{props.title}

${para.map(inline).join(\"
\")}

`);\n para.length = 0;\n };\n let i = 0;\n while (i < lines.length) {\n const line = lines[i];\n\n if (line.trim() === \"\") {\n flush();\n i++;\n continue;\n }\n\n const fence = line.match(/^(`{3,}|~{3,})\\s*(\\S*)/);\n if (fence) {\n flush();\n const marker = fence[1];\n const lang = fence[2];\n const isClose = (l: string) => {\n const t = l.trimEnd();\n return t.length >= marker.length && [...t].every((c) => c === marker[0]);\n };\n const buf: string[] = [];\n i++;\n while (i < lines.length && !isClose(lines[i])) buf.push(lines[i++]);\n i++; // closing fence (or EOF)\n out.push(\n `

${escapeHtml(buf.join(\"\\n\"))}

`,\n );\n continue;\n }\n\n const heading = line.match(/^(#{1,6})\\s+(.+?)\\s*#*\\s*$/);\n if (heading) {\n flush();\n const n = heading[1].length;\n out.push(`${inline(heading[2])}`);\n i++;\n continue;\n }\n\n if (/^ {0,3}([-*_])( *\\1){2,}\\s*$/.test(line)) {\n flush();\n out.push(\"

\");\n i++;\n continue;\n }\n\n if (/^ {0,3}>/.test(line)) {\n flush();\n const buf: string[] = [];\n while (i < lines.length && /^ {0,3}>/.test(lines[i])) buf.push(lines[i++].replace(/^ {0,3}> ?/, \"\"));\n out.push(`

${renderBlocks(buf)}

`);\n continue;\n }\n\n if (\n line.includes(\"|\") && i + 1 < lines.length &&\n /^ {0,3}\\|?[ :|-]+\\|[ :|-]*$/.test(lines[i + 1]) && lines[i + 1].includes(\"-\")\n ) {\n flush();\n const row = (l: string) =>\n l.trim().replace(/^\\|/, \"\").replace(/\\|\\s*$/, \"\").split(\"|\").map((c) => inline(c.trim()));\n const head = row(line);\n i += 2;\n const rows: string[][] = [];\n while (i < lines.length && lines[i].includes(\"|\") && lines[i].trim() !== \"\") rows.push(row(lines[i++]));\n out.push(\n `${head.map((c) => ``).join(\"\")}` +\n rows.map((r) => `${r.map((c) => ``).join(\"\")}`).join(\"\") +\n `

${c}
${c}

`,\n );\n continue;\n }\n\n if (/^(\\s*)([-*+]|\\d{1,9}[.)])\\s+/.test(line)) {\n flush();\n const list = parseList(lines, i);\n out.push(list.html);\n i = list.next;\n continue;\n }\n\n para.push(line);\n i++;\n }\n flush();\n return out.join(\"\\n\");\n}\n\nfunction parseList(lines: string[], start: number): { html: string; next: number } {\n const itemRe = /^(\\s*)(?:([-*+])|(\\d{1,9})[.)])\\s+(.*)$/;\n type Frame = { indent: number; ordered: boolean; start: number; items: string[] };\n const html: string[] = [];\n const stack: Frame[] = [];\n const close = () => {\n const f = stack.pop()!;\n const tag = f.ordered ? \"ol\" : \"ul\";\n const attrs = f.ordered && f.start > 1 ? ` start=\"${f.start}\"` : \"\";\n const rendered = `<${tag}${attrs}>${f.items.map((x) => `

${x}

`).join(\"\")}`;\n if (stack.length) {\n const p = stack[stack.length - 1];\n p.items[p.items.length - 1] += rendered;\n } else html.push(rendered);\n };\n let i = start;\n while (i < lines.length) {\n const m = lines[i].match(itemRe);\n if (m) {\n const indent = m[1].length;\n const ordered = m[3] !== undefined;\n const num = ordered ? parseInt(m[3], 10) : 1;\n if (!stack.length || indent > stack[stack.length - 1].indent) {\n stack.push({ indent, ordered, start: num, items: [] });\n } else {\n while (stack.length > 1 && indent < stack[stack.length - 1].indent) close();\n if (ordered !== stack[stack.length - 1].ordered) {\n if (stack.length > 1) {\n close();\n stack.push({ indent, ordered, start: num, items: [] });\n } else {\n break; // top-level bullet-type switch: end here, caller starts a fresh list\n }\n }\n }\n stack[stack.length - 1].items.push(inline(m[4]));\n i++;\n continue;\n }\n if (lines[i].trim() === \"\") {\n // A blank line only continues the list when another item follows directly.\n if (i + 1 < lines.length && itemRe.test(lines[i + 1])) {\n i++;\n continue;\n }\n break;\n }\n const ind = lines[i].match(/^\\s*/)![0].length;\n if (ind > stack[stack.length - 1].indent) {\n const f = stack[stack.length - 1];\n f.items[f.items.length - 1] += \"
