型的文字回答带回对话。
-2. **图片桥（v0.2.1）** —— 粘贴的图片在**进入持久化日志之前**就被转换成 `inspect_image` 指引文本，拦截点是 `agent/pre-step` waterfall（这是 harness 唯一允许插件替换"进入某一步的消息"的缝；替换后的消息会**成为**持久化的 `user/message` 日志，所以请求重建 invariant 天然满足）。旧版本已经写进日志的图片消息，会在该会话下一次 pre-step 时用 surface `replace` 惰性修复。只有 `multimodalModels` 白名单内的模型直收图片块；**不参考模型的 `inputModalities` 声明**——因为很多配置为了通过 prompt 准入检查，会给纯文本模型声明 `input: [text, image]`（那只是声明，不代表上游真的能吃 `image_url`）。
-
-- 除 dsh SDK 外零依赖 —— 兼容任意端点：OpenAI GPT-4o、Qwen-VL（DashScope）、GLM-4V（智谱）、Moonshot、Gemini 兼容端点、本地 Ollama 等。
-- 注册在**全局工具层**：进程内所有 Agent 都能调用 `inspect_image`。
-- **Web UI 设置栏（v0.3.0）**：插件 → dsh-tool-vision 编辑 `tool-vision` 配置（API 地址、只写密钥、模型、桥接选项），写入当前 Profile patch，**改动即时生效无需重启**。插件按包名挂载（`name: 'dsh-tool-vision'`）以便 web 端发现客户端 bundle。
-
-## 兼容性（v0.9.7）
-
-已在 DSH `0.1.7-rc.2`（Web）与 `0.2.0-rc.2`（桌面端运行时）的一次性 Profile 验证。桌面端使用独立的 `desktop` Profile；其他预发布版本暂未验证。
-
-## 桌面端安装
-
-在桌面端的“插件”页面安装，或使用桌面端“应用 → 管理 dsh 命令”注册的命令：
-
-```powershell
-dsh plugin --profile desktop add dsh-tool-vision@0.9.7
-```
-
-重启桌面端以加载客户端插件。配置位于 `$DSH_HOME/profiles/desktop`。
-
-
-## v0.9.4：修复 Windows 图片桥导出文件名
-
-- 导出文件名改为从完整附件 ID 生成稳定哈希；`sha256:...` 等 ID 不再写入 NTFS 备用数据流，避免出现可见主文件为 0 字节的情况。
-- 写入前确保导出目录存在；进程内缓存按 `bridgeExportDir` 区分。
-- 已通过 DSH `0.1.5-rc.3` 一次性 Profile 的安装、Web 启动、首页与客户端 Bundle HTTP 检查，以及卸载验证。
-
-## 安装
-
-在 profile patch（`$DSH_HOME/profiles/<name>/cordis.patch.yml`）里挂载：
-
-```yaml
-- insert:
-    - id: tool-vision
-      name: 'dsh-tool-vision'     # 前置：在 profile 里 pnpm add dsh-tool-vision
-      config:
-        baseURL: 'https://api.openai.com/v1'
-        apiKeyEnv: 'VISION_API_KEY'
-        model: 'gpt-4o-mini'
-```
-
-不装 npm 包、直接加载本地路径：
-
-```yaml
-    - id: tool-vision
-      name: './plugins/dsh-tool-vision/index.js'
-```
-
-## 配置
-
-| 字段 | 默认值 | 含义 |
-|---|---|---|
-| `enabled` | `true` | **总开关（v0.8.0）**。关闭后插件不注册任何东西：`inspect_image`、14 个 `vision_*` 工具、图片桥、预览路由、图片能力声明统统下线；设置栏本身保留，所以随时能再打开。热生效，无需重启 dsh |
-| `baseURL` | `https://api.openai.com/v1` | OpenAI 兼容 API 基地址 |
-| `apiKey` | `''` | API 密钥（优先于环境变量） |
-| `apiKeyEnv` | `VISION_API_KEY` | 存放密钥的环境变量名 |
-| `model` | `gpt-4o-mini` | 视觉模型 id |
-| `maxTokens` | `1024` | 视觉调用最大输出 token |
-| `timeoutMs` | `60000` | 单次请求超时 |
-| `maxImageBytes` | `10MB` | 本地图片大小上限 |
