[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_154":155,"_156":157},"routes/_app.x.$handle.$name",{"_11":12,"_13":14,"_15":16,"_18":19,"_17":-7,"_117":-5,"_20":118,"_22":119,"_120":121,"_122":123,"_25":124,"_24":-5,"_125":-5,"_126":-5,"_127":53,"_128":129,"_130":44,"_131":-5,"_132":-5,"_133":134},"routes/_app.x.$handle.$name.branch.$branch.code.$",{"_11":12,"_13":14,"_15":16,"_17":-7,"_18":19,"_20":21,"_22":23,"_24":-7,"_25":26,"_27":-5,"_28":-5,"_29":-5,"_30":-5,"_31":-5},"handle","rafalwiktorowicz","name","chatgpt55","branchName","main","version","path","","project",{"_33":61,"_13":14,"_45":62,"_63":-5,"_64":53,"_65":66,"_67":66,"_68":69,"_70":-5,"_71":72,"_73":-5,"_74":53,"_75":-5,"_76":-5,"_77":78,"_79":80,"_81":-5,"_82":-5,"_83":-5,"_84":-5,"_85":86,"_87":86,"_88":89,"_90":89,"_91":89,"_92":86,"_93":86,"_94":95,"_96":97,"_98":99,"_100":53},"tree",[56],"file","contents",[32],"upstream","pull","ahead","behind","envVars",{"_33":34,"_13":35,"_36":37,"_38":-5,"_39":40,"_41":42,"_43":44,"_45":46,"_17":47,"_48":49,"_50":51,"_52":53,"_54":-5,"_18":55},"id","d5a3a576-bdfc-4546-b9b2-49014d26b080","main.ts","fileId","b17d7cd6-548b-11f1-813b-ee650bb23af1","parentId","branchId","b17acc98-548b-11f1-813b-ee650bb23af1","branchVersion",25,"latest",true,"createdAt",["D",1785697259467],24,"fileType","http","content","// ============================================================\n// CHATGPT (OpenAI) — MCP serwer (Val Town / Deno) v3.7 ★ GPT-5.6 (Sol/Terra/Luna)\n// Generacja gpt-image-2 + EDYCJA Z MASKĄ + WIZJA + TEKST + batch (pLimit) + BATCH API async\n// Hosting obrazów w blob, serwowane pod /img/. ID trwałe między edycjami kodu.\n// ENV: OPENAI_API_KEY (opcjonalnie GEMINI_ORIGIN dla edycji finałów z banana-pro)\n//\n// ZMIANA v3.7 (vs v3.6) — JEDNA, minimalna, naprawia podłączanie connectora w Claude:\n// ★ /.well-known/* zwraca teraz 404 zamiast strony powitalnej.\n// DLACZEGO: Claude przy dodawaniu connectora pyta o metadane OAuth pod\n// /.well-known/oauth-authorization-server. Ten serwer odpowiadał na KAŻDE GET\n// statusem 200 i tekstem powitalnym — Claude czytał to jako \"OAuth istnieje\",\n// próbował dynamicznej rejestracji klienta i dostawał tekst zamiast JSON-a.\n// Efekt: \"Couldn't register with Chatgpr's sign-in service\".\n// Serwer nie wymaga logowania — musi to POWIEDZIEĆ WPROST, czyli 404.\n// Reszta kodu bez zmian.\n//\n// ZMIANY v3.6 (vs v3.5) — zgodne ze specyfikacją OpenAI z 09.07.2026:\n// ★ RODZINA GPT-5.6: gpt-5.6-sol (flagowiec, alias gpt-5.6), gpt-5.6-terra (balans),\n// gpt-5.6-luna (tanie, masowe). Kontekst 1.05M, max output 128K.\n// Aliasy: sol|terra|luna|5.6. Domyślny model tekst/wizja: gpt-5.6-sol.\n// ★ REASONING EFFORT: none|low|medium|high|xhigh|max (doszły xhigh i max).\n// ★ PRO MODE: to TRYB WYKONANIA, nie model. reasoning.mode:\"pro\" działa z każdym 5.6.\n// Stary slug \"pro\" -> gpt-5.5-pro USUNIĘTY (mapuje się teraz na sol + pro_mode).\n// ★ REASONING CONTEXT: reasoning.context = auto|all_turns|current_turn (reużycie\n// rozumowania między turami). Przy store:false wymaga include reasoning.encrypted_content.\n// ★ SAFETY_IDENTIFIER: opcjonalny identyfikator użytkownika (klasyfikatory cyber/bio).\n// ★ PROMPT CACHE: prompt_cache_options.ttl (zastąpiło prompt_cache_retention).\n// ★ Wszystkie modele 5.6 chodzą WYŁĄCZNIE przez /v1/responses (RESPONSES_ONLY),\n// w tym vision_plan — na chat/completions ich nie ma.\n//\n// ZMIANY v3.5: TEXT_PLAN + TEXT_PLAN_BATCH + BATCH_SUBMIT_TEXTS (Responses API).\n// ZMIANY v3.4: MASK_RECTS (maska z prostokątów, zero base64 od klienta).\n// ZMIANY v3.3: BATCH API (-50%) dla generacji obrazów.\n// ZMIANY v3.2: mask_is_alpha / mask_white_erases.\n// ZMIANY v3.1: pngSize() z IHDR + UPNG.js + nearest-neighbor resize maski.\n// ZMIANY v3.0: input_fidelity tylko dla gpt-image-1; maska; erase_mode; fallback_model.\n// ============================================================\nimport { blob } from \"https://esm.town/v/std/blob\";\n\nconst IMG_MODEL_DEFAULT = \"gpt-image-2\";\nconst VIS_MODEL_DEFAULT = \"gpt-5.6-sol\";\n\nconst IMG_MODELS: Record = {\n \"image2\": \"gpt-image-2\",\n \"gpt-image-2\": \"gpt-image-2\",\n \"image1\": \"gpt-image-1\",\n \"gpt-image-1\": \"gpt-image-1\",\n \"image1.5\": \"gpt-image-1.5\",\n \"gpt-image-1.5\": \"gpt-image-1.5\",\n \"mini\": \"gpt-image-1-mini\",\n};\n\nconst VIS_MODELS: Record = {\n // GPT-5.6 (aktualna rodzina)\n \"5.6\": \"gpt-5.6-sol\",\n \"sol\": \"gpt-5.6-sol\",\n \"terra\": \"gpt-5.6-terra\",\n \"luna\": \"gpt-5.6-luna\",\n \"pro\": \"gpt-5.6-sol\", // \"pro\" = tryb wykonania, nie model -> patrz pro_mode\n // starsze\n \"5.5\": \"gpt-5.5\",\n \"5.4\": \"gpt-5.4\",\n \"4o\": \"gpt-4o\",\n \"mini\": \"gpt-5.4-mini\",\n};\n\nconst resolveImg = (m?: string) =>\n !m ? IMG_MODEL_DEFAULT : (IMG_MODELS[m] ?? m);\nconst resolveVis = (m?: string) =>\n !m ? VIS_MODEL_DEFAULT : (VIS_MODELS[m] ?? m);\n\n// modele które WSPIERAJĄ input_fidelity (reszta dostaje HTTP 400, więc nie wysyłamy)\nconst SUPPORTS_INPUT_FIDELITY = new Set([\"gpt-image-1\"]);\n\n// modele dostępne WYŁĄCZNIE w Responses API (nie istnieją w chat/completions)\nconst RESPONSES_ONLY = new Set([\n \"gpt-5.6-sol\",\n \"gpt-5.6-terra\",\n \"gpt-5.6-luna\",\n \"gpt-5.6\",\n \"gpt-5.5-pro\",\n]);\n\n// dozwolone poziomy reasoning effort w 5.6\nconst EFFORTS = new Set([\"none\", \"low\", \"medium\", \"high\", \"xhigh\", \"max\"]);\n\n// klauzula ochrony konturu doklejana w erase_mode (karty how-to-draw)\nconst ERASE_GUARD =\n \" Output the SAME bold black line drawing on a pure white background. \" +\n \"Keep EVERY other line pixel-for-pixel identical — do NOT redraw, move, thicken, \" +\n \"re-center, restyle or add anything. No color, no shading.\";\n\nconst RETRY = {\n maxAttempts: 6,\n baseMs: 800,\n factor: 1.8,\n capMs: 20_000,\n retryStatuses: new Set([408, 409, 425, 429, 500, 502, 503, 504]),\n};\nconst TIMEOUT_MS = 120_000;\nconst TIMEOUT_PRO_MS = 600_000; // pro mode / max effort potrafią mielić długo\nconst DEFAULT_CONCURRENCY = 4, MAX_CONCURRENCY = 8;\nconst sleep = (ms: number) => new Promise((r) => setTimeout(r, ms));\nconst jitter = (ms: number) => Math.round(ms * (0.7 + Math.random() * 0.6));\n\nclass OErr extends Error {\n status: number;\n retryable: boolean;\n retryAfterMs?: number;\n constructor(m: string, s: number, r: boolean) {\n super(m);\n this.status = s;\n this.retryable = r;\n }\n}\n\nfunction pLimit(limit: number) {\n let active = 0;\n const q: (() => void)[] = [];\n const next = () => {\n active--;\n q.shift()?.