@goodandready/dsh-image-gen 0.11.2 → 0.11.4

package/lib/attachment-helper.js

@@ -81,10 +81,15 @@ export async function saveAndAttachResult(ctx, exec, cfg, {
   })
 }
 
+/** Raster media types safe for multimodal LLM context (DeepSeek, OpenAI, Claude, Gemini). */
+const SAFE_IMAGE_TYPES = new Set(['image/png', 'image/jpeg', 'image/webp', 'image/gif'])
+
 export function renderToolOutput(value) {
   const text = (value && value.summary) || (typeof value === "string" ? value : JSON.stringify(value || ""))
   const blocks = [{ type: "text", text }]
-  if (value && value.attachment) {
+  if (value && value.attachment
+      && value.attachment.attachmentId
+      && SAFE_IMAGE_TYPES.has(value.attachment.mediaType)) {
     blocks.push({ type: "image", attachment: value.attachment })
   }
   return blocks

package/lib/index.js

@@ -409,57 +409,43 @@ function migrateLegacySettings(sctx, scope) {
   }
 }
 
-/** Collect text chunks from llm.stream iterator. */
-export async function collectText(iterable) {
-  let out = ''
-  let sawDelta = false
-  for await (const chunk of iterable) {
-    if (chunk && chunk.type === 'text-delta' && typeof chunk.text === 'string') {
-      out += chunk.text
-      sawDelta = true
-    } else if (
-      !sawDelta && chunk && chunk.type === 'block-end'
-      && chunk.block && chunk.block.type === 'text' && typeof chunk.block.text === 'string'
-    ) {
-      out += chunk.block.text
+/** Collect text chunks from llm.stream iterator.
+ *  Canonical implementation lives in prompt-enhancer.js; re-exported for backwards compat. */
+export { collectText } from './prompt-enhancer.js'
+
+
+/** Expand short prompt via chat model; return original prompt on error.
+ *  Canonical implementation lives in prompt-enhancer.js; re-exported for backwards compat. */
+export { buildEnhancePromptSystemMessage } from './prompt-enhancer.js'
+
+
+/** Expand short prompt via chat model; return original prompt on error.
+ *  Canonical implementation lives in prompt-enhancer.js; re-exported for backwards compat. */
+export { enhancePrompt } from './prompt-enhancer.js'
+
+
+function sanitizeRaw(raw) {
+  if (!raw || typeof raw !== "object" || Array.isArray(raw)) return {}
+  const cleaned = {}
+  for (const [k, v] of Object.entries(raw)) {
+    if (v && typeof v === "object" && !Array.isArray(v) && typeof v.get !== "function" && Object.keys(v).length === 0) {
+      continue
     }
+    cleaned[k] = v
   }
-  return out.trim()
+  return cleaned
 }
 
-/** Expand short prompt via chat model; return original prompt on error. */
-export function buildEnhancePromptSystemMessage(provider, model) {
-  const isFlux = String(model || '').toLowerCase().includes('flux') || provider === 'fal'
-  if (isFlux) {
-    return 'You are an expert prompt engineer for FLUX image models. Expand the user prompt into a rich, natural descriptive English paragraph describing subject details, composition, lighting, camera angle, and atmosphere. Reply with ONLY the expanded prompt, no commentary, no markdown quotes.'
-  }
-  return 'You are an expert prompt engineer for Stable Diffusion models. Expand the user prompt into detailed comma-separated descriptive visual tags including subject, composition, studio lighting, materials, and artistic medium. Reply with ONLY the expanded prompt, no commentary.'
+export function ensureConfig(raw) {
+  const cleaned = sanitizeRaw(raw)
+  const plain = plainConfig(cleaned)
+  return Config(plain)
 }
 
-/** Expand short prompt via chat model; return original prompt on error. */
-export async function enhancePrompt(ctx, cfg, prompt, signal, provider) {
-  if (!cfg.enhancePrompt) return { prompt, enhanced: false }
-  if (String(prompt).length >= (cfg.enhanceBelowChars || 200)) return { prompt, enhanced: false }
-  try {
-    const sysMsg = buildEnhancePromptSystemMessage(provider || cfg.provider, cfg.enhanceModel || cfg.model)
-    const chunks = ctx.llm.stream({
-      ...(signal ? { signal } : {}),
-      ...(cfg.enhanceModel ? { model: cfg.enhanceModel } : {}),
-      messages: [
-        { role: 'system', content: sysMsg },
-        { role: 'user', content: prompt },
-      ],
-    })
-    const text = await collectText(chunks)
-    if (!text) return { prompt, enhanced: false }
-    return { prompt: text, enhanced: true }
-  } catch (_) {
-    return { prompt, enhanced: false }
-  }
-}
 
 export function apply(ctx, rawConfig) {
-  const config = Config(rawConfig ?? {})
+  const config = ensureConfig(rawConfig)
+  const baseConfig = plainConfig(config)
   if (ctx.systemPrompt && typeof ctx.systemPrompt.section === 'function') {
     ctx.systemPrompt.section({
       name: 'tool:image-generation',
@@ -474,7 +460,7 @@ export function apply(ctx, rawConfig) {
   let settingsApi
   let currentLiveConfig = null
   let getConfig = () => currentLiveConfig || config
-  const live = () => Config(structuredClone(plainConfig(getConfig() ?? {}))) ?? config
+  const live = () => plainConfig(getConfig() ?? {})
 
   const createSettingsAdapter = (svc) => {
     if (!svc) return undefined
@@ -691,7 +677,7 @@ export function apply(ctx, rawConfig) {
 
