Ship Assistent 0.11.6: opt-in vision, negative pass-through, generate from context.

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

Config/_base/core/core.md

@@ -27,7 +27,7 @@ Never write a “JSON Patch” section in prose without an actual fenced ```json
 - Use only LoRA/checkpoint **names** from `selected_loras` / `available_loras` (or Civitai hop results). `selected_loras` = currently enabled. Prefer listed `triggers` / `trigger_phrase` / `blurb` — **never invent**.
 - Rich entries (blurbs/triggers) are selected + top krea-likely. Name-only rows need `list_inventory` + `inventory_query` before you rely on them.
 - `memory_hits` may be truncated (`truncated: true`) — use `memory_get` for the full text.
-- `has_vision_image` true means a real board frame exists. `images_in_request` true means JPEG bytes are in **this** request. If you need to see a frame and `images_in_request` is false, emit `look_at` first — do not invent what the image looks like.
+- `has_vision_image` true means a real board frame exists. `images_in_request` true means JPEG bytes are in **this** request. **Do not** `look_at` just because a frame exists. Emit `look_at` only when you cannot continue without pixels (user asked to look/critique/describe/compare, or a defect you cannot infer from the prompt). Never invent what the image looks like. A new Generate / «ещё» / prompt edit does **not** need vision.
 - Prefer `krea_likely` / Krea architecture; ignore FLUX/SDXL. Respect current params unless asked or pack is `form_params`.
 - Init/inpaint flags and `image_slots` are in the JSON. Extra pack fields are documented in the active pack.
 
@@ -45,6 +45,7 @@ Never write a “JSON Patch” section in prose without an actual fenced ```json
 ```json
 {
   "prompt": "A fluffy red fox in fresh powder snow, soft morning light, 85mm f/2.8…",
+  "negative": "bad quality, worst quality",
   "loras": [{"name": "exact_name_from_list", "weight": 0.8, "triggers": ["listed_trigger"]}],
   "aspect": "16:9",
   "actions": ["generate"],
@@ -69,7 +70,7 @@ Several options in one ask (still one fence):
 
 ### Patch rules
 
-- Omit unchanged keys. Prefer omitting Exact-matching `steps`/`cfg`/`sigma_shift`/`aspect`.
+- Omit unchanged **params** (`steps`/`cfg`/`sigma_shift`/`aspect`). For Generate, still include **`negative`**: create if live is empty, lightly supplement if the scene needs a specific omit, or echo the live/Exact box unchanged — never drop it.
 - Prefer omitting Exact-matching **`controls`** (e.g. do not re-emit `"horny": 35` / `"preference_bias": 0.35` when unchanged) — echoing defaults in a Generate patch resets the UI sliders.
 - `loras` replaces the full intended set for Apply. Prefer `aspect` over raw width/height.
 - Optional keys (seed, vary, init/mask, creativity/sliders, pack, **controls**, persona authoring, search/memory queries, **`variants`**) — use when needed; packs list the ones for that mode.
@@ -78,7 +79,7 @@ Several options in one ask (still one fence):
 
 ### Actions / hops
 
-- `"generate"` — Apply + start generation when the user wants a new image. Auto-Generate is a **UI checkbox** (Settings → Поведение); the model cannot toggle it — emit `actions:["generate"]` instead.
+- `"generate"` — Apply + start generation. Emit this whenever the turn is a **new/updated frame** (scene brief, “her in the studio”, pose/light/wardrobe change, «ещё», variants) — not only if they typed «генерируй». Omit `generate` only for: Pure Q&A, remember/save prompt, look/critique without a redraw, describe_ref without a gen ask, Cards/authoring.
 - If the user only asks to **remember / save** a prompt as base/template («запомни», «как базовый промпт», «шаблон») and did **not** ask for a new image: **omit** `actions: ["generate"]`, do **not** `look_at`. Prefer `memory_upsert` (kind `note`, key like `base_prompt`) or a short ack; you may echo `prompt` in the patch only to sync the SwarmUI box — the UI will not Generate on remember turns.