();\n };\n return (fn: () => Promise) =>\n new Promise((res, rej) => {\n const run = () => {\n active++;\n fn().then(res, rej).finally(next);\n };\n active < limit ? run() : q.push(run);\n });\n}\n\nasync function withRetry(fn: () => Promise): Promise {\n let a = 0, wait = RETRY.baseMs, last: unknown;\n while (a < RETRY.maxAttempts) {\n try {\n return await fn();\n } catch (e) {\n const o = e as OErr;\n if (!o.retryable) throw e;\n if (++a >= RETRY.maxAttempts) break;\n await sleep(o.retryAfterMs ?? jitter(Math.min(wait, RETRY.capMs)));\n wait *= RETRY.factor;\n last = e;\n }\n }\n throw last;\n}\n\nconst KEY = () => {\n const k = Deno.env.get(\"OPENAI_API_KEY\");\n if (!k) throw new OErr(\"brak OPENAI_API_KEY w ENV\", 500, false);\n return k;\n};\n\nlet SELF_ORIGIN = \"\";\n\nasync function storeImage(b64: string, mime = \"image/png\") {\n const id = crypto.randomUUID().replace(/-/g, \"\");\n await blob.setJSON(`img:${id}`, { b64, mime });\n return { id, url: `${SELF_ORIGIN}/img/${id}` };\n}\n\n// image = http(s) URL | data-URL | val id | \"gem:\" | surowy base64\nasync function loadBytes(\n image: string,\n): Promise<{ bytes: Uint8Array; mime: string }> {\n if (image.startsWith(\"gem:\")) {\n const gorigin = Deno.env.get(\"GEMINI_ORIGIN\");\n if (!gorigin) {\n throw new OErr(\n \"podano gem: ale brak ENV GEMINI_ORIGIN (origin vala Gemini)\",\n 400,\n false,\n );\n }\n const gid = image.slice(4);\n return await loadBytes(`${gorigin.replace(/\\/$/, \"\")}/img/${gid}`);\n }\n if (/^https?:/i.test(image)) {\n const r = await fetch(image);\n if (!r.ok) {\n throw new OErr(\n `pobranie obrazu HTTP ${r.status}`,\n r.status,\n RETRY.retryStatuses.has(r.status),\n );\n }\n return {\n bytes: new Uint8Array(await r.arrayBuffer()),\n mime: r.headers.get(\"content-type\") || \"image/png\",\n };\n }\n if (image.startsWith(\"data:\")) {\n const mime = image.slice(5, image.indexOf(\";\")) || \"image/png\";\n const b = image.replace(/^data:[^,]+,/, \"\");\n return { bytes: Uint8Array.from(atob(b), (c) => c.charCodeAt(0)), mime };\n }\n const o = await blob.getJSON(`img:${image}`);\n if (o) {\n return {\n bytes: Uint8Array.from(atob(o.b64), (c) => c.charCodeAt(0)),\n mime: o.mime || \"image/png\",\n };\n }\n return {\n bytes: Uint8Array.from(atob(image), (c) => c.charCodeAt(0)),\n mime: \"image/png\",\n };\n}\n\n// === MASKA: konwersja czarno-białej maski (BIAŁE=zmień) na alfa-maskę OpenAI ===\nfunction pngSize(bytes: Uint8Array): { w: number; h: number } | null {\n if (\n bytes.length < 24 || bytes[0] !== 0x89 || bytes[1] !== 0x50 ||\n bytes[2] !== 0x4e || bytes[3] !== 0x47\n ) return null;\n const dv = new DataView(bytes.buffer, bytes.byteOffset, bytes.byteLength);\n return { w: dv.getUint32(16), h: dv.getUint32(20) };\n}\n\nlet _UPNG: any = null;\nasync function loadUPNG() {\n if (_UPNG) return _UPNG;\n const mod: any = await import(\"https://esm.sh/upng-js@2.1.0?target=deno\");\n const cand = [mod?.default, mod, mod?.UPNG, mod?.default?.default].filter(\n Boolean,\n );\n const U = cand.find((c) => c && typeof c.decode === \"function\");\n if (!U) {\n throw new OErr(\n \"UPNG nie załadował się poprawnie (brak metody decode) — sprawdź import esm.sh\",\n 500,\n false,\n );\n }\n _UPNG = U;\n return U;\n}\n\nasync function decodeToRGBA(\n bytes: Uint8Array,\n): Promise<{ w: number; h: number; rgba: Uint8Array }> {\n const UPNG = await loadUPNG();\n const ab = bytes.buffer.slice(\n bytes.byteOffset,\n bytes.byteOffset + bytes.byteLength,\n );\n const img = UPNG.decode(ab);\n const frames = UPNG.toRGBA8(img);\n if (!frames || !frames[0]) {\n throw new OErr(\"UPNG.toRGBA8 nie zwrócił klatek\", 500, false);\n }\n return { w: img.width, h: img.height, rgba: new Uint8Array(frames[0]) };\n}\n\nasync function maskWhiteToAlphaPng(\n maskBytes: Uint8Array,\n targetW: number,\n targetH: number,\n invert = false,\n): Promise {\n const UPNG = await loadUPNG();\n const { w: mw, h: mh, rgba } = await decodeToRGBA(maskBytes);\n\n const out = new Uint8Array(targetW * targetH * 4);\n for (let y = 0; y < targetH; y++) {\n const sy = mh === targetH ? y : Math.min(mh - 1, (y * mh / targetH) | 0);\n for (let x = 0; x < targetW; x++) {\n const sx = mw === targetW ? x : Math.min(mw - 1, (x * mw / targetW) | 0);\n const si = (sy * mw + sx) * 4;\n const r = rgba[si], g = rgba[si + 1], b = rgba[si + 2];\n const lum = 0.299 * r + 0.587 * g + 0.114 * b;\n let erase = lum > 127; // jasny (białe pole) = obszar do zmiany\n if (invert) erase = !erase;\n const di = (y * targetW + x) * 4;\n out[di] = 0;\n out[di + 1] = 0;\n out[di + 2] = 0;\n out[di + 3] = erase ? 0 : 255; // 0 = edytuj (przezroczyste), 255 = zostaw\n }\n }\n const png = UPNG.encode([out.buffer], targetW, targetH, 0);\n return new Uint8Array(png);\n}\n\n// === MASKA Z PROSTOKĄTÓW: klient wysyła [[x1,y1,x2,y2],...] zamiast base64 ===\nasync function rectsToAlphaMask(\n targetW: number,\n targetH: number,\n rects: number[][],\n pad = 0,\n): Promise {\n const UPNG = await loadUPNG();\n const out = new Uint8Array(targetW * targetH * 4);\n for (let i = 0; i < targetW * targetH; i++) out[i * 4 + 3] = 255;\n for (const r of rects) {\n if (!r || r.length < 4) continue;\n const x1 = Math.max(0, Math.min(targetW - 1, (r[0] | 0) - pad));\n const y1 = Math.max(0, Math.min(targetH - 1, (r[1] | 0) - pad));\n const x2 = Math.max(0, Math.min(targetW, (r[2] | 0) + pad));\n const y2 = Math.max(0, Math.min(targetH, (r[3] | 0) + pad));\n for (let y = y1; y < y2; y++) {\n for (let x = x1; x < x2; x++) {\n out[(y * targetW + x) * 4 + 3] = 0; // edytuj\n }\n }\n }\n const png = UPNG.encode([out.buffer], targetW, targetH, 0);\n return new Uint8Array(png);\n}\n\n// === GENERACJA (images/generations) ===\nasync function genOne(\n o: { prompt: string; model?: string; size?: string; quality?: string },\n) {\n return await withRetry(async () => {\n const ctrl = new AbortController();\n const t = setTimeout(() => ctrl.abort(), TIMEOUT_MS);\n let res: Response;\n try {\n res = await fetch(\"https://api.openai.com/v1/images/generations\", {\n method: \"POST\",\n headers: {\n Authorization: `Bearer ${KEY()}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify({\n model: resolveImg(o.model),\n prompt: o.prompt,\n n: 1,\n ...