-| `description` | 默认描述 | 工具描述（模型可见） |
-| `bridgeTextOnly` | `true` | 把粘贴图片转成文本指引（发给看不懂图片的模型时） |
-| `bridgeExportDir` | 临时目录 | 桥接图片导出目录（`os.tmpdir()/dsh-vision-bridge`）；文件名使用附件 ID 的安全哈希 |
-| `multimodalModels` | `[]` | 模型名单（逗号分隔）。每项按「完整 id / 末段裸 id / `provider/id`」三种写法匹配，大小写不敏感，支持 `*` `?` 通配（如 `*vl*`、`deepseek/*`）。含义由下面的模式决定 |
-| `multimodalListMode` | `whitelist` | **名单模式（v0.9.0）**：`whitelist` 名单内模型直收图片块（旧行为）；`blacklist` 名单内模型强制走桥接（用来纠正"声明支持图片但实际不支持"的模型）；`off` 忽略名单。未知值一律回退 `whitelist` |
-| `autoDetectMultimodal` | `true` | **自动识别（v0.9.0）**：按当前路由自己声明的 `inputModalities` 判定，再与名单合成（白名单取并集、黑名单取差集）。默认开启＝识别到纯文本就交给桥接、识别到多模态就等同白名单成员直发图片；声明永远读"包装前"的真值，不会被 `bridgeAutoImage` 的假声明污染。设为 `false` 可退回"只认名单"的纯手工行为 |
-| `probeResults` | `{}` | **实测结果（v0.9.0）**：`"provider/model" → "yes"`/`"no"`，由 `vision_probe_model` 工具写入，不要手改。实测优先级**高于声明**（真实请求 > 自称），但**低于** `multimodalModels`（人的明确意图有最终话语权） |
-| `bridgePreview` | `true` | 桥接图片内联预览：用户气泡内显示缩略图，点击放大 |
-| `bridgePreviewScanIntervalMs` | `2000` | 预览兜底扫描间隔（毫秒）；`0` 关闭兜底 |
-| `bridgePreviewHideHint` | `true` | 图片加载成功后隐藏桥接提示文本（失败时保留，安全降级） |
-| `bridgeAutoImage` | `true` | 桥接开启时向宿主准入检查报告**所有**模型都支持图片输入，纯文本模型也能直接粘贴图片，无需手动改 provider 配置 |
-| `sendSessionHeader` | `true` | 给视觉请求发送稳定的会话标识头。OpenCode Go 等网关要求每请求带 `x-opencode-session`（一个对话一个稳定 ID），缺失的请求在 2026-09-06 起可能报错 |
-| `sessionHeaderName` | `x-opencode-session` | 承载会话 ID 的头名 |
-| `sessionId` | `''` | 固定会话 ID（后台/无 dsh 会话调用时用）；留空则自动：优先取当前 dsh 会话 ID，否则用进程级稳定随机 ID |
-
-## 图片桥配置
-
-1. （一般不需要）只有关闭 `bridgeAutoImage` 时才需要手动给模型声明图片输入（pi-ai 风格），让 harness 放行图片消息：
-   ```yaml
-   llm-pi-ai:
-     providers:
-       your-provider:
-         models:
-           - id: deepseek-v4-flash
-             input: [text, image]
-   ```
-2. 在插件配置里列出真正多模态的模型，让它们直收图片块（名单模式见下一节）：
-   ```yaml
-   - id: tool-vision
-     name: 'dsh-tool-vision'
-     config:
-       multimodalListMode: whitelist   # 默认，名单内直发图片
-       multimodalModels: ['mimo-v2.5', 'grok-4.5']
-   ```
-
-之后在文本模型下贴图，转录里会留下一条指引：
-`[User sent an image, exported to: <path>. Inspect it with the inspect_image tool...]`
-（该消息不再以像素图形式渲染），Agent 会调用视觉端点查看并把结果带回对话。
-
-> 为什么不用 `llm/stream`？harness 会冻结每个请求，且 agent-loop invariant 会拒绝任何与会话日志推导不一致的请求；这个 cordis 版本的 waterfall `next()` 也无法替换请求参数。`agent/pre-step` 才是受支持的缝：它的决策消息**会成为**持久化日志，invariant 天然成立。
