dsh-vision-recognizer - npm Package Compare versions

Comparing version 0.1.1 to 0.1.2

client/client.js

@@ -48,2 +48,6 @@ window.__ModuleLoader__.load({ id: "dsh-vision-recognizer", factory: (require) => {
   saved: '已保存，立即生效',
+  test: '测试',
+  testing: '测试中…',
+  testOk: '连接正常',
+  testFailed: '测试失败',
   saveFailed: '保存失败',
@@ -81,2 +85,6 @@ loadFailed: '配置加载失败，请稍后重试',
   saved: 'Saved — takes effect immediately',
+  test: 'Test',
+  testing: 'Testing…',
+  testOk: 'Connection OK',
+  testFailed: 'Test failed',
   saveFailed: 'Save failed',
@@ -106,2 +114,3 @@ loadFailed: 'Failed to load configuration',
 .vrg-btn:disabled{opacity:.6;cursor:default}
+.vrg-btn-ghost{background:var(--dsw-alias-bg-layer-2,#f3f4f6);color:var(--dsw-alias-label-primary,#1f2328)}
 .vrg-status{font-size:12px}
@@ -136,2 +145,3 @@ .vrg-status.ok{color:var(--dsw-alias-state-success-primary,#16a34a)}
   const [saving, setSaving] = useState(false)
+  const [testing, setTesting] = useState(false)
   const [status, setStatus] = useState(null)
@@ -172,2 +182,32 @@
 
+  const runTest = useCallback(() => {
+    if (form === null || testing) return
+    setTesting(true)
+    setStatus(null)
+    fetch('/dsh-vision-recognizer/test', {
+      method: 'POST',
+      headers: { 'content-type': 'application/json' },
+      body: JSON.stringify({
+        provider: form.provider,
+        apiKey: form.apiKey,
+        model: form.model,
+        baseURL: form.baseURL,
+      }),
+    })
+      .then((res) => res.text().then((text) => {
+        let body = {}
+        try { body = text ? JSON.parse(text) : {} } catch { body = { error: text.slice(0, 200) } }
+        return body
+      }))
+      .then((body) => {
+        if (body.ok) {
+          setStatus({ kind: 'ok', message: t('testOk') + (typeof body.durationMs === 'number' ? ' · ' + body.durationMs + 'ms' : '') })
+        } else {
+          setStatus({ kind: 'err', message: t('testFailed') + (body.error ? ': ' + body.error : '') })
+        }
+      })
+      .catch((error) => setStatus({ kind: 'err', message: t('testFailed') + ': ' + String(error) }))
+      .finally(() => setTesting(false))
+  }, [form, testing, t])
+
   const save = useCallback(() => {
@@ -206,2 +246,5 @@ if (form === null || saving) return
           setStatus({ kind: 'ok', message: t('saved') })
+          // Auto-test the just-saved config so the user immediately sees
+          // whether the provider / key / model actually works.
+          runTest()
         } else {
@@ -213,3 +256,3 @@ setStatus({ kind: 'err', message: t('saveFailed') + (body.error ? ': ' + body.error : '') })
       .finally(() => setSaving(false))
-  }, [form, saving, t])
+  }, [form, saving, t, runTest])
 
@@ -302,2 +345,3 @@ if (loadError) return h('p', { className: 'vrg-intro' }, t('loadFailed'))
       h('button', { className: 'vrg-btn', disabled: saving, onClick: save }, saving ? t('saving') : t('save')),
+      h('button', { className: 'vrg-btn vrg-btn-ghost', disabled: testing, onClick: runTest }, testing ? t('testing') : t('test')),
       status !== null && h('span', { className: 'vrg-status ' + status.kind }, status.message)))
@@ -304,0 +348,0 @@ }