\" + inline(lines[i].trim());\n i++;\n continue;\n }\n break;\n }\n while (stack.length) close();\n return { html: html.join(\"\"), next: i };\n}\n",[18,105],"4cf8f2e1-1c5f-43d4-9d85-d0a5196ebf72",["D",1785513777448],"// Format-sniffing importer for raw agent-CLI transcripts. Supported today:\n// Claude Code session JSONL, Codex rollout JSONL, and OpenCode `opencode export`\n// JSON. Each converter is tolerant — unknown entry types are skipped, so minor\n// format drift degrades to a sparser transcript rather than an error.\nimport type { Message } from \"./store.ts\";\nimport { convertClaudeCodeJsonl, trim } from \"./claude-code.ts\";\n\nconst INPUT_LIMIT = 2_000;\nconst OUTPUT_LIMIT = 6_000;\n\nexport type ConvertedChat = { title: string | null; source: string; messages: Message[] };\n\n// Codex wraps injected context (env info, plugin ads, AGENTS.md) in XML-ish tags\n// inside user messages; none of it is conversation.\nconst CODEX_CONTEXT_TAGS = /<(environment_context|recommended_plugins|user_instructions|turn_context)>[\\s\\S]*?<\\/\\1>/g;\n\nfunction pushAssistant(messages: Message[], text: string) {\n const last = messages[messages.length - 1];\n if (last?.role === \"assistant\") last.content += \"\\n\\n\" + text;\n else messages.push({ role: \"assistant\", content: text });\n}\n\n// Codex tool outputs are often a JSON-encoded [{type:\"input_text\", text}] array.\nfunction codexOutputText(output: unknown): string {\n if (typeof output !== \"string\") return \"\";\n try {\n const parsed = JSON.parse(output);\n if (Array.isArray(parsed)) {\n const texts = parsed.map((b) => (b && typeof b.text === \"string\" ? b.text : \"\")).filter(Boolean);\n if (texts.length) return texts.join(\"\\n\");\n }\n } catch { /* plain text output */ }\n return output;\n}\n\nexport function convertCodexJsonl(jsonl: string): ConvertedChat {\n const messages: Message[] = [];\n const toolsByCallId = new Map();\n\n for (const line of jsonl.split(\"\\n\")) {\n if (!line.trim()) continue;\n let entry: any;\n try {\n entry = JSON.parse(line);\n } catch {\n continue;\n }\n // Newer rollouts wrap items in {type:\"response_item\", payload}; be tolerant\n // of older files where the item is the line itself.\n const item = entry.type === \"response_item\" ? entry.payload : entry;\n if (!item || typeof item !== \"object\") continue;\n\n if (item.type === \"message\" && Array.isArray(item.content)) {\n const text = item.content\n .map((b: any) => (typeof b?.text === \"string\" ? b.text : \"\"))\n .filter(Boolean)\n .join(\"\\n\")\n .replace(CODEX_CONTEXT_TAGS, \"\")\n .trim();\n if (!text) continue;\n if (item.role === \"user\") messages.push({ role: \"user\", content: text });\n else if (item.role === \"assistant\") pushAssistant(messages, text);\n // developer messages are injected instructions — skip\n } else if (item.type === \"function_call\" || item.type === \"custom_tool_call\") {\n const msg: Message = { role: \"tool\", name: String(item.name ?? \"tool\") };\n const input = typeof item.arguments === \"string\" ? item.arguments : typeof item.input === \"string\" ? item.input : \"\";\n if (input.trim()) msg.input = trim(input, INPUT_LIMIT);\n messages.push(msg);\n if (typeof item.call_id === \"string\") toolsByCallId.set(item.call_id, msg);\n } else if (item.type === \"function_call_output\" || item.type === \"custom_tool_call_output\") {\n const msg = toolsByCallId.get(item.call_id);\n const out = codexOutputText(item.output);\n if (msg && out.trim()) msg.output = trim(out, OUTPUT_LIMIT);\n } else if (item.type === \"local_shell_call\") {\n const cmd = Array.isArray(item.action?.command) ? item.action.command.join(\" \") : \"\";\n const msg: Message = { role: \"tool\", name: \"shell\" };\n if (cmd) msg.input = trim(cmd, INPUT_LIMIT);\n messages.push(msg);\n if (typeof item.call_id === \"string\") toolsByCallId.set(item.call_id, msg);\n }\n // reasoning, event_msg (duplicates of messages), session_meta, etc. — skip\n }\n\n return { title: null, source: \"codex\", messages };\n}\n\n// `opencode export` shape: {info: {title?}, messages: [{info: {role}, parts: [...]}]}\n// where parts are {type:\"text\", text}, {type:\"tool\", tool, state:{input, output}}, etc.\nexport function convertOpenCodeExport(doc: any): ConvertedChat {\n const messages: Message[] = [];\n for (const m of doc?.messages ?? []) {\n const role = m?.info?.role ?? m?.role;\n for (const p of m?.parts ?? []) {\n if (p?.type === \"text\" && typeof p.text === \"string\" && p.text.trim()) {\n if (role === \"user\") messages.push({ role: \"user\", content: p.text });\n else if (role === \"assistant\") pushAssistant(messages, p.text);\n } else if (p?.type === \"tool\") {\n const msg: Message = { role: \"tool\", name: String(p.tool ?? \"tool\") };\n const state = p.state ?? {};\n const input = typeof state.input === \"string\" ? state.input : state.input ? JSON.stringify(state.input, null, 2) : \"\";\n const output = typeof state.output === \"string\" ? state.output : state.output ? JSON.stringify(state.output, null, 2) : \"\";\n if (input.trim()) msg.input = trim(input, INPUT_LIMIT);\n if (output.trim()) msg.output = trim(output, OUTPUT_LIMIT);\n messages.push(msg);\n }\n // reasoning / step-start / snapshot parts — skip\n }\n }\n const title = typeof doc?.info?.title === \"string\" ? doc.info.title : null;\n return { title, source: \"opencode\", messages };\n}\n\n// pi (`@earendil-works/pi-coding-agent`) session JSONL: a flat event log.\n// Conversation lives in `message` events carrying `{message: {role, content,\n// toolCallId, ...}}`. Content parts are text / thinking / toolCall; tool\n// results are separate `toolResult` messages matched by toolCallId.\n// session / model_change / thinking_level_change / custom_message events are\n// scaffolding (or subagent sidechains) — skipped.\nexport function convertPiJsonl(jsonl: string): ConvertedChat {\n const messages: Message[] = [];\n const toolByCallId = new Map();\n\n const textOf = (content: unknown): string => {\n if (!Array.isArray(content)) return \"\";\n return content\n .map((b: any) => (typeof b?.text === \"string\" ? b.text : \"\"))\n .filter(Boolean)\n .join(\"\\n\")\n .trim();\n };\n\n for (const line of jsonl.split(\"\\n\")) {\n if (!line.trim()) continue;\n let entry: any;\n try {\n entry = JSON.parse(line);\n } catch {\n continue;\n }\n if (entry?.type !== \"message\") continue;\n const msg = entry.message;\n if (!msg || typeof msg !