-
-密钥解析顺序：`config.apiKey` → `process.env[apiKeyEnv]` → `process.env.OPENAI_API_KEY`。
-
-## 桥接图片预览（v0.4.0）
-
-纯文本模型下，被桥接的图片在对话里只显示一段 `[User sent an image...]` 文本指引。开启 `bridgePreview`（默认开）后，浏览器端会在**展示层**把指引渲染成气泡内缩略图：
-
-- **缩略图 + 灯箱**：点击缩略图全屏放大，点击任意处或按 `Esc` 关闭；
-- **即时 + 兜底**：新消息由 MutationObserver 即时处理，历史消息由周期兜底扫描补齐（间隔见 `bridgePreviewScanIntervalMs`）；
-- **隐藏提示文本（P2）**：`bridgePreviewHideHint` 开启时，图片加载成功后桥接文本自动隐藏，气泡只留图片；加载失败则保留文本（安全降级，绝不出现"既无图又无字"）；
-- **识别机制**：桥接文本带不可见前缀标记（`\u200b[bridge]`），客户端据此精确识别桥接块——用户正常发言中出现"exported to:"字样不会被误伤；
-- **纯展示层红线**：不修改持久化消息、不修改转录、不修改模型侧文本、不碰 `inspect_image` 调用链。
-
-预览图片由同源回环路由 `/plugins/dsh-tool-vision/image` 提供，只读桥接导出目录、仅本机 Host、仅图片扩展名、单文件 ≤ 20MB、防目录穿越。
-
-## 工具：`inspect_image`
-
-| 参数 | 必填 | 含义 |
-|---|---|---|
-| `path` | ✅ | 图片路径（绝对路径，或相对当前工作区）或 http(s) URL |
-| `question` | – | 可选的具体问题 |
-| `detail` | – | `auto` / `low` / `high` 分辨率提示 |
-
-示例端点（`baseURL`）：
-
-- **OpenAI**：`https://api.openai.com/v1` —— `gpt-4o`、`gpt-4o-mini`
-- **阿里云 DashScope（Qwen-VL）**：`https://dashscope.aliyuncs.com/compatible-mode/v1` —— `qwen-vl-plus`、`qwen-vl-max`
-- **智谱（GLM-4V）**：`https://open.bigmodel.cn/api/paas/v4` —— `glm-4v-flash`（免费档）、`glm-4v-plus`
-- **Moonshot（Kimi）**：`https://api.moonshot.cn/v1` —— `moonshot-v1-8k-vision-preview`
-- **Ollama 本地**：`http://localhost:11434/v1` —— `llama3.2-vision`（无需密钥）
-
-## 像素级视觉工具(v0.6.0,移植自 dsh-vision-router)
-
-14 个 `vision_*` 工具由**同一个** `inspect_image` 配置的端点驱动
-(baseURL/apiKey/model)——无 provider 链、无本地模型、零新增配置:
-
-| 工具 | 用途 |
-|---|---|
-| `vision_describe` | 看图问答 / 多图对比(可选结构化 JSON) |
-| `vision_ground` | 定位目标,返回原图像素坐标框 |
-| `vision_detect` | 枚举元素(按钮/输入框/图标…),带编号框 |
-| `vision_crop` | 按像素区域裁剪出 PNG 产物 |
-| `vision_pixel_diff` | 逐像素对比:差异比例、最差区域、热图、报告 |
-| `vision_colors` | 主色量化,还原 UI 调色板 |
-| `vision_ocr` | 逐字转写文字(只读字,不做场景识别) |
-| `vision_long_screenshot_ocr` | 长截图分块转写为 Markdown |
-| `vision_trace` | potrace 矢量化输出彩色 SVG(worker 线程,安全) |
-| `vision_extract_foreground` | 纯色背景抠图 → 透明 PNG |
-| `vision_html_screenshot` | 本地 HTML 无头渲染截图(禁网) |
-| `vision_screenshot` | 桌面截屏(隐私门控:需在设置中开启 `desktopScreenshot`;Win: PowerShell / macOS: screencapture / Linux: import/scrot) |
-| `vision_present` | 通过宿主附件库把生成的图片正式展示给用户 |
