Fix broken chat: slim Exact prompt and restore JSON discipline.

Stop duplicating Exact into live context, tighten preview URL matching, force write_prompt on "поправь", and gate critique behind vision + a real JSON fence.

Co-authored-by: Cursor <cursoragent@cursor.com>

Assets/assistent.js

@@ -1,6 +1,6 @@
 /**
  * Swarm Assistent — Krea 2 collaborative chat (Ollama via Swarm API).
- * v0.7.1: Exact KV memory; block SwarmUI ViewSpecial/model-card previews from Generate+vision.
+ * v0.7.2: Slim Exact in prompt; no Exact blob in live ctx; tighter preview filter; critique requires vision+JSON.
  */
 (function () {
     const LS_BASE = 'swarm_assistent_base_url';
@@ -1681,10 +1681,10 @@
             };
         }
 
-        ctx.exact = state.exact || state.config?.exact || null;
         ctx.session_exact = state.sessionExact && Object.keys(state.sessionExact).length
             ? { ...state.sessionExact }
             : {};
+        // Exact KV is already in the system prompt — do not duplicate the full blob into live context.
 
         return ctx;
     }
@@ -2088,6 +2088,9 @@
         if (!t.trim()) {
             return null;
         }
+        if (/\b(поправь|исправь|перепиши|улучши|fix\s+it|make\s+it\s+better|rewrite)\b/i.test(t)) {
+            return 'write_prompt';
+        }
         if (/\b(опиши\s+реф|опиши\s+изображ|prompt\s+from\s+image|describe\s+(this|the|ref|image)|reverse\s*prompt)\b/i.test(t)
             || /опиши\s+(этот|эту|картинк|референс)/i.test(t)) {
             return 'describe_ref';
@@ -2846,17 +2849,14 @@
         if (!s) {
             return false;
         }
-        // SwarmUI checkpoint/LoRA cards: ViewSpecial/*, View/Models/*, *.preview.*, PLACEHOLDER.
+        // Only SwarmUI checkpoint/LoRA card routes — not bare "/models/" (matches Civitai page URLs).
-        if (s.includes('.preview.')
+        return s.includes('.preview.')
             || s.includes('placeholder')
             || s.includes('/viewspecial/')
             || s.includes('viewspecial/')
-            || s.includes('/models/')
+            || s.includes('/view/models/')
-            || s.includes('view/models/')
+            || /\/view\/models\//.test(s)
-            || /[?&](?:path|file)=[^&]*\.preview\./i.test(s)) {
+            || /[?&](?:path|file)=[^&]*\.preview\./i.test(s);
-            return true;
-        }
-        return false;
     }
 
     function findCurrentGenerateSrc({ allowPreview = false } = {}) {

AssistentConfig.cs

@@ -315,6 +315,46 @@ public sealed class AssistentConfig
         return def;
     }
 
+    /// <summary>Exact KV for the system prompt: params tables only (facts stay short / in RAG).</summary>
+    public JObject LoadExactForPrompt(string personaId)
+    {
+        JObject full = LoadExact(personaId);
+        if (full is null || full.Count == 0)
+        {
+            return full ?? new JObject();
+        }
+        JObject slim = new();
+        foreach (string key in new[] { "generation", "profiles", "aspect_table" })
+        {
+            if (full[key] is not null)
+            {
+                slim[key] = full[key].DeepClone();
+            }
+        }
+        if (full["facts"] is JObject facts)
+        {
+            // Keep short pointers only — long prose bloats 7B context and kills JSON discipline.
+            JObject shortFacts = new();
+            foreach (JProperty prop in facts.Properties())
+            {
+                string text = prop.Value?.ToString() ?? "";
+                if (text.Length <= 160)
+                {
+                    shortFacts[prop.Name] = text;
+                }
+                else
+                {
+                    shortFacts[prop.Name] = text.Substring(0, 157) + "…";
+                }
+            }
+            if (shortFacts.Count > 0)
+            {
+                slim["facts"] = shortFacts;
+            }
+        }
+        return slim;
+    }
+
     public JObject LoadAssistant(string personaId) => MergeJsonLayers("assistant.json", LayerRoots(personaId));
 
     public JObject LoadUi(string personaId) => MergeJsonLayers("ui.json", LayerRoots(personaId));

