NL → Code Agent

Describe what you want in plain English. Get TypeScript code back.

\n \n

\n \n \n \n \n \n \n \n

\n ${code ? `

Generated Code

${code.replace(//g, \">\")}

` : \"\"}\n

\n\n`);\n});\n\n// ── Dataset routes ──\n\n// Browse dataset with pagination (JSON API)\napp.get(\"/dataset\", async (c) => {\n const offset = parseInt(c.req.query(\"offset\") ?? \"0\", 10);\n const length = Math.min(parseInt(c.req.query(\"length\") ?? \"10\", 10), 100);\n const page = await loadDatasetPage(offset, length);\n return c.json(page);\n});\n\n// Dataset metadata\napp.get(\"/dataset/info\", async (c) => {\n const info = await getDatasetInfo();\n return c.json(info);\n});\n\n// Search dataset by keyword\napp.get(\"/dataset/search\", async (c) => {\n const q = c.req.query(\"q\") ?? \"\";\n const limit = parseInt(c.req.query(\"limit\") ?? \"10\", 10);\n if (!q) return c.json({ error: \"Provide ?q=keyword\" }, 400);\n const results = await searchDataset(q, limit);\n return c.json({ query: q, results });\n});\n\n// Get a specific row by index\napp.get(\"/dataset/row/:idx\", async (c) => {\n const idx = parseInt(c.req.param(\"idx\"), 10);\n const page = await loadDatasetPage(idx, 1);\n if (page.rows.length === 0) return c.json({ error: \"Row not found\" }, 404);\n return c.json(page.rows[0]);\n});\n\n// ── Analytics routes ──\n\n// JSON 분석 데이터 (전체 4,665행 fetch → 통계 계산)\napp.get(\"/dataset/analytics\", async (c) => {\n const result = await computeAnalytics();\n return c.json(result);\n});\n\n// 분석 대시보드 (HTML)\napp.get(\"/dataset/analytics/dashboard\", async (c) => {\n return await serveFile(\"analytics-dashboard.html\");\n});\n\n// Always add this for full stack traces on errors:\napp.onError((err) => Promise.reject(err));\n\nexport default app.fetch;\n","deleted","label",[115],"9aa65419-198a-48d3-9b42-6aface389761","analytics-dashboard.html","bc334f7a-7a83-11f1-8dbf-1607ee4eb77e",9,["D",1783484164189],"\n\n\n\n\nFable-5-Traces Analytics\n\n\n\n

\n ← NL → Code Agent\n

Fable-5-Traces 데이터셋 분석

Glint-Research/Fable-5-traces — 4,665개 AI 에이전트 실행 기록 통계

\n\n

전체 4,665개 행을 병렬로 fetch 중... (약 5-15초 소요)