-| `vision_materialize` | 把附件/本地图片落盘为工作区真实路径 |
-
-质量与安全细节:
-
-- **内容哈希缓存**按 端点+模型+图片+问题 取键(切模型不吃旧答案,失败结果
-  不入缓存);
-- **统一 4MP 降采样**后再调用模型;超大输入 stat 预检直接拒绝(文件与
-  附件路径统一 20MB 上限);
-- **限流/5xx 自动重试**(感知 Retry-After 退避);端点**内容安全拒绝**明确
-  返回 `VISION_CONTENT_FILTERED`,不再误报后端不可用;
-- **长截图 OCR 边界**:120s 总预算、40 块上限、取消检查、首块失败即停;
-- **路径 containment**(相对输入禁止逃逸工作区);产物写入
-  `<工作区>/.dsh-tool-vision/`。
-
-依赖 `sharp` / `potrace` / `puppeteer-core`(声明为**可选依赖**:平台安装失败
-不会阻断插件安装;缺失时懒加载降级并给出安装提示,不影响其他工具)。
-
-`vision_screenshot` 属于隐私敏感能力,**默认不注册**——在 tool-vision 设置中
-开启 `desktopScreenshot: true` 后才会注册桌面截屏工具。
-
-## v0.9.0：模型名单模式与自动识别
-
-桥接要回答一个问题：**当前这个模型能不能直接看图片？** v0.9.0 把它拆成两个相互独立的输入。
-
-```
-base = autoDetectMultimodal ? (路由声明含 image) : 空集
-off        → direct = base              名单不参与
-whitelist  → direct = base ∪ 名单        名单只做"加"
-blacklist  → direct = base \ 名单        名单只做"减"
-```
-
-**命中名单时以名单为准**：名单是用户明确的意图，优先级高于模型自己的声明，所以永远能纠正误判。
-
-**黑名单绝不会退化成"未列出的一律直发"**：黑名单的基准是自动识别集；没开自动识别时基准是空集，未列出的模型照样走桥接。这是刻意设计的——否则一次误配就能把图片硬塞给纯文本端点。
-
-**匹配规则**：`mimo-v2.5`、`xiaomi/mimo-v2.5`、`commandcode/xiaomi/mimo-v2.5` 指向同一个路由；`*` / `?` 为通配符；大小写不敏感。v0.8.1 只做字符串全等，README 自己举的 `mimo-v2.5` 例子在 `xiaomi/mimo-v2.5` 这类路由上其实**静默无效**，v0.9.0 修好了（只会多放行，不会收回任何旧配置已经放行的模型）。
-
-**面板**（插件 → dsh-tool-vision）：
-
-- 名单字段下面是一份**可直接勾选的模型清单**：候选来自 dsh 已配置的模型（`llm.listProviders()` + `llm.listModels()`），按 provider 分组，标注哪些**声明**支持图片；勾选加入、取消勾选移出，上方输入框仍可手打通配符（两者共用同一份草稿，改完按「保存」生效，未保存时清单头部会标"未保存"）；
-- 顶部常驻**当前路由判定**：`provider / model`、直发还是桥接、依据是什么（名单命中 / 自动识别 / 默认）。
-
-> 为什么不只用 `<datalist>`：原生 datalist **必须聚焦输入框并敲字才会弹出**，v0.9.0 第一版因此看起来像"面板没反应"。现在清单常驻可见，datalist 只作打字辅助。
->
-> 取消勾选移除的是**命中的那条记录本身**（名单里写 `mimo-v2.5` 命中了 `xiaomi/mimo-v2.5`，取消勾选就移除 `mimo-v2.5`，不会凭空虚增一个完整 id）。"哪条命中了"由服务端用与桥接**同一套匹配器**算出（payload 的 `matchedEntries`），客户端不复制规则，所以面板显示的状态永远不会和实际判定打架。
-
-**判定依据的读取路径**（这是本版最容易做错的地方）：`autoDetectMultimodal` **必须**读 `resolveModelInfo` 被包装之前的真值，否则 `bridgeAutoImage` 给所有模型贴上的"支持图片"就成了自证。代码里由 `unwrappedResolveModelInfo()` 保证，并有专门的回归测试。
-
-候选项由插件自己的回环路由提供：`GET /plugins/dsh-tool-vision/models`（仅本机 Host、只读、`no-store`），只返回 provider/model id 与声明能力，**不含任何密钥或端点地址**。它挂在插件主 fiber 而非总开关的子 fiber 上，所以插件关闭时面板依然可用。
-