(o.size ? { size: o.size } : {}),\n ...(o.quality ? { quality: o.quality } : {}),\n }),\n signal: ctrl.signal,\n });\n } catch (e) {\n throw new OErr(`network: ${(e as Error).message}`, 0, true);\n } finally {\n clearTimeout(t);\n }\n if (!res.ok) {\n const b = await res.text().catch(() => \"\");\n throw new OErr(\n `HTTP ${res.status}: ${b.slice(0, 300)}`,\n res.status,\n RETRY.retryStatuses.has(res.status),\n );\n }\n const data = await res.json();\n const b64 = data?.data?.[0]?.b64_json;\n if (!b64) throw new OErr(\"brak b64_json w odpowiedzi\", 200, false);\n return await storeImage(b64, \"image/png\");\n });\n}\n\n// === EDYCJA (images/edits) — model-aware fidelity, MASKA, mask_rects, erase_mode, fallback ===\nasync function editOnce(\n o: {\n image: string;\n prompt: string;\n model?: string;\n size?: string;\n mask?: string;\n mask_rects?: number[][];\n mask_pad?: number;\n mask_white_erases?: boolean;\n mask_is_alpha?: boolean;\n mask_invert?: boolean;\n input_fidelity?: string;\n erase_mode?: boolean;\n },\n modelOverride?: string,\n) {\n const model = modelOverride ?? resolveImg(o.model);\n const { bytes, mime } = await loadBytes(o.image);\n const prompt = o.erase_mode ? (o.prompt + ERASE_GUARD) : o.prompt;\n\n const fd = new FormData();\n fd.append(\"model\", model);\n fd.append(\"prompt\", prompt);\n if (o.size) fd.append(\"size\", o.size);\n\n if (SUPPORTS_INPUT_FIDELITY.has(model)) {\n fd.append(\"input_fidelity\", o.input_fidelity || \"high\");\n }\n\n fd.append(\n \"image\",\n new Blob([bytes], { type: mime || \"image/png\" }),\n \"image.png\",\n );\n\n // MASKA Z PROSTOKĄTÓW — priorytet nad `mask`, zero base64 od klienta\n if (o.mask_rects && o.mask_rects.length && !o.mask) {\n let bw = 0, bh = 0;\n const hdr = pngSize(bytes);\n if (hdr) {\n bw = hdr.w;\n bh = hdr.h;\n } else {\n const dec = await decodeToRGBA(bytes);\n bw = dec.w;\n bh = dec.h;\n }\n const maskBytes = await rectsToAlphaMask(\n bw,\n bh,\n o.mask_rects,\n o.mask_pad ?? 0,\n );\n fd.append(\"mask\", new Blob([maskBytes], { type: \"image/png\" }), \"mask.png\");\n }\n\n if (o.mask) {\n const mk = await loadBytes(o.mask);\n let maskBytes = mk.bytes;\n if (o.mask_white_erases && !o.mask_is_alpha) {\n try {\n let bw = 0, bh = 0;\n const hdr = pngSize(bytes);\n if (hdr) {\n bw = hdr.w;\n bh = hdr.h;\n } else {\n const dec = await decodeToRGBA(bytes);\n bw = dec.w;\n bh = dec.h;\n }\n maskBytes = await maskWhiteToAlphaPng(\n maskBytes,\n bw,\n bh,\n Boolean(o.mask_invert),\n );\n } catch (_e) {\n maskBytes = mk.bytes;\n }\n }\n fd.append(\"mask\", new Blob([maskBytes], { type: \"image/png\" }), \"mask.png\");\n }\n\n const ctrl = new AbortController();\n const t = setTimeout(() => ctrl.abort(), TIMEOUT_MS);\n let res: Response;\n try {\n res = await fetch(\"https://api.openai.com/v1/images/edits\", {\n method: \"POST\",\n headers: { Authorization: `Bearer ${KEY()}` },\n body: fd,\n signal: ctrl.signal,\n });\n } catch (e) {\n throw new OErr(`network: ${(e as Error).message}`, 0, true);\n } finally {\n clearTimeout(t);\n }\n if (!res.ok) {\n const b = await res.text().catch(() => \"\");\n throw new OErr(\n `HTTP ${res.status}: ${b.slice(0, 300)}`,\n res.status,\n RETRY.retryStatuses.has(res.status),\n );\n }\n const data = await res.json();\n const b64 = data?.data?.[0]?.b64_json;\n if (!b64) throw new OErr(\"brak b64_json w odpowiedzi\", 200, false);\n return await storeImage(b64, \"image/png\");\n}\n\nasync function editOne(o: any) {\n return await withRetry(async () => {\n try {\n return await editOnce(o);\n } catch (e) {\n const oe = e as OErr;\n if (\n o.fallback_model && !oe.retryable && oe.status >= 400 && oe.status < 500\n ) {\n return await editOnce(o, resolveImg(o.fallback_model));\n }\n throw e;\n }\n });\n}\n\n// === TEKST (Responses API /v1/responses) — v3.6, GPT-5.6 ===\ntype TextOpts = {\n prompt: string;\n model?: string;\n system?: string;\n as_json?: boolean;\n json_schema?: any;\n schema_name?: string;\n reasoning_effort?: string; // none|low|medium|high|xhigh|max\n pro_mode?: boolean; // reasoning.mode = \"pro\"\n reasoning_context?: string; // auto|all_turns|current_turn\n safety_identifier?: string;\n cache_ttl?: string; // prompt_cache_options.ttl, np. \"24h\"\n web_search?: boolean;\n previous_response_id?: string;\n store?: boolean;\n max_output_tokens?: number;\n image_url?: string;\n image_b64?: string;\n mime?: string;\n};\n\nasync function textOne(o: TextOpts) {\n return await withRetry(async () => {\n if (o.reasoning_effort && !EFFORTS.has(o.reasoning_effort)) {\n throw new OErr(\n `reasoning_effort=\"${o.reasoning_effort}\" — dozwolone: ${\n [...EFFORTS].join(\"|\")\n }`,\n 400,\n false,\n );\n }\n\n // input: sam tekst albo multimodal (tekst + obraz)\n let input: any = o.prompt;\n let imgUrl = o.image_url;\n if (!imgUrl && o.image_b64) {\n const b = o.image_b64.replace(/^data:[^,]+,/, \"\");\n imgUrl = `data:${o.mime || \"image/png\"};base64,${b}`;\n }\n if (imgUrl) {\n input = [{\n role: \"user\",\n content: [\n { type: \"input_text\", text: o.prompt },\n { type: \"input_image\", image_url: imgUrl },\n ],\n }];\n }\n\n const store = Boolean(o.store);\n const body: any = {\n model: resolveVis(o.model),\n input,\n store, // domyślnie bezstanowo; true -> wątek po stronie OpenAI\n ...(o.system ? { instructions: o.system } : {}),\n ...(o.previous_response_id\n ? { previous_response_id: o.previous_response_id }\n : {}),\n ...(o.max_output_tokens\n ? { max_output_tokens: o.max_output_tokens }\n : {}),\n ...(o.safety_identifier\n ? { safety_identifier: o.safety_identifier }\n : {}),\n ...(o.cache_ttl ? { prompt_cache_options: { ttl: o.cache_ttl } } : {}),\n };\n\n // reasoning: effort + pro mode + context (GPT-5.6)\n if (o.reasoning_effort || o.pro_mode || o.reasoning_context) {\n body.reasoning = {\n ...(o.reasoning_effort ? { effort: o.reasoning_effort } : {}),\n ...(o.pro_mode ? { mode: \"pro\" } : {}),\n ...(o.reasoning_context ? { context: o.reasoning_context } : {}),\n };\n }\n // reużycie rozumowania bez store -> trzeba zaszyfrowanego bloku w odpowiedzi\n if (\n o.reasoning_context && o.reasoning_context !