== \"object\") continue;\n\n if (msg.role === \"user\") {\n const text = textOf(msg.content);\n if (text) messages.push({ role: \"user\", content: text });\n } else if (msg.role === \"assistant\") {\n const text = textOf(msg.content);\n if (text) pushAssistant(messages, text);\n for (const part of msg.content ?? []) {\n if (part?.type !== \"toolCall\") continue;\n const input =\n typeof part.arguments === \"string\"\n ? part.arguments\n : part.arguments\n ? JSON.stringify(part.arguments, null, 2)\n : \"\";\n const tool: Message = { role: \"tool\", name: String(part.name ?? \"tool\") };\n if (input.trim()) tool.input = trim(input, INPUT_LIMIT);\n messages.push(tool);\n if (typeof part.id === \"string\") toolByCallId.set(part.id, tool);\n }\n } else if (msg.role === \"toolResult\") {\n const tool =\n typeof msg.toolCallId === \"string\"\n ? toolByCallId.get(msg.toolCallId)\n : undefined;\n const out = textOf(msg.content);\n if (tool && out.trim()) tool.output = trim(out, OUTPUT_LIMIT);\n }\n // other roles (system, …) — skip\n }\n\n return { title: null, source: \"pi\", messages };\n}\n\nexport function detectAndConvert(text: string): ConvertedChat | { error: string } {\n const trimmed = text.trim();\n if (!trimmed) return { error: \"Empty body.\" };\n\n // A single JSON document (not JSONL) → OpenCode export.\n if (trimmed.startsWith(\"{\") && !trimmed.includes(\"\\n{\")) {\n try {\n const doc = JSON.parse(trimmed);\n if (Array.isArray(doc?.messages)) return convertOpenCodeExport(doc);\n return { error: \"JSON object without a `messages` array — not a recognized transcript format.\" };\n } catch { /* fall through to JSONL sniffing */ }\n }\n\n // JSONL: sniff the first few parseable lines for a format signature.\n for (const line of trimmed.split(\"\\n\").slice(0, 25)) {\n try {\n const e = JSON.parse(line);\n if (e.type === \"response_item\" || e.type === \"session_meta\" || e.type === \"event_msg\" || e.type === \"turn_context\") {\n return convertCodexJsonl(text);\n }\n if (e.type === \"message\" && typeof e?.message?.role === \"string\") {\n return convertPiJsonl(text);\n }\n if ((e.type === \"user\" || e.type === \"assistant\") && e.message || e.type === \"ai-title\" || e.sessionId) {\n return convertClaudeCodeJsonl(text);\n }\n } catch {\n continue;\n }\n }\n return {\n error: \"Unrecognized transcript format. Supported: Claude Code session .jsonl, Codex rollout .jsonl, `opencode export` JSON, pi session .jsonl.\",\n };\n}\n",[18,197],"2b2672f2-02f7-4cda-abfa-6cc6596317a2",["D",1784821029227],"// Converts a raw Claude Code session transcript (~/.claude/projects//.jsonl)\n// into sharechat's chat-input shape, so agents/scripts can POST the file verbatim\n// instead of hand-retyping the conversation (which costs the model output tokens).\n//\n// Kept: user turns, assistant text, tool calls with their results (trimmed).\n// Dropped: thinking blocks, sidechain (subagent) traffic, system/meta entries,\n// system-reminder blocks injected into user turns, and local-command echoes.\nimport type { Message } from \"./store.ts\";\n\nconst INPUT_LIMIT = 2_000;\n// Output truncation tiers: retried smaller until the payload fits the 2 MB cap.\nconst OUTPUT_TIERS = [6_000, 2_000, 600];\nconst PAYLOAD_BUDGET = 1_900_000;\n\ntype ToolMsg = Message & { role: \"tool\" };\n\nexport function trim(s: string, limit: number): string {\n if (s.length <= limit) return s;\n return s.slice(0, limit) + `\\n… [truncated, ${s.length.toLocaleString()} chars total]`;\n}\n\nfunction stripSystemReminders(s: string): string {\n return s.replace(/[\\s\\S]*?