-> ⚠️ `inputModalities` 是**声明**，不是保证——profile 里常见"为了过准入检查而写 `input: [text, image]`"的纯文本模型。上游 `dsh-llm-pi-ai` 自己也把"未声明"一律当作纯文本，理由写在源码注释里：**两种误判的代价不对等**——少声明会在贴图前就拒绝并点名模型，多声明则会放行一张"上游中途拒收、而消息已经落盘"的图片。
->
-> 所以自动识别是默认开启的，同时带三层兜底：① 某个路由**首次**因"仅凭自己的声明"被放行时，日志会打一条明确提示并点名怎么改；② 面板顶部常驻当前判定与依据；③ 把该模型写进 `multimodalModels` 并切到 `blacklist` 模式即可强制回到桥接。
-
-## 测试
-
-```bash
-npm test           # 服务端单测（不需要额外依赖）
-npm run test:render  # 面板渲染测试（需要 devDependencies）
-```
-
-`npm run test:render` 在 jsdom 里加载**真实客户端 bundle**、走**真实注册路径**（`apply` → `slots.register` → 组件），数据由**真实的服务端路由处理器**产出，最后断言真实 DOM 与真实的设置写入。
-
-它单独成命令、**不并入 `npm test`**：它需要 `react` / `react-dom` / `jsdom`，而一个"依赖缺失就静默跳过"的 DOM 测试只会带来虚假的安全感。需要时先 `npm i -D react@18 react-dom@18 jsdom`。
-
-它存在的理由很具体：v0.9.0 第一版把模型清单只渲染进原生 `<datalist>` —— 服务端单测**全绿**，面板却看起来完全没反应。这类 bug 在 DOM 之下根本抓不到。
-
-## 实测探测：`vision_probe_model`（v0.9.0）
-
-每次请求默认超时 60 秒、输出预算 2048 token。响应在流式读取时限制为 1 MiB，
-超限即停止读取并返回 `unknown`。鉴权、限流和服务器故障即使提到图片输入，
-也保持 `unknown`；错误响应仅在 HTTP 400、415、422 且明确拒绝图片时判为不支持。
-
-前面所有判定都建立在"模型**说自己**能不能看图"之上。这个工具改为**真的发一张图过去试**，是插件里唯一的地面真相。
-
-```
-base = autoDetect ? 路由声明含 image : ∅
-探测结果（若有）覆盖 base                  实测 > 自称
-名单命中（若有）覆盖一切                    人的意图 > 实测
-```
-
-**为什么一次请求不够**（都是实测踩出来的）：
-
-- **会瞎猜**：模型可能对橙色方块回答"blue"。所以探测跑**两个不同颜色**，两个都答对才算通过；
-- **推理模型会返回空 `content`**：思考预算不够时 `max_tokens` 全被烧光。默认给 2048，并回退读 `reasoning_content`；
-- **必须要有对照组**：先发一条纯文本请求确认"这条路本身是通的"，否则一个 401/超时会被误读成"它不能看图"——那会把一条好端端的多模态路由**永久**推回桥接；
-- **端点不报错 ≠ 能看图**。真实网关实测：`meituan/LongCat-2.0:free` 收到图片请求返回 200，但回答是 `"I can't see any image."`。只有"端点主动拒绝图片部分"才是确定性的负面信号。
-
-结论分三档：`yes`（对照组通过 + 两色全对）、`no`（端点拒绝图片，或答了但没读图）、`unknown`（网络/鉴权/协议问题——**绝不**当作能力结论）。
-
-**它怎么拿到主模型的端点与凭据**（不新增任何配置）：`llm.listConfigurableProviders()` 给出 provider 的 settings 命名空间与路径 → `settings.get(ns)` 读出 `baseURL`/`apiKeyEnv`/`api` → `credentials.resolve(apiKeyEnv)` 取出密钥（`.value`），与 `dsh-llm-pi-ai` 真实调用同一条路。只读，且**端点与密钥绝不出现在探测结果里**。
-
-**联动**：探测结果写入 `probeResults` 并**立即参与桥接判定**——`yes` 直接放行图片、`no` 强制回到桥接，且**与名单模式无关**。（这比"自动往名单里写一笔"更正确：名单在 `blacklist` 模式下含义是反的，自动写入会得到完全相反的效果。）面板清单里每个模型带徽章：
-
-- `已实测可看图` / `已实测不可看图`（蓝 / 红）——实测优先展示；
-- 没探测过的才回退显示声明的 `声明支持图片`；
-- 顶部"依据"栏显示 `实测（真实发图验证通过/未通过）`。
-
-> 未知协议的路由（如 `openai-responses`）会直接返回 `unknown` 并说明原因，而不是用 `chat/completions` 去猜一个自信的错误答案。