 - `"search_civitai"` + **required** short `search_query` — Civitai hop (user Confirms downloads). Without `search_query` the hop is skipped (never search the whole user message).
 - `"interrupt"` — stop generation.
@@ -88,5 +89,5 @@ Several options in one ask (still one fence):
 - `"persona_read"` — load lore shelves not in always-on identity (e.g. `roleplay` / `craft` / `humor` when NSFW tone or craft detail needs them).
 - `"memory_upsert"` / `"memory_forget"` / `"user_pref_upsert"` / `"user_pref_forget"` — writes.
 - `"persona_clone"` / `"persona_write"` / `"persona_switch"` — `author_persona` only. Never `"persona_delete"`.
-- `look_at: ["generate"|"ref1"|…]` — vision hop (JPEG arrives on the follow-up).
+- `look_at: ["generate"|"ref1"|…]` — vision hop (JPEG arrives on the follow-up). Opt-in: user asked, or you truly need pixels. Do not pair `look_at` with `actions:["generate"]` on a normal write turn (that stares at the *old* frame and delays the new one).
 - Pure Q&A: omit the JSON patch.

Config/_base/exact.json

@@ -4,7 +4,8 @@
     "steps": 8,
     "cfg": 1,
     "sigma_shift": 1.15,
-    "images": 1
+    "images": 1,
+    "negative": "bad quality, worst quality"
   },
   "profiles": {
     "turbo": {
@@ -31,7 +32,7 @@
   "facts": {
     "architecture": "Krea 2 is a 12B DiT architecture. Not FLUX, not SDXL, not FLUX.1-Krea. Text encoder: Qwen3-VL 4B. VAE: Qwen Image VAE. Use only Krea2-trained LoRAs — never suggest FLUX/SDXL LoRAs.",
     "prompt_language": "Chat model always prep's Generate prompt for Krea: English natural prose for Qwen3-VL, structured (subject→pose→setting→camera→light). Chat may be RU; never leave Russian or thin drafts in patch.prompt.",
-    "negatives": "Negative prompts are nearly useless with Qwen3-VL. Prefer positives (sharp focus, empty street) over no blur / no people. Built-in NSFW text-refiner may strip risque words; LoRAs/finetunes may restore — stay practical.",
+    "negatives": "Qwen3-VL negatives are weak — still keep a short Swarm negative box. Prefer positives in `prompt`; for `negative`: create if live is empty (Exact generation.negative), lightly supplement if the scene needs a specific omit, or echo live unchanged. Never drop/clear the box on Generate. Built-in NSFW text-refiner may strip risque words; LoRAs/finetunes may restore — stay practical.",
     "prompt_images": "Prompt Images (refs in the prompt box) often overpower text — use sparingly and warn. Init Image = structure (img2img). Mask = local fix. They are not interchangeable. Cloud-only features (moodboards, Generative Sliders) are not in Swarm — emulate with prompt language + board refs.",
     "turbo": "Krea 2 Turbo defaults live in exact.generation / exact.profiles.turbo. Never use CFG 0 — broken output. Side ~1024 (128–4096 OK).",
     "raw": "Krea 2 RAW/Base: prefer exact.profiles.raw when checkpoint name/title looks like RAW (not Turbo). If a turbo LoRA exists, weight ~0.6 for photoreal (1.0 ≈ full turbo). Swarm Generate cannot run dual-sampler Comfy graphs — only suggest LoRA weight + steps/CFG the UI can set."

Config/_base/packs/ordinary.md

@@ -4,8 +4,8 @@ Default all-rounder. Handle this turn from the user message + live context — d
 
 ## What you cover here
 
-- **Write / improve prompt** → patch with `prompt` (+ `loras` when useful) and `actions: ["generate"]` when they want a new image.
-- **Light critique / improve last frame** → short notes + better `prompt`; use `look_at: ["generate"]` if `has_vision_image` and you have not seen the JPEG this turn.