<\\/system-reminder>/g, \"\").trim();\n}\n\n// tool_result content is a string or an array of {type:\"text\", text} blocks.\nfunction resultText(content: unknown): string {\n if (typeof content === \"string\") return content;\n if (Array.isArray(content)) {\n return content\n .map((b) => (b && typeof b === \"object\" && (b as any).type === \"text\" ? (b as any).text : \"\"))\n .filter(Boolean)\n .join(\"\\n\");\n }\n return \"\";\n}\n\nfunction inputText(name: string, input: unknown): string {\n if (input && typeof input === \"object\") {\n const o = input as Record;\n if (name === \"Bash\" && typeof o.command === \"string\") return o.command;\n if (typeof o.file_path === \"string\" && Object.keys(o).length === 1) return o.file_path;\n }\n try {\n return JSON.stringify(input, null, 2);\n } catch {\n return String(input);\n }\n}\n\nexport function convertClaudeCodeJsonl(\n jsonl: string,\n): { title: string | null; source: string; messages: Message[] } {\n let title: string | null = null;\n const messages: Message[] = [];\n const pendingTools = new Map();\n\n for (const line of jsonl.split(\"\\n\")) {\n if (!line.trim()) continue;\n let entry: any;\n try {\n entry = JSON.parse(line);\n } catch {\n continue;\n }\n if (entry.isSidechain) continue;\n\n if (entry.type === \"ai-title\" && typeof entry.aiTitle === \"string\") {\n title = entry.aiTitle;\n continue;\n }\n\n if (entry.type === \"assistant\") {\n const blocks = entry.message?.content;\n if (!Array.isArray(blocks)) continue;\n for (const b of blocks) {\n if (b?.type === \"text\" && typeof b.text === \"string\" && b.text.trim()) {\n const last = messages[messages.length - 1];\n if (last?.role === \"assistant\") last.content += \"\\n\\n\" + b.text;\n else messages.push({ role: \"assistant\", content: b.text });\n } else if (b?.type === \"tool_use\") {\n const msg: ToolMsg = { role: \"tool\", name: String(b.name ?? \"tool\") };\n const input = inputText(msg.name, b.input);\n if (input.trim()) msg.input = trim(input, INPUT_LIMIT);\n messages.push(msg);\n if (typeof b.id === \"string\") pendingTools.set(b.id, { msg, raw: \"\" });\n }\n }\n continue;\n }\n\n if (entry.type === \"user\" && !entry.isMeta) {\n const content = entry.message?.content;\n if (typeof content === \"string\") {\n const text = stripSystemReminders(content);\n if (!text || text.startsWith(\"\")) continue;\n messages.push({ role: \"user\", content: text });\n } else if (Array.isArray(content)) {\n for (const b of content) {\n if (b?.type === \"tool_result\") {\n const t = pendingTools.get(b.tool_use_id);\n if (t) t.raw = resultText(b.content);\n } else if (b?.type === \"text\" && typeof b.text === \"string\") {\n const text = stripSystemReminders(b.text);\n if (text) messages.push({ role: \"user\", content: text });\n }\n }\n }\n }\n }\n\n // Fill tool outputs at the largest truncation tier that keeps the payload under budget.\n for (const limit of OUTPUT_TIERS) {\n for (const { msg, raw } of pendingTools.values()) {\n if (raw.trim()) msg.output = trim(raw, limit);\n }\n if (new TextEncoder().encode(JSON.stringify(messages)).length <= PAYLOAD_BUDGET) break;\n }\n\n return { title, source: \"claude-code\", messages };\n}\n",[18,173],"b2126331-ae2d-417e-b33b-d5742d05990c",["D",1784820965379],"// Reads a request body as parsed JSON, transparently decompressing gzip/deflate.