-
-## v0.9.3：适配器快照模型准入（pi-ai 门禁放行）
-
-**问题**。`v0.9.2` 的 `installDispatchImageAdmission` 成功满足了 DSH 核心层 `LlmService.generate` 的模态检查，将图片放行给底层适配器。但在以 `dsh-llm-pi-ai` 作为适配器时，其内部在 `streamWithSnapshot` 中执行了第二道门禁校验：
-
-```js
-const model = this.modelOf(snapshot, options.provider, options.model);
-if (containsImage && !model.input.includes("image"))
-  throw new LlmError(`pi-ai model "${model.id}" does not support image input`, "UNSUPPORTED_CONTENT");
-```
-
-`this.modelOf` 直接读取适配器自身维护的 `snapshot.models` 目录，若用户未在 `settings.yaml` 中显式声明 `input: [text, image]`，此处默认只有 `["text"]`。导致未声明的模型在放行图片后，在流式发起前被 `pi-ai` 适配器抛出 `UNSUPPORTED_CONTENT`，且由于该图片已被写进会话历史，后续所有轮次（哪怕发纯文本）都会彻底报错锁死。
-
-**修复**。在判定为直通路由（`direct: true`）时，`installDispatchImageAdmission` 增加了双重深度注入：
-1. 包装 `adapter.modelOf`（若存在），使其对直通路由返回携带 `"image"` 模态的模型对象；
-2. 动态向当前快照模型（`snapshot.models.getModel(route, model).input`）追加 `"image"` 模态；
-3. `dispose` 时完整还原 `adapter.modelOf` 并清理被追加的模态数组。
-
-## v0.9.2：准入不等于派发
-
-**问题**。插件收集的每一个能力信号 —— `probeResults`、`multimodalModels`、
-`autoDetectMultimodal` —— 都只决定**桥接**要不要拦截图片，没有一个决定模型**收不收得到**它。
-
-`LlmService.generate` 的能力来源是适配器，不是插件：
-
-```js
-const adapterCall = await adapter.prepareCall(provider, model, signal);
-modelInfo = this.normalizeModelInfo(registration, model, adapterCall.model);
-if (modelInfo.inputModalities !== undefined
-    && !modelInfo.inputModalities.includes("image")
-    && projectedMessages.some((message) => contentHasImage(message.content)))
-  projectedMessages = projectImagesForTextModel(projectedMessages);
-```
-
-`projectImagesForTextModel` 会在**调用适配器之前**把每个图像块改写成
-`[image omitted because this model accepts text only; attachment sha256:…]`。
-于是在一条插件已实测为「能读图」的路由上 —— 桥接因此让开放行 —— 图片依然被销毁，
-而销毁它的那个判断只读适配器的声明，不读任何别的东西。`resolveModelInfo` 那层 wrap
-（`bridgeAutoImage`）是**准入**：它决定谁可以「提交」图片，从不改变适配器实际流出的内容。
-插件里没有任何东西够到真正做决定的那一层。
-
-症状就是：一条探测结果为 `yes`、也列在 `multimodalModels` 里的路由，依然回
-`[image omitted …]`。
-
-**修复**。`installDispatchImageAdmission` 包装 `llm.registration` —— 也就是核心自己
-调用的那个访问器 —— 于是每个适配器（包括安装之后才注册的）都会经由桥接所用的**同一个**
-`routeDirectDecision` 来回答 `prepareCall`。一份优先级规则、两条缝，两者永不可能对同一
-条路由给出不同答案。判定为「走桥接」时行为不变：桥接早已把图片变成 `inspect_image`
-提示，核心的投影仍然是任何漏到派发层的图片的正确兜底。
-
-三条性质值得写下来，因为每一条都有测试：
-