   // Tool registrations live in register-tools.js; each tool is a labeled ctx.effect (#216).
   registerAllTools(ctx, {
-    config,
+    config: baseConfig,
     live,
     saveAndAttachResult,
     resolveSource,

package/lib/prompt-enhancer.js

@@ -0,0 +1,51 @@
+// prompt-enhancer.js — Prompt enhancement via LLM (#4 fix)
+// Extracted from index.js to avoid circular dependency with tools/generation.js
+
+/** Collect text chunks from llm.stream iterator. */
+export async function collectText(iterable) {
+  let out = ''
+  let sawDelta = false
+  for await (const chunk of iterable) {
+    if (chunk && chunk.type === 'text-delta' && typeof chunk.text === 'string') {
+      out += chunk.text
+      sawDelta = true
+    } else if (
+      !sawDelta && chunk && chunk.type === 'block-end'
+      && chunk.block && chunk.block.type === 'text' && typeof chunk.block.text === 'string'
+    ) {
+      out += chunk.block.text
+    }
+  }
+  return out.trim()
+}
+
+/** Build system message for prompt enhancement. */
+export function buildEnhancePromptSystemMessage(provider, model) {
+  const isFlux = String(model || '').toLowerCase().includes('flux') || provider === 'fal'
+  if (isFlux) {
+    return 'You are an expert prompt engineer for FLUX image models. Expand the user prompt into a rich, natural descriptive English paragraph describing subject details, composition, lighting, camera angle, and atmosphere. Reply with ONLY the expanded prompt, no commentary, no markdown quotes.'
+  }
+  return 'You are an expert prompt engineer for Stable Diffusion models. Expand the user prompt into detailed comma-separated descriptive visual tags including subject, composition, studio lighting, materials, and artistic medium. Reply with ONLY the expanded prompt, no commentary.'
+}
+
+/** Expand short prompt via chat model; return original prompt on error. */
+export async function enhancePrompt(ctx, cfg, prompt, signal, provider) {
+  if (!cfg.enhancePrompt) return { prompt, enhanced: false }
+  if (String(prompt).length >= (cfg.enhanceBelowChars || 200)) return { prompt, enhanced: false }
+  try {
+    const sysMsg = buildEnhancePromptSystemMessage(provider || cfg.provider, cfg.enhanceModel || cfg.model)
+    const chunks = ctx.llm.stream({
+      ...(signal ? { signal } : {}),
+      ...(cfg.enhanceModel ? { model: cfg.enhanceModel } : {}),
+      messages: [
+        { role: 'system', content: sysMsg },
+        { role: 'user', content: prompt },
+      ],
+    })
+    const text = await collectText(chunks)
+    if (!text) return { prompt, enhanced: false }
+    return { prompt: text, enhanced: true }
+  } catch (_) {
+    return { prompt, enhanced: false }
+  }
+}

package/lib/providers.js

@@ -128,9 +128,16 @@ export async function pixelDiff(a, b) {
 
 /** Extract detailed ComfyUI node error messages from /history/{pid} response. */
 
-/** Safe attachment persistence with fallback when ctx.attachments is unavailable. */
+/** Raster media types that DSH attachment store (Sharp) can process. */
+const RASTER_IMAGE_TYPES = new Set(['image/png', 'image/jpeg', 'image/webp', 'image/gif'])
+
+/** Safe attachment persistence with fallback when ctx.attachments is unavailable.
+ *  SVG and other non-raster types are never sent to saveImage (Sharp rejects them)
+ *  and never emitted as image content blocks (LLMs reject them). */
 export async function saveAttachmentSafe(ctx, { bytes, mediaType, name }) {
-  if (ctx && ctx.attachments && typeof ctx.attachments.saveImage === 'function') {
+  // Only attempt saveImage for raster formats that Sharp/LLMs can handle
+  if (RASTER_IMAGE_TYPES.has(mediaType)
+      && ctx && ctx.attachments && typeof ctx.attachments.saveImage === 'function') {
     try {
       const att = await ctx.attachments.saveImage({
         data: new Uint8Array(bytes),

package/lib/tools/processing-basic.js

@@ -198,16 +198,12 @@ export function registerProcessingTools(ctx, deps) {
             const filePath = path.join(outDir, name)
             await writeFile(filePath, res.bytes)
 
-            let attachment = null
-            let localUrl = ''
-            if (ctx.attachments?.saveImage) {
-              attachment = await ctx.attachments.saveImage({
-                data: new Uint8Array(res.bytes),
-                mediaType: 'image/svg+xml',
-                name,
-              })
-              localUrl = '/dsh-image-gen/image?id=' + encodeURIComponent(attachment.attachmentId)
-            }
+            // SVG is not a raster format — do not create image attachment (#5)
+            // saveAttachmentSafe already guards against SVG, but we skip entirely
+            // to avoid even a fallback stub that could confuse renderToolOutput
+            const attachment = null
+            const localUrl = ''
+
 
             const summary = `### Vectorized SVG Export\n- **File**: ${filePath}\n- **Format**: SVG (XML vector)\n- **Color mode**: ${args.color_mode || 'color'}\n- **Palette size**: ${res.palette ? res.palette.length : 16} colors\n`
             return toLosslessJson({

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@goodandready/dsh-image-gen",
-  "version": "0.11.2",
+  "version": "0.11.4",
   "description": "Image generation for DeepSeek Harness: a generate_image tool with pluggable providers — the FAL queue, any OpenAI-compatible images API, or a ChatGPT/Grok subscription with no API key at all. The picture is shown inline in the conversation; the model receives either a link (works with any chat model) or the image itself (needs dsh-vision-bridge or a vision-capable model).",
   "keywords": [
     "deepseek-harness",