\n\n \n

\n\n\n\n\n",[131],"f5b2d57b-faaa-4c43-a93c-2caf59f2ff91","lib","2df540fa-e4d5-401b-b625-bc7ee89bbdac",3,["D",1783484057587],"directory",[138],"pulls","branches",1,"forks","envVars",{"_152":42,"_154":42},"connections",{"_152":42,"_153":42},"count","unfilledCount","emptyCount",{"_18":137,"_20":138,"_116":139,"_117":-5,"_118":71,"_119":140,"_120":38,"_22":183,"_68":42,"_123":142,"_125":-5,"_127":26,"_128":-5,"_161":184,"_89":185},{"_18":175,"_20":176,"_116":177,"_117":139,"_118":71,"_119":178,"_120":38,"_22":179,"_68":146,"_123":171,"_125":180,"_127":26,"_128":-5,"_161":181,"_89":182},{"_18":167,"_20":168,"_116":169,"_117":139,"_118":71,"_119":140,"_120":38,"_22":170,"_68":42,"_123":171,"_125":172,"_127":26,"_128":-5,"_161":173,"_89":174},{"_18":130,"_20":131,"_116":132,"_117":-5,"_118":71,"_119":133,"_120":38,"_22":164,"_68":42,"_123":100,"_125":135,"_127":26,"_128":-5,"_161":165,"_89":166},{"_18":114,"_20":115,"_116":36,"_117":-5,"_118":71,"_119":69,"_120":38,"_22":160,"_68":122,"_123":124,"_125":126,"_127":26,"_128":-5,"_161":162,"_89":163},["D",1783484191819],"children",[],[115],["D",1783484164189],[],[131],"2f30095a-e804-4664-a765-f595141ec8aa","hf-dataset.ts","7ca9292e-7a83-11f1-b50b-1607ee4eb77e",["D",1783484057587],"script","// TypeScript equivalent of:\n// from datasets import load_dataset\n// ds = load_dataset(\"Glint-Research/Fable-5-traces\")\n//\n// Uses the Hugging Face datasets-server REST API:\n// https://huggingface.co/docs/datasets-server/\n//\n// Val Town runs Deno (not Python), so we fetch via HTTP instead.\n\nconst DATASET = \"Glint-Research/Fable-5-traces\";\nconst CONFIG = \"pi_agent\";\nconst SPLIT = \"train\";\nconst BASE = \"https://datasets-server.huggingface.co\";\n\nexport interface FableTrace {\n harness: string;\n session_id: string;\n prompt: string;\n messages: any[];\n tools: any[];\n metadata: any;\n sent_at: string;\n num_user_messages: number;\n num_tool_calls: number;\n trace: any;\n file_path: string;\n}\n\nexport interface HFRowsResponse {\n features: any[];\n rows: { row_idx: number; row: FableTrace }[];\n num_rows_total: number;\n num_rows_per_page: number;\n}\n\n/** Fetch a page of rows from the dataset (max 100 per call). */\nexport async function loadDatasetPage(\n offset = 0,\n length = 100,\n): Promise {\n const url = `${BASE}/rows?dataset=${encodeURIComponent(DATASET)}&config=${CONFIG}&split=${SPLIT}&offset=${offset}&length=${length}`;\n const res = await fetch(url);\n if (!res.ok) {\n throw new Error(`HF API error ${res.status}: ${await res.text()}`);\n }\n return res.json();\n}\n\n/** Get dataset metadata (splits, features, num_examples, etc.) */\nexport async function getDatasetInfo() {\n const res = await fetch(\n `${BASE}/info?dataset=${encodeURIComponent(DATASET)}&config=${CONFIG}`,\n );\n if (!res.ok) {\n throw new Error(`HF API error ${res.status}: ${await res.text()}`);\n }\n return res.json();\n}\n\n/** Load the full dataset by paginating through all rows.\n * Returns an async generator so you can iterate without loading everything\n * into memory at once.\n *\n * Usage:\n * for await (const row of loadDataset()) {\n * console.log(row.prompt);\n * }\n */\nexport async function* loadDataset(): AsyncGenerator {\n const pageSize = 100;\n let offset = 0;\n let total = Infinity;\n\n while (offset < total) {\n const page = await loadDatasetPage(offset, pageSize);\n total = page.num_rows_total;\n for (const r of page.rows) {\n yield r.row;\n }\n offset += page.rows.length;\n if (page.rows.length === 0) break;\n }\n}\n\n/** Search the dataset by keyword in the prompt field.