-- **失败即保守**。判定无法作出时（配置抛错）保持调用原样，而不是把像素发给一条可能拒绝它的路由。
-- **可逆**。dispose 会还原它改过的每个适配器**以及**访问器本身；dispose 后再安装会重新包装。
-- **后注册的 provider 也算**。包装挂在访问器上而不是适配器快照上，所以安装之后注册的 provider 同样覆盖。
-
-`routeDirectDecision` 现在是这条优先级规则的唯一实现，桥接与派发共用 —— 两份实现必然
-漂移，而这份有三个输入。
-
-## v0.8.0：总开关，以及保存修复
-
-**总开关。** `enabled` 字段加设置栏顶部的一键按钮（`一键关闭` / `重新启用`）。
-注册本身就是绑定在「发起注册的 cordis fiber」上的副作用，所以插件把所有工具、
-图片桥、预览路由和图片能力声明都放进一个**子 fiber**：关掉开关就是销毁这个子
-fiber，15 个工具一并从模型侧消失。设置栏留在父 fiber 上，因此关掉之后还能再打开。
-无需重启 dsh。
-
-**保存修复。** 此前表单把 18 个字段作为**并行** `scope.set()/unset()` 提交。每个
-写入各自携带一个 revision 栅栏，而栅栏落后于宿主文档的写入会被 `settings/conflict`
-拒绝——**且被拒绝的写入仍然 resolve**（scope 的契约是「完成写入与恢复读取后结算」，
-不是「被拒就抛错」）。于是设置栏显示「已保存」，而编辑内容静默回退，看起来就是
-「完全存不了」。
-
-现在改为**一次原子 `mutate()`**，整批修改共用一个栅栏和一次持久化决策；写入结算后
-再回读命名空间section做校验，只有确认生效才报「已保存」，否则弹出「写入未生效」并
-重新载入表单。没有 `mutate()` 的旧宿主回退为**串行**写入（每个等前一个完成，revision
-链条依然正确），绝不并行。
-
-另外删掉了 3 处 `if (typeof scope.load === "function") scope.load()`：`SettingsScope` 接口
-从来没有 `load()`（完整 seam 只有 getSnapshot / subscribe / mutate / set / unset，读走的是
-共享 describe 镜像，由宿主 `settings/document-updated` 驱动刷新）。这些守卫是照臆测 API 写的
-死代码，读起来像"已经刷新过了"，反而掩盖了缺失的写入校验。
-
-## 限制
-
-- 从 `0.6.3` 起最低支持 DSH `0.1.0-rc.7`，已针对 `0.1.0-rc.7`、
-  `0.1.0-rc.8` 和 `0.1.1-rc.1` 测试。仍使用 DSH `0.1.0-rc.6` 的用户请锁定
-  `dsh-tool-vision@0.6.1`；这是最后一个包含旧 settings 白名单兼容补丁的版本。
-- 被桥接的图片以文本指引进入对话（转录而非像素）——文本模型无法做像素级上下文推理；视觉模型的描述通过 `inspect_image` 回传。
-- 桥接是**单向门**：文本模型下贴的图会在 `agent/pre-step` 被写进持久化日志，之后切到多模态模型也不会还原成图片块（反方向——多模态切到文本模型——会由 `repairLoggedImages` 自动补上桥接）。
-- 图片以 base64 传输；注意隐私与大小限制。
-- 独立于 dsh-llm 的路由/重试体系；失败会向 Agent 返回明确错误。
-
-## License
-
-MIT —— 桥接预览与整合:xing666173。像素级视觉工具移植自
-[dsh-vision-router](https://github.com/ysr666/dsh-vision-router)(© ysr666,MIT),
-在此致谢。
+[English](README.en.md)

package/index.js

@@ -975,11 +975,12 @@ async function routeDirectDecision(provider, model, config, llm) {
 /**
  * Whether the session's current model may receive image blocks directly.
  *
- * {@link routeDirectDecision} keyed by the session's route, with the answer
+ * {@link routeDirectDecision} keyed by the step's fixed route when available,
+ * otherwise by the session's route on older hosts, with the answer
  * recorded in {@link lastModelDecision} for the settings panel.