lib/core.js

@@ -19,2 +19,7 @@ /**
 
+/** Short prompt used by the API-key test request. */
+export const TEST_PROMPT = 'Reply with exactly: OK';
+/** A 32x32 solid-color PNG used by the API-key test (valid for every vision model, cheap). */
+export const TEST_IMAGE_BASE64 = 'iVBORw0KGgoAAAANSUhEUgAAACAAAAAgCAYAAABzenr0AAAAL0lEQVR4nO3OIQEAAAgDMBrShPpIiHEzMb/q2UsqAQEBAQEBAQEBAQEBAQGBdOABt9TMl+gurJYAAAAASUVORK5CYII=';
+
 /** The transcription prompt: a complete, faithful text rendition for a text-only recipient. */
@@ -73,2 +78,5 @@ export const TRANSCRIBE_PROMPT = `You are an image-to-text transcription service. The recipient is a text-only LLM that cannot see the image. Produce a faithful, complete text rendition covering:
     }
+    if (status === 400 && /image|invalid/i.test(text) && /length|width|dimension|size|larger than|smaller than|resolution/i.test(text)) {
+        return { kind: 'image_dimensions', hint: 'the image does not meet the model\'s dimension requirements — it is probably too small (each side must usually be larger than ~10px); use a normal screenshot or resize the image' };
+    }
     if (status === 400 && /context|length|too (large|long)|token/i.test(text)) {
@@ -118,3 +126,3 @@ return { kind: 'context_too_large', hint: 'input is too large for this model — try a smaller image or a longer-context model' };
 /** Build the OpenAI-compatible request for a transcription. */
-export function buildOpenAIRequest(attempt, base64, mediaType) {
+export function buildOpenAIRequest(attempt, base64, mediaType, prompt = TRANSCRIBE_PROMPT) {
     return {
@@ -133,3 +141,3 @@ url: `${attempt.baseURL.replace(/\/+$/, '')}/chat/completions`,
                     { type: 'image_url', image_url: { url: `data:${mediaType};base64,${base64}` } },
-                    { type: 'text', text: TRANSCRIBE_PROMPT },
+                    { type: 'text', text: prompt },
                 ],
@@ -142,3 +150,3 @@ }],
 /** Build the Anthropic Messages request for a transcription. */
-export function buildAnthropicRequest(attempt, base64, mediaType) {
+export function buildAnthropicRequest(attempt, base64, mediaType, prompt = TRANSCRIBE_PROMPT) {
     return {
@@ -159,3 +167,3 @@ url: `${attempt.baseURL.replace(/\/+$/, '')}/messages`,
                     { type: 'image', source: { type: 'base64', media_type: mediaType, data: base64 } },
-                    { type: 'text', text: TRANSCRIBE_PROMPT },
+                    { type: 'text', text: prompt },
                 ],
@@ -188,3 +196,3 @@ }],
 /** Send the transcription request for already-read image bytes; returns the text. */
-export async function transcribeRequest(attempt, data, mediaType, signal) {
+export async function transcribeRequest(attempt, data, mediaType, signal, prompt = TRANSCRIBE_PROMPT) {
     const base64 = Buffer.from(data).toString('base64');
@@ -196,3 +204,3 @@ const effectiveTimeout = isLocalBaseURL(attempt.baseURL)
     const parse = attempt.protocol === PROTOCOL_ANTHROPIC ? parseAnthropicResponse : parseOpenAIResponse;
-    const request = build(attempt, base64, mediaType);
+    const request = build(attempt, base64, mediaType, prompt);
     const post = () => fetch(request.url, {
@@ -232,2 +240,20 @@ method: 'POST',
 /**
+ * Verify a provider connection: sends a tiny valid image with a short prompt
+ * through the configured protocol and throws on any failure, so the Settings
+ * UI can report exactly why the key/endpoint/model does not work.
+ * @param attempt - resolved attempt (provider/model/baseURL/apiKey).
+ * @returns the resolved attempt on success.
+ */
+export async function testProvider(attempt) {
+    const apiKey = resolveApiKey(attempt);
+    if (apiKey === '' && !isLocalBaseURL(attempt.baseURL)) {
+        throw new Error('no API key configured — enter the key in Settings → Plugins → 识图, or export the provider env var');
+    }
+    const effective = { ...attempt, apiKey };
+    const data = Buffer.from(TEST_IMAGE_BASE64, 'base64');
+    await transcribeRequest(effective, data, 'image/png', undefined, TEST_PROMPT);
+    return effective;
+}
+
+/**
  * Fully resolve one config entry (primary or fallback) into an attempt object.
@@ -237,4 +263,3 @@ * @param entry - {provider?, model?, baseURL?, apiKey?} plus shared defaults.
  */
-export function resolveAttempt(entry, shared) {
-    const preset = providerById(entry.provider ?? 'custom');
+export function resolveAttempt(entry, shared) {    const preset = providerById(entry.provider ?? 'custom');
     const baseURL = (entry.baseURL && entry.baseURL !== '') ? entry.baseURL : preset.baseURL;
@@ -241,0 +266,0 @@ const model = (entry.model && entry.model !== '') ? entry.model : preset.defaultModel;