\n * Iterates through all 4,665 rows (47 API calls).\n */\nexport async function searchDataset(\n query: string,\n limit = 10,\n): Promise<{ row_idx: number; prompt: string; session_id: string }[]> {\n const q = query.toLowerCase();\n const results: { row_idx: number; prompt: string; session_id: string }[] = [];\n\n for await (const row of loadDataset()) {\n if (row.prompt.toLowerCase().includes(q)) {\n results.push({\n row_idx: results.length,\n prompt: row.prompt.slice(0, 200),\n session_id: row.session_id,\n });\n if (results.length >= limit) break;\n }\n }\n\n return results;\n}\n",[],[138,168],"227d4fa4-fbed-4ac8-ac31-a2c52f76d33c","analytics.ts","aae9f28c-7a83-11f1-a869-1607ee4eb77e",7,["D",1783484137390],"// Fable-5-traces 데이터셋 분석 모듈\n// 전체 4,665개 행을 병렬로 fetch하여 통계 추출\n\nimport { loadDatasetPage, type FableTrace } from \"./hf-dataset.ts\";\n\nexport interface AnalyticsResult {\n totalTraces: number;\n totalToolCalls: number;\n totalUserMessages: number;\n avgToolCalls: number;\n avgUserMessages: number;\n avgPromptLength: number;\n maxToolCalls: number;\n maxUserMessages: number;\n maxPromptLength: number;\n // 분포\n toolCallsBuckets: { range: string; count: number }[];\n userMessagesBuckets: { range: string; count: number }[];\n promptLengthBuckets: { range: string; count: number }[];\n // 카테고리별\n harnessDist: Record;\n modelDist: Record;\n thinkingLevelDist: Record;\n // 상위 도구\n topTools: { name: string; count: number }[];\n // 날짜별\n tracesByDate: { date: string; count: number }[];\n // turn_count 분포\n turnCountBuckets: { range: string; count: number }[];\n // 처리 시간\n fetchedAt: string;\n fetchDurationMs: number;\n}\n\n/** 전체 데이터셋을 병렬로 fetch (47개 API 호출을 배치로 처리) */\nasync function fetchAllRows(): Promise {\n const PAGE_SIZE = 100;\n const firstPage = await loadDatasetPage(0, PAGE_SIZE);\n const total = firstPage.num_rows_total;\n const numPages = Math.ceil(total / PAGE_SIZE);\n\n const rows: FableTrace[] = firstPage.rows.map((r) => r.row);\n\n // 나머지 페이지를 병렬로 fetch (10개씩 배치)\n const batchSize = 10;\n for (let i = 1; i < numPages; i += batchSize) {\n const batch = [];\n for (let j = i; j < Math.min(i + batchSize, numPages); j++) {\n batch.push(\n loadDatasetPage(j * PAGE_SIZE, PAGE_SIZE).then((p) =>\n p.rows.map((r) => r.row)\n ),\n );\n }\n const results = await Promise.all(batch);\n for (const r of results) rows.push(...r);\n }\n\n return rows;\n}\n\nfunction makeBuckets(\n values: number[],\n edges: number[],\n): { range: string; count: number }[] {\n const buckets = edges.map((e, i) => ({\n range: i === 0 ? `0-${e}` : `${edges[i - 1] + 1}-${e}`,\n count: 0,\n }));\n buckets.push({\n range: `${edges[edges.length - 1] + 1}+`,\n count: 0,\n });\n\n for (const v of values) {\n let placed = false;\n for (let i = 0; i < edges.length; i++) {\n if (v <= edges[i]) {\n buckets[i].count++;\n placed = true;\n break;\n }\n }\n if (!placed) buckets[buckets.length - 1].count++;\n }\n\n return buckets;\n}\n\nexport async function computeAnalytics(): Promise {\n const start = Date.now();\n const rows = await fetchAllRows();\n const fetchDurationMs = Date.now() - start;\n\n const totalTraces = rows.length;\n let totalToolCalls = 0;\n let totalUserMessages = 0;\n let totalPromptLength = 0;\n let maxToolCalls = 0;\n let maxUserMessages = 0;\n let maxPromptLength = 0;\n\n const toolCallsValues: number[] = [];\n const userMessagesValues: number[] = [];\n const promptLengthValues: number[] = [];\n const turnCountValues: number[] = [];\n\n const harnessDist: Record = {};\n const modelDist: Record = {};\n const thinkingLevelDist: Record = {};\n const toolNames: Record = {};\n const tracesByDate: Record = {};\n\n for (const row of rows) {\n // 기본 통계\n totalToolCalls += row.num_tool_calls;\n totalUserMessages += row.num_user_messages;\n const promptLen = row.prompt.length;\n totalPromptLength += promptLen;\n\n maxToolCalls = Math.max(maxToolCalls, row.num_tool_calls);\n maxUserMessages = Math.max(maxUserMessages, row.num_user_messages);\n maxPromptLength = Math.max(maxPromptLength, promptLen);\n\n toolCallsValues.push(row.num_tool_calls);\n userMessagesValues.push(row.num_user_messages);\n promptLengthValues.push(promptLen);\n\n // harness 분포\n const h = row.harness || \"unknown\";\n harnessDist[h] = (harnessDist[h] || 0) + 1;\n\n // metadata에서 정보 추출\n const meta = row.metadata || {};\n const model = meta.model || \"unknown\";\n modelDist[model] = (modelDist[model] || 0) + 1;\n\n const thinkingLevel = meta.thinking_level || \"unknown\";\n thinkingLevelDist[thinkingLevel] = (thinkingLevelDist[thinkingLevel] || 0) +\n 1;\n\n const turnCount = meta.turn_count ?? 0;\n turnCountValues.push(turnCount);\n\n // 도구 이름 추출\n if (Array.isArray(row.tools)) {\n for (const tool of row.tools) {\n const name = tool?.function?.name ?? tool?.name ?? \"unknown\";\n toolNames[name] = (toolNames[name] || 0) + 1;\n }\n }\n\n // 날짜별 분포\n const date = row.sent_at?.slice(0, 10) ?? \"unknown\";\n tracesByDate[date] = (tracesByDate[date] || 0) + 1;\n }\n\n // 상위 15개 도구\n const topTools = Object.entries(toolNames)\n .map(([name, count]) => ({ name, count }))\n .sort((a, b) => b.count - a.count)\n .slice(0, 15);\n\n // 날짜별 정렬\n const tracesByDateArr = Object.entries(tracesByDate)\n .map(([date, count]) => ({ date, count }))\n .sort((a, b) => a.date.localeCompare(b.date));\n\n return {\n totalTraces,\n totalToolCalls,\n totalUserMessages,\n avgToolCalls: Math.round((totalToolCalls / totalTraces) * 100) / 100,\n avgUserMessages: Math.round((totalUserMessages / totalTraces) * 100) / 100,\n avgPromptLength: Math.round((totalPromptLength / totalTraces)),\n maxToolCalls,\n maxUserMessages,\n maxPromptLength,\n toolCallsBuckets: makeBuckets(toolCallsValues, [0, 5, 10, 20, 50, 100]),\n userMessagesBuckets: makeBuckets(userMessagesValues, [1, 3, 5, 10, 20, 50]),\n promptLengthBuckets: makeBuckets(promptLengthValues, [\n 500,\n 1000,\n 2000,\n 5000,\n 10000,\n 20000,\n ]),\n turnCountBuckets: makeBuckets(turnCountValues, [1, 5, 10, 20, 50, 100]),\n harnessDist,\n modelDist,\n thinkingLevelDist,\n topTools,\n tracesByDate: tracesByDateArr,\n fetchedAt: new Date().toISOString(),\n fetchDurationMs,\n };\n}\n",[],[138,176],["D",1783484057587],[156,157],[138],["D",1783483833950],["D",1789525245864],{"_16":17,"_87":83},{"_16":17,"_80":81,"_82":83,"_33":34,"_84":85,"_86":26},{"_68":69,"_20":70,"_18":71,"_72":19,"_73":-5,"_74":-5,"_75":-5,"_76":-5,"_77":-5,"_78":21,"_79":17},"ctx",{"_61":-5,"_108":224,"_225":226,"_227":228,"_229":26,"_230":26,"_231":26,"_232":38,"_233":-5,"_234":-7,"_235":-7,"_236":42,"_237":26,"_238":26,"_239":-5,"_240":26,"_241":26,"_242":26,"_243":-5},"ENV",{"_195":196,"_197":198,"_199":70,"_200":201,"_202":203,"_204":205,"_206":207,"_208":209,"_210":211,"_212":213,"_214":215,"_216":217,"_218":219,"_220":221,"_222":223},"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"]

NL → Code Agent

Fable-5-Traces 데이터셋 분석

도구 호출 횟수 분포

사용자 메시지 수 분포

프롬프트 길이 분포 (문자 수)

턴 카운트 분포

모델 분포

사고 수준 (Thinking Level)

상위 15개 도구

날짜별 트레이스 수