== \"current_turn\" && !store\n ) {\n body.include = [\"reasoning.encrypted_content\"];\n }\n\n if (o.web_search) body.tools = [{ type: \"web_search\" }];\n\n // Structured outputs: w Responses to text.format (NIE response_format!)\n if (o.json_schema) {\n body.text = {\n format: {\n type: \"json_schema\",\n name: o.schema_name || \"wynik\",\n schema: o.json_schema,\n strict: true,\n },\n };\n } else if (o.as_json) {\n body.text = { format: { type: \"json_object\" } };\n }\n\n const heavy = o.pro_mode ||\n o.reasoning_effort === \"max\" || o.reasoning_effort === \"xhigh\";\n const ctrl = new AbortController();\n const t = setTimeout(\n () => ctrl.abort(),\n heavy ? TIMEOUT_PRO_MS : TIMEOUT_MS,\n );\n let res: Response;\n try {\n res = await fetch(\"https://api.openai.com/v1/responses\", {\n method: \"POST\",\n headers: {\n Authorization: `Bearer ${KEY()}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify(body),\n signal: ctrl.signal,\n });\n } catch (e) {\n throw new OErr(`network: ${(e as Error).message}`, 0, true);\n } finally {\n clearTimeout(t);\n }\n if (!res.ok) {\n const b = await res.text().catch(() => \"\");\n throw new OErr(\n `HTTP ${res.status}: ${b.slice(0, 300)}`,\n res.status,\n RETRY.retryStatuses.has(res.status),\n );\n }\n const data = await res.json();\n const out = extractResponsesText(data);\n if (!out.trim()) throw new OErr(\"pusta odpowiedź (Responses)\", 200, false);\n return JSON.stringify({\n text: out.trim(),\n response_id: data?.id ?? null,\n model: data?.model ?? resolveVis(o.model),\n usage: data?.usage ?? null,\n });\n });\n}\n\n// REST /v1/responses NIE zwraca helpera output_text (to SDK) — składamy z output[]\nfunction extractResponsesText(data: any): string {\n return (data?.output ?? [])\n .filter((it: any) => it?.type === \"message\")\n .flatMap((it: any) => it?.content ?? [])\n .filter((c: any) => c?.type === \"output_text\")\n .map((c: any) => c?.text ?? \"\")\n .join(\"\\n\");\n}\n\n// === WIZJA (obraz -> tekst/JSON) ===\n// v3.6: cała rodzina 5.6 istnieje TYLKO w Responses -> delegacja do textOne.\n// Stare modele (5.5/5.4/4o/mini) nadal przez chat/completions.\nasync function visionOne(\n o: {\n prompt: string;\n model?: string;\n image_b64?: string;\n image_url?: string;\n mime?: string;\n as_json?: boolean;\n pro_mode?: boolean;\n reasoning_effort?: string;\n },\n) {\n const resolved = resolveVis(o.model);\n if (RESPONSES_ONLY.has(resolved)) {\n const r = await textOne({\n prompt: o.prompt,\n model: o.model,\n as_json: o.as_json,\n pro_mode: o.model === \"pro\" || Boolean(o.pro_mode),\n reasoning_effort: o.reasoning_effort,\n image_b64: o.image_b64,\n image_url: o.image_url,\n mime: o.mime,\n });\n return JSON.parse(r).text as string;\n }\n return await withRetry(async () => {\n let imgUrl = o.image_url;\n if (!imgUrl && o.image_b64) {\n const b = o.image_b64.replace(/^data:[^,]+,/, \"\");\n imgUrl = `data:${o.mime || \"image/png\"};base64,${b}`;\n }\n if (!imgUrl) throw new OErr(\"podaj image_b64 albo image_url\", 400, false);\n const body: any = {\n model: resolved,\n messages: [{\n role: \"user\",\n content: [{ type: \"text\", text: o.prompt }, {\n type: \"image_url\",\n image_url: { url: imgUrl },\n }],\n }],\n ...(o.as_json ? { response_format: { type: \"json_object\" } } : {}),\n };\n const ctrl = new AbortController();\n const t = setTimeout(() => ctrl.abort(), TIMEOUT_MS);\n let res: Response;\n try {\n res = await fetch(\"https://api.openai.com/v1/chat/completions\", {\n method: \"POST\",\n headers: {\n Authorization: `Bearer ${KEY()}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify(body),\n signal: ctrl.signal,\n });\n } catch (e) {\n throw new OErr(`network: ${(e as Error).message}`, 0, true);\n } finally {\n clearTimeout(t);\n }\n if (!res.ok) {\n const b = await res.text().catch(() => \"\");\n throw new OErr(\n `HTTP ${res.status}: ${b.slice(0, 300)}`,\n res.status,\n RETRY.retryStatuses.has(res.status),\n );\n }\n const data = await res.json();\n const out = data?.choices?.[0]?.message?.content ?? \"\";\n if (!out.trim()) throw new OErr(\"pusta odpowiedź\", 200, false);\n return out.trim();\n });\n}\n\nasync function batch(\n items: any[],\n conc: number,\n fn: (it: any, i: number) => Promise,\n) {\n const lim = pLimit(Math.max(1, Math.min(MAX_CONCURRENCY, conc)));\n const out = await Promise.all(items.map((it, index) =>\n lim(async () => {\n try {\n return { index, ok: true, ...(await fn(it, index) as any) };\n } catch (e) {\n return { index, ok: false, error: (e as Error).message };\n }\n })\n ));\n out.sort((a: any, b: any) => a.index - b.index);\n return out;\n}\n\n// === BATCH API (OpenAI /v1/batches, async, -50% taniej; do ~24h) ===\nasync function batchCreate(lines: string, endpoint: string) {\n if (!lines) throw new OErr(\"puste wejście batcha\", 400, false);\n const fd = new FormData();\n fd.append(\"purpose\", \"batch\");\n fd.append(\n \"file\",\n new Blob([lines], { type: \"application/jsonl\" }),\n \"batch.jsonl\",\n );\n const up = await fetch(\"https://api.openai.com/v1/files\", {\n method: \"POST\",\n headers: { Authorization: `Bearer ${KEY()}` },\n body: fd,\n });\n if (!up.ok) {\n const b = await up.text().catch(() => \"\");\n throw new OErr(\n `files HTTP ${up.status}: ${b.slice(0, 300)}`,\n up.status,\n false,\n );\n }\n const file = await up.json();\n if (!file?.id) throw new OErr(\"upload file: brak id\", 500, false);\n\n const br = await fetch(\"https://api.openai.com/v1/batches\", {\n method: \"POST\",\n headers: {\n Authorization: `Bearer ${KEY()}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify({\n input_file_id: file.id,\n endpoint,\n completion_window: \"24h\",\n }),\n });\n if (!br.ok) {\n const b = await br.text().catch(() => \"\");\n throw new OErr(\n `batches HTTP ${br.status}: ${b.slice(0, 300)}`,\n br.status,\n false,\n );\n }\n const batchObj = await br.json();\n if (!batchObj?.id) throw new OErr(\"create batch: brak id\", 500, false);\n return batchObj;\n}\n\nasync function batchSubmitImages(o: {\n prompts: string[];\n model?: string;\n size?: string;\n quality?: string;\n}) {\n const model = resolveImg(o.model);\n const lines = (o.prompts ?? []).map((p, i) =>\n JSON.stringify({\n custom_id: `img-${i}`,\n method: \"POST\",\n url: \"/v1/images/generations\",\n body: {\n model,\n prompt: p,\n n: 1,\n ...