lib/index.js

@@ -157,17 +157,47 @@ /**
 
-    // Single exact route; the handler dispatches by HTTP method. The webserver
-    // rejects duplicate (kind, path) registrations, so registering GET and POST
-    // as two routes would throw and leave the POST handler dead.
-    return hostCtx.webServer.register({
-        kind: 'exact',
-        path: '/dsh-vision-recognizer/config',
-        handler: async (request, response) => {
-            if (request.method === 'GET') {
-                sendJson(response, 200, {
-                    providers: providerCatalog(),
-                    config: redactedConfig(state),
-                });
-                return;
-            }
-            if (request.method === 'POST') {
+    // The config route is a single exact route whose handler dispatches by HTTP
+    // method — the webserver rejects duplicate (kind, path) registrations, so
+    // registering GET and POST as two routes would throw and leave POST dead.
+    // The test route is a separate path and may be its own registration.
+    const disposers = [
+        hostCtx.webServer.register({
+            kind: 'exact',
+            path: '/dsh-vision-recognizer/config',
+            handler: async (request, response) => {
+                if (request.method === 'GET') {
+                    sendJson(response, 200, {
+                        providers: providerCatalog(),
+                        config: redactedConfig(state),
+                    });
+                    return;
+                }
+                if (request.method === 'POST') {
+                    if (!sameOrigin(request)) {
+                        sendJson(response, 403, { error: 'untrusted origin' });
+                        return;
+                    }
+                    try {
+                        const body = await readJsonBody(request);
+                        const patch = core.sanitizePatch(body);
+                        const next = persistConfig({ ...state.config, ...patch });
+                        state.config = next;
+                        sendJson(response, 200, { ok: true, config: redactedConfig(state) });
+                    } catch (error) {
+                        sendJson(response, 400, { ok: false, error: error instanceof Error ? error.message : String(error) });
+                    }
+                    return;
+                }
+                response.writeHead(405, { allow: 'GET, POST' });
+                response.end();
+            },
+        }),
+        hostCtx.webServer.register({
+            kind: 'exact',
+            path: '/dsh-vision-recognizer/test',
+            handler: async (request, response) => {
+                if (request.method !== 'POST') {
+                    response.writeHead(405, { allow: 'POST' });
+                    response.end();
+                    return;
+                }
                 if (!sameOrigin(request)) {
@@ -179,15 +209,28 @@ sendJson(response, 403, { error: 'untrusted origin' });
                     const body = await readJsonBody(request);
-                    const patch = core.sanitizePatch(body);
-                    const next = persistConfig({ ...state.config, ...patch });
-                    state.config = next;
-                    sendJson(response, 200, { ok: true, config: redactedConfig(state) });
+                    const c = state.config;
+                    const candidate = {
+                        provider: typeof body.provider === 'string' && body.provider !== '' ? body.provider : c.provider,
+                        model: typeof body.model === 'string' ? body.model : c.model,
+                        baseURL: typeof body.baseURL === 'string' ? body.baseURL : c.baseURL,
+                        apiKey: typeof body.apiKey === 'string' && body.apiKey !== '' ? body.apiKey : c.apiKey,
+                    };
+                    const shared = { maxTokens: 32, timeoutMs: Math.min(c.timeoutMs ?? DEFAULT_TIMEOUT_MS, 30_000) };
+                    const attempt = core.resolveAttempt(candidate, shared);
+                    const started = Date.now();
+                    await core.testProvider(attempt);
+                    sendJson(response, 200, {
+                        ok: true,
+                        durationMs: Date.now() - started,
+                        model: attempt.model,
+                        baseURL: attempt.baseURL,
+                    });
                 } catch (error) {
-                    sendJson(response, 400, { ok: false, error: error instanceof Error ? error.message : String(error) });
+                    sendJson(response, 200, { ok: false, error: error instanceof Error ? error.message : String(error) });
                 }
-                return;
-            }
-            response.writeHead(405, { allow: 'GET, POST' });
-            response.end();
-        },
-    });
+            },
+        }),
+    ];
+    return () => {
+        for (const dispose of disposers) dispose();
+    };
 }
@@ -194,0 +237,0 @@

package.json

 {
   "name": "dsh-vision-recognizer",
-  "version": "0.1.1",
+  "version": "0.1.2",
   "description": "DeepSeek Harness image recognition plugin: keep DeepSeek as the conversation brain while a configurable vision model (15+ providers, OpenAI-compatible + Anthropic) transcribes attached images. Configurable from Settings → Plugins.",
@@ -5,0 +5,0 @@ "private": false,

New alerts

Long strings

Supply chain risk

Contains long string literals, which may be a sign of obfuscated or packed code.

Fixed alerts

Long strings

Supply chain risk

Contains long string literals, which may be a sign of obfuscated or packed code.

Improved metrics

Total package byte prevSize

74696

8.15%

Lines of code

1189

10.09%

Worsened metrics

Number of medium supply chain risk alerts

5

25%

No dependency changes