Config/_base/core/core.md

@@ -16,14 +16,17 @@ When instructions conflict, apply this order (highest wins):
 
 Exact = encyclopedia of defaults. RAG = soft notes. Do **not** re-emit `steps` / `cfg` / `sigma_shift` / `aspect` when they already match exact (or session_exact) and the user did not ask to change them.
 
+Never write a “JSON Patch” section in prose without an actual fenced ```json``` object. If you propose UI changes, the fence is mandatory. Keep the prose short (a few lines) — do not paste long critique templates.
+
 ## Live context
 
 A JSON block named "Live SwarmUI context" is attached. Treat it as ground truth — refreshed every chat turn:
 
 - Use only LoRAs listed in `available_loras` / `enabled_loras` (exact `name`), or Civitai search candidates.
 - Prefer listed `trigger_phrase` / `triggers` — **never invent** trigger words.
-- `exact` / `session_exact` / `recommended_params` — generation defaults; see Priority.
+- `session_exact` / `recommended_params` — session overrides and defaults (Exact KV is in the system block above).
-- `memory_hits` are retrieved notes (LoRA tips, pitfalls). Trust them over guesses, but **not** over exact or the user.
+- `memory_hits` are retrieved notes (LoRA tips, pitfalls). Trust them over guesses, but **not** over Exact or the user.
+- `has_vision_image` — if false, do not invent what the image looks like; emit `look_at` first when you need to see it.
 - `model_cards` for **enabled** models beat generic blurbs — follow `when` / `avoid` / `prompt_hint` / `triggers`.
 - `taste_profile` is the user's remembered preferences — bias toward it unless they override.
 - Prefer `krea_likely` / Krea architecture entries; ignore FLUX/SDXL LoRAs.

Config/_base/packs/critique_image.md

@@ -1,28 +1,20 @@
 # Mode: critique_image
 
-Goal: look at the attached / requested board image(s) and improve the next generation for **Krea 2**.
+Goal: look at the attached board image(s) and improve the next **Krea 2** generation.
 
-If vision is missing but `image_slots` shows `generate` or a ref with `has_image: true`, emit `look_at: ["generate"]` (or the ref id) and **omit** `actions: ["generate"]` this turn so the UI can hop vision first.
+## Vision gate (mandatory)
 
-## Critique checklist
+- If `has_vision_image` is false: emit **only** a one-line note + JSON with `look_at: ["generate"]` (or the ref id). Do **not** write a critique checklist. Do **not** invent defects.
+- If vision is present: short critique, then a real fenced JSON patch.
 
-- Subject, composition, lighting — what works / what fails.
+## Critique (keep short)
-- **Anatomy** (hands, limbs, face).
-- **Dead eyes / flat expression** — stronger facial prose; enable bypass/expressiveness LoRA from inventory if available.
-- **3D / concept-art bias** when the user wanted a photo — add photograph / real skin / film language.
-- **Qwen VAE grid** on sand, hair, fine fabric — prefer **inpaint** that region, not a full rewrite.
-- **Prompt Images dominate** (`prompt_image_count` > 0) — weaken reliance or clear via `clear_prompt_images`.
-- Aspect too tight/wide — set `aspect` or width/height.
-- Missing / wrong LoRA triggers or weights.
 
-## Fix strategy
+Bullet the real issues you see (anatomy, eyes, lighting, aspect, LoRA triggers). Do not paste a generic template of pitfalls you did not observe.
-
-- **Local defect** (hands, face, object): inpaint (`use_init_image` + mask). No mask → ask to paint / pack `inpaint_edit`.
-- **Global restyle:** img2img with `init_creativity` ≈ 0.4–0.6.
-- **From-scratch rewrite:** only when composition is wrong.
 