\n//\n// Why gzip matters here: Val Town sits behind Cloudflare, whose managed WAF\n// scans *uncompressed* POST bodies and anomaly-scores them. Rich technical chat\n// transcripts (shell commands, URLs, code) reliably trip that scoring and get a\n// 403 \"Blocked\" HTML page before the request ever reaches this val. Cloudflare\n// does not decompress request bodies, so a gzipped body sails through untouched —\n// we just have to decompress it ourselves. Agents posting coding chats should\n// prefer the gzip path (documented on the homepage).\n//\n// Is skipping the WAF safe? The WAF is defense-in-depth, not our security\n// boundary. Input safety lives in *our* code and holds regardless: parameterized\n// SQL (lib/store.ts) and an escape-first markdown renderer (lib/markdown.ts) that\n// escapes every char before emitting a tag, images downgraded to links, zero\n// client JS on chat pages. The WAF was only false-positiving on benign chat\n// content. Keep that safety intact and this stays sound.\n\nconst MAX_DECOMPRESSED = 8_000_000; // hard cap so a zip bomb can't OOM the isolate\n\nasync function readBounded(stream: ReadableStream): Promise {\n const reader = stream.getReader();\n const chunks: Uint8Array[] = [];\n let total = 0;\n for (;;) {\n const { done, value } = await reader.read();\n if (done) break;\n total += value.byteLength;\n if (total > MAX_DECOMPRESSED) {\n reader.cancel();\n throw new Error(\"decompressed body exceeds 8 MB\");\n }\n chunks.push(value);\n }\n const out = new Uint8Array(total);\n let off = 0;\n for (const c of chunks) {\n out.set(c, off);\n off += c.byteLength;\n }\n return out;\n}\n\nexport async function readTextBody(req: Request): Promise {\n const enc = (req.headers.get(\"content-encoding\") ?? \"\").toLowerCase();\n if (enc.includes(\"gzip\") || enc.includes(\"deflate\")) {\n const format = enc.includes(\"gzip\") ? \"gzip\" : \"deflate\";\n const raw = req.body;\n if (!raw) throw new Error(\"empty body\");\n const stream = raw.pipeThrough(new DecompressionStream(format));\n return new TextDecoder().decode(await readBounded(stream));\n }\n return await req.text();\n}\n\nexport async function readJsonBody(req: Request): Promise {\n return JSON.parse(await readTextBody(req));\n}\n",[18,115],"pulls","branches","forks","envVars",{"_344":26,"_346":26},"connections",{"_344":26,"_345":26},"count","unfilledCount","emptyCount",{"_49":463,"_13":454,"_78":449,"_59":-5,"_51":52,"_300":464,"_301":24,"_53":465,"_55":26,"_81":221,"_303":-5,"_83":30,"_304":-5,"_371":466,"_17":467},{"_49":455,"_13":456,"_78":457,"_59":449,"_51":52,"_300":458,"_301":24,"_53":459,"_55":26,"_81":21,"_303":460,"_83":30,"_304":-5,"_371":461,"_17":462},{"_49":446,"_13":447,"_78":448,"_59":449,"_51":52,"_300":262,"_301":24,"_53":450,"_55":99,"_81":21,"_303":451,"_83":30,"_304":-5,"_371":452,"_17":453},{"_49":299,"_13":18,"_78":220,"_59":-5,"_51":52,"_300":56,"_301":24,"_53":443,"_55":26,"_81":221,"_303":-5,"_83":30,"_304":-5,"_371":444,"_17":445},{"_49":333,"_13":115,"_78":114,"_59":220,"_51":52,"_300":176,"_301":24,"_53":440,"_55":56,"_81":82,"_303":335,"_83":30,"_304":-5,"_371":441,"_17":442},{"_49":329,"_13":173,"_78":172,"_59":220,"_51":52,"_300":183,"_301":24,"_53":437,"_55":318,"_81":82,"_303":331,"_83":30,"_304":-5,"_371":438,"_17":439},{"_49":325,"_13":197,"_78":196,"_59":220,"_51":52,"_300":215,"_301":24,"_53":434,"_55":90,"_81":82,"_303":327,"_83":30,"_304":-5,"_371":435,"_17":436},{"_49":321,"_13":105,"_78":104,"_59":220,"_51":52,"_300":99,"_301":24,"_53":431,"_55":26,"_81":82,"_303":323,"_83":30,"_304