(o.size ? { size: o.size } : {}),\n ...(o.quality ? { quality: o.quality } : {}),\n },\n })\n ).join(\"\\n\");\n const batchObj = await batchCreate(lines, \"/v1/images/generations\");\n return {\n batch_id: batchObj.id,\n status: batchObj.status,\n n: o.prompts.length,\n note:\n \"Status: batch_status(batch_id). Wynik: batch_results(batch_id) gdy completed (do 24h).\",\n };\n}\n\n// batch TEKSTOWY przez /v1/responses (-50%)\nasync function batchSubmitTexts(o: {\n items?: { prompt: string; system?: string; as_json?: boolean }[];\n prompts?: string[];\n model?: string;\n reasoning_effort?: string;\n pro_mode?: boolean;\n max_output_tokens?: number;\n}) {\n const model = resolveVis(o.model);\n const items = o.items ?? (o.prompts ?? []).map((p) => ({ prompt: p }));\n const reasoning = (o.reasoning_effort || o.pro_mode)\n ? {\n reasoning: {\n ...(o.reasoning_effort ? { effort: o.reasoning_effort } : {}),\n ...(o.pro_mode ? { mode: \"pro\" } : {}),\n },\n }\n : {};\n const lines = items.map((it, i) =>\n JSON.stringify({\n custom_id: `txt-${i}`,\n method: \"POST\",\n url: \"/v1/responses\",\n body: {\n model,\n input: String(it.prompt ?? \"\"),\n store: false,\n ...(it.system ? { instructions: it.system } : {}),\n ...(it.as_json ? { text: { format: { type: \"json_object\" } } } : {}),\n ...reasoning,\n ...(o.max_output_tokens\n ? { max_output_tokens: o.max_output_tokens }\n : {}),\n },\n })\n ).join(\"\\n\");\n const batchObj = await batchCreate(lines, \"/v1/responses\");\n return {\n batch_id: batchObj.id,\n status: batchObj.status,\n n: items.length,\n note:\n \"Status: batch_status(batch_id). Wynik: batch_results_texts(batch_id) gdy completed (do 24h).\",\n };\n}\n\nasync function batchStatus(batch_id: string) {\n const r = await fetch(`https://api.openai.com/v1/batches/${batch_id}`, {\n headers: { Authorization: `Bearer ${KEY()}` },\n });\n if (!r.ok) {\n const b = await r.text().catch(() => \"\");\n throw new OErr(`HTTP ${r.status}: ${b.slice(0, 300)}`, r.status, false);\n }\n const j = await r.json();\n return {\n status: j.status,\n request_counts: j.request_counts,\n output_file_id: j.output_file_id ?? null,\n error_file_id: j.error_file_id ?? null,\n };\n}\n\nasync function fetchBatchOutput(batch_id: string) {\n const r = await fetch(`https://api.openai.com/v1/batches/${batch_id}`, {\n headers: { Authorization: `Bearer ${KEY()}` },\n });\n if (!r.ok) {\n const b = await r.text().catch(() => \"\");\n throw new OErr(`HTTP ${r.status}: ${b.slice(0, 300)}`, r.status, false);\n }\n const j = await r.json();\n if (j.status !== \"completed\") {\n return { pending: true, status: j.status, text: \"\" };\n }\n if (!j.output_file_id) {\n return {\n pending: false,\n status: j.status,\n text: \"\",\n error_file_id: j.error_file_id ?? null,\n };\n }\n const fr = await fetch(\n `https://api.openai.com/v1/files/${j.output_file_id}/content`,\n { headers: { Authorization: `Bearer ${KEY()}` } },\n );\n return { pending: false, status: \"completed\", text: await fr.text() };\n}\n\nasync function batchResults(batch_id: string) {\n const out = await fetchBatchOutput(batch_id);\n if (out.pending) {\n return { status: out.status, results: [], note: \"Jeszcze nie completed.\" };\n }\n if (!out.text) {\n return {\n status: out.status,\n results: [],\n error_file_id: (out as any).error_file_id ?? null,\n note: \"Brak output_file_id (sprawdź error_file_id).\",\n };\n }\n const results: any[] = [];\n for (const line of out.text.split(\"\\n\").filter(Boolean)) {\n try {\n const row = JSON.parse(line);\n const data = row?.response?.body?.data?.[0];\n if (data?.b64_json) {\n const st = await storeImage(data.b64_json, \"image/png\");\n results.push({ custom_id: row.custom_id, ...st });\n } else if (data?.url) {\n results.push({ custom_id: row.custom_id, url: data.url });\n } else {\n results.push({\n custom_id: row.custom_id,\n error: row?.response?.body?.error ?? \"no image\",\n });\n }\n } catch (e) {\n results.push({ error: (e as Error).message });\n }\n }\n return { status: \"completed\", results };\n}\n\nasync function batchResultsTexts(batch_id: string) {\n const out = await fetchBatchOutput(batch_id);\n if (out.pending) {\n return { status: out.status, results: [], note: \"Jeszcze nie completed.\" };\n }\n if (!out.text) {\n return {\n status: out.status,\n results: [],\n error_file_id: (out as any).error_file_id ?? null,\n note: \"Brak output_file_id (sprawdź error_file_id).\",\n };\n }\n const results: any[] = [];\n for (const line of out.text.split(\"\\n\").filter(Boolean)) {\n try {\n const row = JSON.parse(line);\n const body = row?.response?.body;\n if (body?.error) {\n results.push({ custom_id: row.custom_id, error: body.error });\n continue;\n }\n results.push({\n custom_id: row.custom_id,\n text: extractResponsesText(body),\n });\n } catch (e) {\n results.push({ error: (e as Error).message });\n }\n }\n return { status: \"completed\", results };\n}\n\nconst TOOLS = [\n {\n name: \"generate_image\",\n description:\n \"Generuje obraz (gpt-image-2). size: 1024x1024|1536x1024|1024x1536 albo dowolne WxH (dzielne przez 16, proporcje 1:3–3:1, max 3840x2160). Zwraca {id,url}. base64 przez get_image(id).\",\n inputSchema: {\n type: \"object\",\n properties: {\n prompt: { type: \"string\" },\n model: { type: \"string\" },\n size: { type: \"string\" },\n quality: { type: \"string\" },\n },\n required: [\"prompt\"],\n },\n },\n {\n name: \"generate_images_batch\",\n description:\n \"★ WIELOWĄTKOWO generuje LISTĘ promptów (pLimit). Zwraca [{index,id,url}]. base64 przez get_images.\",\n inputSchema: {\n type: \"object\",\n properties: {\n prompts: { type: \"array\", items: { type: \"string\" } },\n model: { type: \"string\" },\n size: { type: \"string\" },\n quality: { type: \"string\" },\n max_concurrent: { type: \"number\" },\n },\n required: [\"prompts\"],\n },\n },\n {\n name: \"edit_image\",\n description:\n \"Edytuje obraz. image = url | data-url | id | base64 | gem:. \" +\n \"★ MASK_RECTS (zalecane): mask_rects=[[x1,y1,x2,y2],...] (+mask_pad px) — val buduje alfa-maskę SAM, \" +\n \"zero base64 od klienta. Region-lock: edycja TYLKO w prostokątach. \" +\n \"Albo: mask (jak image) + mask_white_erases:true (BIAŁE=obszar do zmiany). \" +\n \"erase_mode:true dokleja klauzulę ochrony konturu (karty how-to-draw). \" +\n \"model: image1/image1.5/image2 (input_fidelity auto, tylko image1). fallback_model opcjonalnie. Zwraca {id,url}.