  */
-async function currentModelAcceptsImage(agent, config, llm) {
-  const { provider, model } = currentRoute(agent);
+async function currentModelAcceptsImage(agent, config, llm, modelSelection) {
+  const { provider, model } = modelSelection ?? currentRoute(agent);
   const decision = await routeDirectDecision(provider, model, config, llm);
   Object.assign(lastModelDecision, { provider, model, ...decision });
   return decision.direct;
@@ -1038,7 +1039,7 @@ function attachPreStepBridge(ctx, getConfig, exportDir) {
     const agent = payload?.agent;
     if (!agent?.session) return decision;
     try {
-      const acceptsImage = await currentModelAcceptsImage(agent, getConfig(), ctx.get("llm"));
+      const acceptsImage = await currentModelAcceptsImage(agent, getConfig(), ctx.get("llm"), payload.modelSelection);
       warnAutoPromotion(ctx.logger);
       if (!acceptsImage) {
         let repaired = repairedBySession.get(agent.session.id);

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "dsh-tool-vision",
-    "version": "0.10.4",
+    "version": "0.10.6",
     "description": "DeepSeek Harness 外置视觉模型插件：inspect_image 把本地图片或 http(s) 图片 URL 发给任意 OpenAI 兼容端点，视觉模型看图的文字回答直接带回对话；附 Web UI 设置栏。",
     "type": "module",
     "main": "index.js",
@@ -22,7 +22,7 @@
             "platform": "web"
         },
         "compatibility": {
-            "dsh": ">=0.2.0-rc.2 <0.3.0 || 0.2.1-rc.1 || 0.2.1-rc.2 || 0.2.1-rc.3",
+            "dsh": ">=0.2.0-rc.2 <0.3.0 || 0.2.1-rc.1 || 0.2.1-rc.2 || 0.2.1-rc.3 || 0.2.1-rc.4 || 0.2.1-rc.5 || 0.2.2-rc.1",
             "dshReleases": {
                 "0.1.5-rc.2": "incompatible",
                 "0.1.5-rc.3": "incompatible",
@@ -35,7 +35,10 @@
                 "0.2.0-rc.2": "compatible",
                 "0.2.1-rc.1": "compatible",
                 "0.2.1-rc.2": "compatible",
-                "0.2.1-rc.3": "compatible"
+                "0.2.1-rc.3": "compatible",
+                "0.2.1-rc.4": "compatible",
+                "0.2.1-rc.5": "compatible",
+                "0.2.2-rc.1": "compatible"
             },
             "node": ">=20.18",
             "profiles": [
@@ -71,7 +74,7 @@
     "license": "MIT",
     "engines": {
         "node": ">=20.18",
-        "dsh": ">=0.2.0-rc.2 <0.3.0 || 0.2.1-rc.1 || 0.2.1-rc.2 || 0.2.1-rc.3"
+        "dsh": ">=0.2.0-rc.2 <0.3.0 || 0.2.1-rc.1 || 0.2.1-rc.2 || 0.2.1-rc.3 || 0.2.1-rc.4 || 0.2.1-rc.5 || 0.2.2-rc.1"
     },
     "repository": {
         "type": "git",
@@ -84,8 +87,8 @@
         "@deepseek-ai/schemastery": "^3.18.4"
     },
     "peerDependencies": {
-        "@deepseek-ai/dsh-settings": ">=0.2.0-rc.2 <0.3.0 || 0.2.1-rc.1 || 0.2.1-rc.2 || 0.2.1-rc.3",
-        "@deepseek-ai/dsh-tools": ">=0.2.0-rc.2 <0.3.0 || 0.2.1-rc.1 || 0.2.1-rc.2 || 0.2.1-rc.3",
+        "@deepseek-ai/dsh-settings": ">=0.2.0-rc.2 <0.3.0 || 0.2.1-rc.1 || 0.2.1-rc.2 || 0.2.1-rc.3 || 0.2.1-rc.4 || 0.2.1-rc.5 || 0.2.2-rc.1",
+        "@deepseek-ai/dsh-tools": ">=0.2.0-rc.2 <0.3.0 || 0.2.1-rc.1 || 0.2.1-rc.2 || 0.2.1-rc.3 || 0.2.1-rc.4 || 0.2.1-rc.5 || 0.2.2-rc.1",
         "react": "^18.2.0"
     },
     "optionalDependencies": {
@@ -94,9 +97,9 @@
         "sharp": ">=0.35.3 <1"
     },
     "devDependencies": {
-        "@deepseek-ai/dsh": "0.2.1-rc.3",
-        "@deepseek-ai/dsh-settings": "0.2.1-rc.3",
-        "@deepseek-ai/dsh-tools": "0.2.1-rc.3",
+        "@deepseek-ai/dsh": "0.2.2-rc.1",
+        "@deepseek-ai/dsh-settings": "0.2.2-rc.1",
+        "@deepseek-ai/dsh-tools": "0.2.2-rc.1",
         "jsdom": "^30.0.1",
         "react": "^18.3.1",
         "react-dom": "^18.3.1"