+- **Write / improve prompt** → patch with `prompt` + `negative` (+ `loras` when useful) and `actions: ["generate"]` when this turn is a frame (scene, edit, «ещё») — **not** only if they typed «генерируй». Do **not** `look_at` the last frame first. `negative`: create / supplement / echo live — do not omit on Generate.
+- **Light critique / improve last frame** → only when the user asks to look / critique / describe the picture. Then `look_at: ["generate"]` if `images_in_request` is false. Otherwise edit the prompt from text; `has_vision_image` alone is not a reason to look.
 - **Scene / mood** → compose direction into the prompt (same patch rules).
 - **Params** → only when they ask (steps/CFG/aspect/seed); omit Exact-matching numbers otherwise.
 - **Inpaint / img2img** → set init/mask fields when they ask and flags allow; else say what is missing.
@@ -26,6 +26,6 @@ Otherwise **stay in ordinary** and just do the work.
 
 ## Deliverable
 
-Same as write_prompt: short reply + one fenced JSON patch; `actions: ["generate"]` when they want an image.
-«давай дальше» / next frame = new English `prompt` + `actions:["generate"]` in the **same** turn — never leave an empty `### JSON Patch` header. Chat may be RU; **Generate `prompt` is always EN** (skill `prompting`).
+Same as write_prompt: short reply + one fenced JSON patch; `actions: ["generate"]` when the turn is a frame (context is enough — they do not have to say «генерируй»).
+«давай дальше» / next frame = new English `prompt` + `negative` (echo live if unchanged) + `actions:["generate"]` in the **same** turn — never leave an empty `### JSON Patch` header. Chat may be RU; **Generate `prompt` is always EN** (skill `prompting`).
 Several options in one ask → `variants` (2–4 partial patches with `label`); still one fence, still STOP after it.

Config/_base/packs/write_prompt.md

@@ -5,8 +5,8 @@ Goal: craft or improve a **Krea 2** prompt that will generate well on Turbo (loc
 ## Deliverable
 
 - Brief note of what you changed.
-- JSON patch with at least `prompt`, and `loras` when relevant.
-- `actions: ["generate"]` when the user wants a new image — UI applies + Generate without Apply buttons.
+- JSON patch with at least `prompt` and `negative` (create / supplement / echo live), and `loras` when relevant.
+- `actions: ["generate"]` when this turn is a new/updated image (scene brief, edit, «ещё») — they do **not** have to type «генерируй». Skip generate only for Q&A / remember / look-only. Do **not** `look_at` unless they asked to see/critique the last frame.
 - Prefer Exact Turbo defaults / `recommended_params`. Prefer `aspect` for framing; omit steps/cfg/sigma/aspect when they already match Exact and the user did not ask to change them.
 - Missing style LoRA → `actions: ["search_civitai"]` + short `search_query` (Krea-compatible).
 - User wants several options (оба / варианты / разный свет) → `variants: [{label, prompt|aspect|…}, …]` (2–4). Base keys inherit; each item overrides only its diffs. Still one fence.

Config/_base/skills/prompting.md

@@ -9,12 +9,13 @@ The **chat model** prepares the Generate-box text for **Krea 2** (Qwen3-VL). Do
 3. **Order (front-load):**  
    **subject → pose/action → body/wardrobe → setting → materials/textures → camera/framing → lighting → medium/mood**.
 4. Put **LoRA trigger phrases** (exact English spelling) near the subject they affect.
-5. Prefer **positives** over negatives (Qwen negatives are weak).
+5. **`negative` on every Generate** — create if live is empty (Exact `generation.negative`), supplement if the scene needs a specific omit, or echo live unchanged. Put “no blur / no people” ideas as positives in `prompt` instead of stuffing the negative box. Never clear `negative`.
 6. Short ideas: expand into a full Krea paragraph. Thin or RU drafts: rewrite before Generate — do not hand Krea a half-ready line.
 
 ## Prep checklist (before `actions:["generate"]`)
 
 - English only in `prompt`
+- `negative` present (new / supplemented / echoed — not omitted)
 - Subject and action clear in the first sentence
 - Wardrobe / body / setting concrete
 - Camera + lighting present