\",\n inputSchema: {\n type: \"object\",\n properties: {\n image: { type: \"string\" },\n prompt: { type: \"string\" },\n model: { type: \"string\" },\n size: { type: \"string\" },\n mask: { type: \"string\" },\n mask_rects: {\n type: \"array\",\n items: { type: \"array\", items: { type: \"number\" } },\n },\n mask_pad: { type: \"number\" },\n mask_white_erases: { type: \"boolean\" },\n mask_is_alpha: { type: \"boolean\" },\n mask_invert: { type: \"boolean\" },\n input_fidelity: { type: \"string\" },\n erase_mode: { type: \"boolean\" },\n fallback_model: { type: \"string\" },\n },\n required: [\"image\", \"prompt\"],\n },\n },\n {\n name: \"edit_images_batch\",\n description: \"★ WIELOWĄTKOWO edytuje LISTĘ (pLimit). \" +\n \"edits=[{image,prompt,model,size,mask,mask_rects,mask_pad,mask_white_erases,mask_invert,erase_mode,fallback_model}]. \" +\n \"Zwraca [{index,id,url}].\",\n inputSchema: {\n type: \"object\",\n properties: {\n edits: { type: \"array\", items: { type: \"object\" } },\n max_concurrent: { type: \"number\" },\n },\n required: [\"edits\"],\n },\n },\n {\n name: \"get_image\",\n description: \"Zwraca base64 JEDNEGO obrazu wg id.\",\n inputSchema: {\n type: \"object\",\n properties: { id: { type: \"string\" } },\n required: [\"id\"],\n },\n },\n {\n name: \"get_images\",\n description:\n \"★ Zwraca DUŻY payload {images:[{id,b64,mime}]} dla ids (zrzut na dysk).\",\n inputSchema: {\n type: \"object\",\n properties: { ids: { type: \"array\", items: { type: \"string\" } } },\n required: [\"ids\"],\n },\n },\n {\n name: \"text_plan\",\n description:\n \"★ TEKST (Responses API, GPT-5.6): prompt -> TEKST/JSON. Opcjonalnie obraz (image_url/image_b64). \" +\n \"model: sol|terra|luna (alias 5.6 = sol) | 5.5 | 5.4 | mini | 4o. \" +\n \"sol = frontier (reasoning, kod, agenty), terra = balans cena/jakość, luna = tanio i masowo. \" +\n \"reasoning_effort: none|low|medium|high|xhigh|max. pro_mode:true = tryb pro (więcej obliczeń, jeden wynik, dłuższa latencja). \" +\n \"reasoning_context: auto|all_turns|current_turn (reużycie rozumowania między turami). \" +\n \"system=instructions; as_json | json_schema(+schema_name) = structured outputs; \" +\n \"web_search:true = wbudowane wyszukiwanie; previous_response_id + store:true = wątek po stronie OpenAI; \" +\n 'cache_ttl (np. \"24h\") = prompt_cache_options.ttl; safety_identifier = id użytkownika. ' +\n \"Zwraca {text, response_id, model, usage}.\",\n inputSchema: {\n type: \"object\",\n properties: {\n prompt: { type: \"string\" },\n model: { type: \"string\" },\n system: { type: \"string\" },\n as_json: { type: \"boolean\" },\n json_schema: { type: \"object\" },\n schema_name: { type: \"string\" },\n reasoning_effort: { type: \"string\" },\n pro_mode: { type: \"boolean\" },\n reasoning_context: { type: \"string\" },\n safety_identifier: { type: \"string\" },\n cache_ttl: { type: \"string\" },\n web_search: { type: \"boolean\" },\n previous_response_id: { type: \"string\" },\n store: { type: \"boolean\" },\n max_output_tokens: { type: \"number\" },\n image_url: { type: \"string\" },\n image_b64: { type: \"string\" },\n mime: { type: \"string\" },\n },\n required: [\"prompt\"],\n },\n },\n {\n name: \"text_plan_batch\",\n description:\n \"★ WIELOWĄTKOWO (pLimit) LISTA zapytań tekstowych przez Responses API. \" +\n \"items=[{prompt,model,system,as_json,json_schema,reasoning_effort,pro_mode,reasoning_context,web_search,max_output_tokens,image_url,image_b64}]. \" +\n \"Zwraca [{index,ok,text,response_id}].\",\n inputSchema: {\n type: \"object\",\n properties: {\n items: { type: \"array\", items: { type: \"object\" } },\n concurrency: { type: \"number\" },\n },\n required: [\"items\"],\n },\n },\n {\n name: \"vision_plan\",\n description:\n \"★ WIZJA: obraz + prompt -> TEKST/JSON. Modele 5.6 (sol/terra/luna) idą automatycznie przez Responses API; \" +\n \"5.5/5.4/4o przez chat/completions. image_b64 ALBO image_url. as_json=true -> JSON. pro_mode/reasoning_effort tylko dla 5.6.\",\n inputSchema: {\n type: \"object\",\n properties: {\n prompt: { type: \"string\" },\n model: { type: \"string\" },\n image_b64: { type: \"string\" },\n image_url: { type: \"string\" },\n mime: { type: \"string\" },\n as_json: { type: \"boolean\" },\n pro_mode: { type: \"boolean\" },\n reasoning_effort: { type: \"string\" },\n },\n required: [\"prompt\"],\n },\n },\n {\n name: \"vision_plan_batch\",\n description:\n \"★ WIELOWĄTKOWO analizuje LISTĘ obrazów (pLimit). items=[{prompt,model,image_b64|image_url,mime,as_json,pro_mode,reasoning_effort}]. Wyniki w kolejności.\",\n inputSchema: {\n type: \"object\",\n properties: {\n items: { type: \"array\", items: { type: \"object\" } },\n concurrency: { type: \"number\" },\n },\n required: [\"items\"],\n },\n },\n {\n name: \"batch_submit_images\",\n description:\n \"★ BATCH API (-50% taniej, async do 24h). Wysyła LISTĘ promptów do generacji OBRAZÓW. Zwraca {batch_id,status}. Do MASOWEJ produkcji baz (np. siatki nocą).\",\n inputSchema: {\n type: \"object\",\n properties: {\n prompts: { type: \"array\", items: { type: \"string\" } },\n model: { type: \"string\" },\n size: { type: \"string\" },\n quality: { type: \"string\" },\n },\n required: [\"prompts\"],\n },\n },\n {\n name: \"batch_submit_texts\",\n description:\n \"★ BATCH API TEKST (-50% taniej, async do 24h) przez /v1/responses. \" +\n \"prompts=[string,...] ALBO items=[{prompt,system,as_json}]. model/reasoning_effort/pro_mode wspólne dla całej paczki. \" +\n \"Zwraca {batch_id,status}. Odbiór: batch_results_texts.\",\n inputSchema: {\n type: \"object\",\n properties: {\n prompts: { type: \"array\", items: { type: \"string\" } },\n items: { type: \"array\", items: { type: \"object\" } },\n model: { type: \"string\" },\n reasoning_effort: { type: \"string\" },\n pro_mode: { type: \"boolean\" },\n max_output_tokens: { type: \"number\" },\n },\n },\n },\n {\n name: \"batch_status\",\n description:\n \"Status batcha (obrazy i tekst): {status, request_counts, output_file_id}. status: validating/in_progress/finalizing/completed/failed/expired.