 ## Deliverable
 
-- Short critique in the user's language.
+1. 2–6 short lines in the user's language.
-- JSON patch with improved `prompt` and any `loras` / size / init tweaks.
+2. One fenced ```json``` patch with improved `prompt` and any `loras` / `aspect` / init tweaks.
-- `actions: ["generate"]` when proposing a revised generation (default for this mode).
+3. `actions: ["generate"]` when proposing a revised generation.
+
+Never describe a patch in prose without the fenced JSON object.

README.md

@@ -2,7 +2,7 @@
 
 SwarmUI extension for **collaborative Krea 2** prompting via **Ollama**: chat + board (Generate | Refs tabs), LoRA chips, **persona presets** (`Config/personas/`), **vector memory**, model cards with Civitai fetch, img2img/inpaint, slash commands, auto Generate.
 
-**Version 0.7.1** — Exact KV memory + persona overlays; block SwarmUI `ViewSpecial`/model-card previews from Generate + vision.
+**Version 0.7.2** — Exact KV (slim in prompt); block `ViewSpecial` model-card previews; critique requires vision + real JSON fence.
 
 ## Layout
 

SwarmAssistentExtension.cs

@@ -49,7 +49,7 @@ public class SwarmAssistentExtension : Extension
         ExtensionAuthor = "mrleo1nid";
         Description = "Collaborative Krea 2 assistant: Ollama chat, persona presets, vector memory, model cards, Generate loop.";
         License = "MIT";
-        Version = "0.7.1";
+        Version = "0.7.2";
         Tags = ["tabs", "ui", "llm", "ollama", "krea", "inpaint", "memory"];
     }
 
@@ -1347,11 +1347,11 @@ public class SwarmAssistentExtension : Extension
             }
         }
 
-        JObject exact = Config.LoadExact(pid);
+        JObject exact = Config.LoadExactForPrompt(pid);
         if (exact is not null && exact.Count > 0)
         {
             system.AppendLine();
-            system.AppendLine("## Exact memory (canonical KV — always trust over RAG guesses)");
+            system.AppendLine("## Exact memory (canonical KV defaults — prefer over RAG for numbers)");
             system.AppendLine("```json");
             system.AppendLine(exact.ToString(Newtonsoft.Json.Formatting.None));
             system.AppendLine("```");
@@ -1574,7 +1574,7 @@ public class SwarmAssistentExtension : Extension
             Logs.Debug($"Assistent memory retrieve: {ex.Message}");
         }
 
-        string enrichedContext = InjectMemoryHits(contextJson, hits, Config.LoadExact(pid));
+        string enrichedContext = InjectMemoryHits(contextJson, hits);
         List<JObject> messages = BuildOllamaMessages(packName, includeBase, enrichedContext, userMessages, personaId: pid, skillIds: skills);
         JArray civitaiResults = [];
         string reply = "";
@@ -1681,29 +1681,12 @@ public class SwarmAssistentExtension : Extension
             ctx = new JObject { ["_raw_context"] = contextJson };
         }
         ctx["memory_hits"] = hits ?? new JArray();
-        if (exact is not null && exact.Count > 0 && ctx["exact"] is null)
+        // Never re-inject full Exact into live context (already in system prompt).
-        {
+        ctx.Remove("exact");
-            ctx["exact"] = exact;
-        }
         if (ctx["session_exact"] is null)
         {
             ctx["session_exact"] = new JObject();
         }
-        if (ctx["recommended_params"] is null && exact?["generation"] is JObject gen)
-        {
-            JObject rec = new();
-            foreach (string key in new[] { "steps", "cfg", "sigma_shift", "aspect", "images" })
-            {
-                if (gen[key] is not null)
-                {
-                    rec[key] = gen[key].DeepClone();
-                }
-            }
-            if (rec.Count > 0)
-            {
-                ctx["recommended_params"] = rec;
-            }
-        }
         // Slim inventory for LLM: keep enabled + current, drop full dump if present
         if (ctx["available_loras"] is JArray allLoras && allLoras.Count > 24)
         {