\",\n inputSchema: {\n type: \"object\",\n properties: { batch_id: { type: \"string\" } },\n required: [\"batch_id\"],\n },\n },\n {\n name: \"batch_results\",\n description:\n \"Pobiera wyniki batcha OBRAZÓW gdy completed: {results:[{custom_id,id,url}]}. Obrazy hostowane pod /img/.\",\n inputSchema: {\n type: \"object\",\n properties: { batch_id: { type: \"string\" } },\n required: [\"batch_id\"],\n },\n },\n {\n name: \"batch_results_texts\",\n description:\n \"Pobiera wyniki batcha TEKSTOWEGO gdy completed: {results:[{custom_id,text}]}.\",\n inputSchema: {\n type: \"object\",\n properties: { batch_id: { type: \"string\" } },\n required: [\"batch_id\"],\n },\n },\n];\n\nasync function dispatch(name: string, a: any): Promise {\n if (name === \"generate_image\") {\n const r = await genOne({\n prompt: String(a?.prompt ?? \"\"),\n model: a?.model,\n size: a?.size,\n quality: a?.quality,\n });\n return JSON.stringify(r);\n }\n if (name === \"generate_images_batch\") {\n const r = await batch(\n (a?.prompts ?? []).map((p: string) => ({ prompt: p })),\n Number(a?.max_concurrent) || DEFAULT_CONCURRENCY,\n (it) =>\n genOne({\n prompt: String(it.prompt),\n model: a?.model,\n size: a?.size,\n quality: a?.quality,\n }),\n );\n return JSON.stringify(r);\n }\n if (name === \"edit_image\") {\n const r = await editOne({\n image: String(a?.image ?? \"\"),\n prompt: String(a?.prompt ?? \"\"),\n model: a?.model,\n size: a?.size,\n mask: a?.mask,\n mask_rects: a?.mask_rects,\n mask_pad: a?.mask_pad,\n mask_white_erases: Boolean(a?.mask_white_erases),\n mask_is_alpha: Boolean(a?.mask_is_alpha),\n mask_invert: Boolean(a?.mask_invert),\n input_fidelity: a?.input_fidelity,\n erase_mode: Boolean(a?.erase_mode),\n fallback_model: a?.fallback_model,\n });\n return JSON.stringify(r);\n }\n if (name === \"edit_images_batch\") {\n const r = await batch(\n a?.edits ?? [],\n Number(a?.max_concurrent) || DEFAULT_CONCURRENCY,\n (it) =>\n editOne({\n image: String(it.image),\n prompt: String(it.prompt),\n model: it.model,\n size: it.size,\n mask: it.mask,\n mask_rects: it.mask_rects,\n mask_pad: it.mask_pad,\n mask_white_erases: Boolean(it.mask_white_erases),\n mask_is_alpha: Boolean(it.mask_is_alpha),\n mask_invert: Boolean(it.mask_invert),\n input_fidelity: it.input_fidelity,\n erase_mode: Boolean(it.erase_mode),\n fallback_model: it.fallback_model,\n }),\n );\n return JSON.stringify(r);\n }\n if (name === \"get_image\") {\n const o = await blob.getJSON(`img:${a.id}`);\n if (!o) throw new OErr(`brak obrazu ${a.id}`, 404, false);\n return JSON.stringify({ id: a.id, b64: o.b64, mime: o.mime });\n }\n if (name === \"get_images\") {\n const ids: string[] = a?.ids ?? [];\n const images: any[] = [];\n for (const id of ids) {\n const o = await blob.getJSON(`img:${id}`);\n if (o) images.push({ id, b64: o.b64, mime: o.mime });\n }\n let payload = JSON.stringify({ images });\n const TARGET = 1150000;\n if (payload.length < TARGET) {\n payload = JSON.stringify({\n images,\n _pad: \" \".repeat(TARGET - payload.length),\n });\n }\n return payload;\n }\n if (name === \"text_plan\") {\n return await textOne({\n prompt: String(a?.prompt ?? \"\"),\n model: a?.model,\n system: a?.system,\n as_json: Boolean(a?.as_json),\n json_schema: a?.json_schema,\n schema_name: a?.schema_name,\n reasoning_effort: a?.reasoning_effort,\n pro_mode: Boolean(a?.pro_mode) || a?.model === \"pro\",\n reasoning_context: a?.reasoning_context,\n safety_identifier: a?.safety_identifier,\n cache_ttl: a?.cache_ttl,\n web_search: Boolean(a?.web_search),\n previous_response_id: a?.previous_response_id,\n store: Boolean(a?.store),\n max_output_tokens: a?.max_output_tokens\n ? Number(a.max_output_tokens)\n : undefined,\n image_url: a?.image_url,\n image_b64: a?.image_b64,\n mime: a?.mime,\n });\n }\n if (name === \"text_plan_batch\") {\n const r = await batch(\n a?.items ?? [],\n Number(a?.concurrency) || DEFAULT_CONCURRENCY,\n async (it) => {\n const res = await textOne({\n prompt: String(it?.prompt ?? \"\"),\n model: it?.model,\n system: it?.system,\n as_json: Boolean(it?.as_json),\n json_schema: it?.json_schema,\n schema_name: it?.schema_name,\n reasoning_effort: it?.reasoning_effort,\n pro_mode: Boolean(it?.pro_mode) || it?.model === \"pro\",\n reasoning_context: it?.reasoning_context,\n safety_identifier: it?.safety_identifier,\n cache_ttl: it?.cache_ttl,\n web_search: Boolean(it?.web_search),\n max_output_tokens: it?.max_output_tokens\n ? Number(it.max_output_tokens)\n : undefined,\n image_url: it?.image_url,\n image_b64: it?.image_b64,\n mime: it?.mime,\n });\n return JSON.parse(res);\n },\n );\n return JSON.stringify(r, null, 2);\n }\n if (name === \"vision_plan\") {\n return await visionOne({\n prompt: String(a?.prompt ?? \"\"),\n model: a?.model,\n image_b64: a?.image_b64,\n image_url: a?.image_url,\n mime: a?.mime,\n as_json: Boolean(a?.as_json),\n pro_mode: Boolean(a?.pro_mode),\n reasoning_effort: a?.reasoning_effort,\n });\n }\n if (name === \"vision_plan_batch\") {\n const r = await batch(\n a?.items ?? [],\n Number(a?.concurrency) || DEFAULT_CONCURRENCY,\n async (it) => ({\n text: await visionOne({\n prompt: String(it?.prompt ?? \"\"),\n model: it?.model,\n image_b64: it?.image_b64,\n image_url: it?.image_url,\n mime: it?.mime,\n as_json: Boolean(it?.as_json),\n pro_mode: Boolean(it?.pro_mode),\n reasoning_effort: it?.reasoning_effort,\n }),\n }),\n );\n return JSON.stringify(r, null, 2);\n }\n if (name === \"batch_submit_images\") {\n return JSON.stringify(\n await batchSubmitImages({\n prompts: a?.prompts ?? [],\n model: a?.model,\n size: a?.size,\n quality: a?.quality,\n }),\n );\n }\n if (name === \"batch_submit_texts\") {\n return JSON.stringify(\n await batchSubmitTexts({\n prompts: a?.prompts,\n items: a?.items,\n model: a?.model,\n reasoning_effort: a?.reasoning_effort,\n pro_mode: Boolean(a?.pro_mode),\n max_output_tokens: a?.max_output_tokens\n ? Number(a.max_output_tokens)\n : undefined,\n }),\n );\n }\n if (name === \"batch_status\") {\n return JSON.stringify(await batchStatus(String(a?.batch_id ?? \"\")));\n }\n if (name === \"batch_results\") {\n return JSON.stringify(await batchResults(String(a?.batch_id ?? \"\")));\n }\n if (name === \"batch_results_texts\") {\n return JSON.stringify(await batchResultsTexts(String(a?.batch_id ?? \"\")));\n }\n throw new OErr(`Nieznane narzędzie: ${name}`, 400, false);\n}\n\nconst ok = (id: any, result: any) =>\n Response.json({ jsonrpc: \"2.0\", id, result });\nconst er = (id: any, code: number, message: string) =>\n Response.json({ jsonrpc: \"2.0\", id, error: { code, message } });\n\nexport default async function (req: Request): Promise {\n const url = new URL(req.url);\n SELF_ORIGIN = url.origin;\n if (req.method === \"GET\" && url.pathname.startsWith(\"/img/\")) {\n const id = url.pathname.slice(5);\n const o = await blob.getJSON(`img:${id}`);\n if (!o) return new Response(\"not found\", { status: 404 });\n const bin = Uint8Array.from(atob(o.b64), (c) => c.charCodeAt(0));\n return new Response(bin, {\n headers: {\n \"Content-Type\": o.mime || \"image/png\",\n \"Cache-Control\": \"public, max-age=31536000\",\n },\n });\n }\n // ★ v3.7 — JEDYNA ZMIANA W CAŁYM PLIKU.\n // Discovery OAuth: mówimy WPROST \"nie mam logowania\" (404), zamiast oddawać\n // stronę powitalną ze statusem 200. Bez tego Claude czyta 200 jako \"OAuth\n // istnieje\", rusza z dynamiczną rejestracją klienta i dostaje tekst zamiast\n // JSON-a -> \"Couldn't register with Chatgpr's sign-in service\".\n // Serwer nie wymaga autoryzacji — i teraz to sygnalizuje poprawnie.\n if (url.pathname.startsWith(\"/.well-known/\")) {\n return new Response(null, { status: 404 });\n }\n if (req.method === \"GET\") {\n return new Response(\n `ChatGPT (OpenAI) MCP v3.7 — działa. Obrazy: ${IMG_MODEL_DEFAULT}. Tekst/Wizja: ${VIS_MODEL_DEFAULT} (sol|terra|luna). ` +\n `Responses API: reasoning effort none..max, pro_mode, reasoning_context, web_search, json_schema, wątki. ` +\n `Edycja z maską (mask_rects bez base64) i erase_mode aktywne. Batch API (-50%): obrazy + tekst.`,\n { status: 200 },\n );\n }\n let body: any = {};\n try {\n body = await req.json();\n } catch {\n return er(null, -32700, \"Parse error\");\n }\n const { id, method, params } = body;\n if (method === \"initialize\") {\n return ok(id, {\n protocolVersion: \"2024-11-05\",\n capabilities: { tools: {} },\n serverInfo: { name: \"chatgpt\", version: \"3.7\" },\n });\n }\n if (\n method === \"notifications/initialized\" ||\n method?.startsWith?.(\"notifications/\")\n ) return new Response(null, { status: 202 });\n if (method === \"tools/list\") return ok(id, { tools: TOOLS });\n if (method === \"tools/call\") {\n try {\n return ok(id, {\n content: [{\n type: \"text\",\n text: await dispatch(params?.name, params?.arguments ?? {}),\n }],\n });\n } catch (e) {\n return ok(id, {\n content: [{\n type: \"text\",\n text: `BŁĄD OpenAI: ${(e as Error).message}`,\n }],\n isError: true,\n });\n }\n }\n return er(id, -32601, \"Method not found\");\n}","deleted",false,"label",[35],{"_33":34,"_13":35,"_36":37,"_38":-5,"_39":40,"_41":42,"_43":44,"_45":57,"_17":47,"_48":49,"_50":51,"_52":53,"_54":-5,"_58":59,"_18":60},["D",1785697259467],"children",[],[35],"b17a01be-548b-11f1-813b-ee650bb23af1",["D",1779309437993],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","chatgpt5","imageUrl","userId","0d65c1b5-f226-4c7a-8c4b-30b17a9d544e","pinnedFileId","pinPreview","pinMigrated","screenshotVersion","likeCount",0,"orgId","019e308b-edd7-77ca-895e-c4f0d8ba1fdb","tursoName","tursoHostname","tursoMeasuredAt","pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_11":12,"_116":112},"org",{"_11":12,"_109":110,"_111":112,"_71":72,"_113":114,"_115":53},"branch",{"_17":42,"_13":16,"_33":40,"_101":61,"_102":-5,"_103":-5,"_104":-5,"_105":-5,"_106":-5,"_107":14,"_108":12},"isOwned","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","displayName","R W","avatar","https://lh3.googleusercontent.com/a/ACg8ocKAmj_qhWrVoXHf18hayjA4wNNLTkmWMqHGPR0qAaQrAmi6AA=s96-c","tier","free","canRestrictVals","profileImageUrl","adoptOrgId",{"_33":61,"_13":14,"_45":150,"_63":-5,"_64":53,"_65":66,"_67":66,"_68":69,"_70":-5,"_71":72,"_73":-5,"_74":53,"_75":-5,"_76":-5,"_77":78,"_79":80,"_81":-5,"_82":-5,"_83":-5,"_84":-5,"_85":86,"_87":86,"_88":89,"_90":89,"_91":89,"_92":86,"_93":86,"_94":151,"_96":152,"_98":153,"_100":53},[147],"files",[147],"counts",{"_137":78,"_138":139,"_140":78,"_31":141,"_142":143},[135],"upstreamProject","upstreamBranch","emptyEnvVars","flags",[],"hasHttpFiles","grants","orgs","customDomains",{},{"_33":34,"_13":35,"_36":37,"_38":-5,"_39":40,"_41":42,"_43":44,"_45":46,"_17":47,"_48":49,"_50":51,"_52":53,"_54":-5,"_18":136},[35],"pulls","branches",1,"forks",{"_144":139,"_146":78},"connections",{"_144":78,"_145":78},"count","unfilledCount","emptyCount",{"_33":34,"_13":35,"_36":37,"_38":-5,"_39":40,"_41":42,"_43":44,"_45":57,"_17":47,"_48":49,"_50":51,"_52":53,"_54":-5,"_58":148,"_18":149},[],[35],["D",1779309437993],{"_11":12,"_116":112},{"_11":12,"_109":110,"_111":112,"_71":72,"_113":114,"_115":53},{"_17":42,"_13":16,"_33":40,"_101":61,"_102":-5,"_103":-5,"_104":-5,"_105":-5,"_106":-5,"_107":14,"_108":12},"ctx",{"_94":-5,"_132":187,"_188":189,"_190":191,"_192":53,"_193":53,"_194":53,"_195":44,"_196":-5,"_197":-7,"_198":-7,"_199":78,"_200":53,"_201":53,"_202":-5,"_203":53,"_204":53,"_205":53,"_206":-5},"ENV",{"_158":159,"_160":161,"_162":16,"_163":164,"_165":166,"_167":168,"_169":170,"_171":172,"_173":174,"_175":176,"_177":178,"_179":180,"_181":182,"_183":184,"_185":186},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","RENDER_GIT_COMMIT","e87eb04e707cf64e7e532c8e8ec395f209e17f15","RENDER_GIT_BRANCH","POSTHOG_API_TOKEN","phc_nmoyuAkStEZuUp69rYlg1DZgrIt19Gx0Y4Hwf8tGgJP","POSTHOG_API_HOST","https://wart.val.town","WEB_API_TEMPLATE","https://{val}.web.val.run","APP_URL","https://www.val.town","MODULE_URL","https://esm.town","VANITY_WEB_API_TEMPLATE","https://{label}.val.run/","DENOLS_URL","https://lsp.val.town","NODE_ENV","production","CLOUDFLARE_TURNSTILE_SITE_KEY","0x4AAAAAACtaIZJH7s9rBMZ6","GOOGLE_CLIENT_ID","875989696899-jtsqkhlbdv5jrlsk8631p7ipjg3c6o62.apps.googleusercontent.com","STRIPE_PUBLISHABLE_KEY","pk_live_51N2zvWEpZTTLipIkVEwXrUG9AT0yeOOEYxszarZT1jpsZV58Ly6wlg3SltZpHN26CUalOBmUSY7adQRQiSkCZz2500qSXNkf8Y","API_URL","https://api.val.town",[],"preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesCloseBrackets","preferencesChatPanelState","preferencesDprint","preferencesDenofmt","notificationCount","flagDenols","isBusinessOrgMember","businessOrgHandle","isBusinessOrgAdmin","isHigherTownieTier","navCollapsed","currentOrgId"]