":-5,"_371":432,"_17":433},{"_49":316,"_13":80,"_78":79,"_59":220,"_51":52,"_300":126,"_301":24,"_53":428,"_55":318,"_81":82,"_303":319,"_83":30,"_304":-5,"_371":429,"_17":430},{"_49":312,"_13":96,"_78":95,"_59":220,"_51":52,"_300":159,"_301":24,"_53":425,"_55":99,"_81":82,"_303":314,"_83":30,"_304":-5,"_371":426,"_17":427},{"_49":420,"_13":401,"_78":394,"_59":-5,"_51":52,"_300":421,"_301":24,"_53":422,"_55":26,"_81":221,"_303":-5,"_83":30,"_304":-5,"_371":423,"_17":424},{"_49":411,"_13":412,"_78":413,"_59":394,"_51":52,"_300":414,"_301":24,"_53":415,"_55":416,"_81":82,"_303":417,"_83":30,"_304":-5,"_371":418,"_17":419},{"_49":402,"_13":403,"_78":404,"_59":394,"_51":52,"_300":405,"_301":24,"_53":406,"_55":407,"_81":82,"_303":408,"_83":30,"_304":-5,"_371":409,"_17":410},{"_49":391,"_13":392,"_78":393,"_59":394,"_51":52,"_300":395,"_301":24,"_53":396,"_55":397,"_81":82,"_303":398,"_83":30,"_304":-5,"_371":399,"_17":400},{"_49":383,"_13":384,"_78":385,"_59":-5,"_51":52,"_300":386,"_301":24,"_53":387,"_55":56,"_81":21,"_303":388,"_83":30,"_304":-5,"_371":389,"_17":390},{"_49":374,"_13":375,"_78":376,"_59":-5,"_51":52,"_300":377,"_301":24,"_53":378,"_55":379,"_81":21,"_303":380,"_83":30,"_304":-5,"_371":381,"_17":382},{"_49":364,"_13":365,"_78":234,"_59":-5,"_51":52,"_300":366,"_301":24,"_53":367,"_55":368,"_81":369,"_303":370,"_83":30,"_304":14,"_371":372,"_17":373},"0fd3f962-b6fc-4784-acc2-b8a0e20b0ed0","main.tsx",88,["D",1785513802940],15,"http","/** @jsxImportSource npm:hono@4/jsx */\n// chat.val.run — share an AI conversation as a clean, unlisted link.\n// Homepage doubles as agent instructions (instructions.md): browsers get HTML,\n// everything else gets raw markdown. No auth, no list endpoint, 128-bit IDs.\n// The docs use a {{BASE}} placeholder substituted with the request origin, so\n// the val works unchanged on any subdomain.\nimport { type Context, Hono } from \"npm:hono@4\";\nimport { parseVal, readFile } from \"https://esm.town/v/std/utils/index.ts\";\nimport { type Chat, createChat, deleteChat, getChat } from \"./lib/store.ts\";\nimport { parseChatInput } from \"./lib/validate.ts\";\nimport { readJsonBody, readTextBody } from \"./lib/body.ts\";\nimport { detectAndConvert } from \"./lib/import.ts\";\nimport { ChatPage } from \"./views/chat.tsx\";\nimport { NotFoundPage } from \"./views/layout.tsx\";\nimport { HomePage } from \"./views/home.tsx\";\n\nconst INSTRUCTIONS_MD = await readFile(\"/instructions.md\", import.meta.url);\nconst CLI_SH = await readFile(\"/cli/sharechat.sh\", import.meta.url);\nconst INSTALL_SH = await readFile(\"/cli/install.sh\", import.meta.url);\nconst MD = \"text/markdown; charset=utf-8\";\n\nconst originOf = (c: Context) => new URL(c.req.url).origin;\nconst hostOf = (c: Context) => new URL(c.req.url).host;\nconst instructionsFor = (c: Context) => INSTRUCTIONS_MD.replaceAll(\"{{BASE}}\", originOf(c));\nconst wantsHtml = (c: Context) => (c.req.header(\"accept\") ?? \"\").includes(\"text/html\");\n\nconst app = new Hono();\napp.onError((err) => Promise.reject(err));\n\napp.get(\"/\", (c) =>\n wantsHtml(c)\n ? c.html()\n : c.text(instructionsFor(c), 200, { \"content-type\": MD }));\n\n// The machine-readable instructions. `/llms.txt` is the convention an agent\n// probes; the bare GET above returns the same thing for a plain fetch/curl.\napp.get(\"/llms.txt\", (c) => c.text(instructionsFor(c), 200, { \"content-type\": MD }));\napp.get(\"/instructions.md\", (c) => c.text(instructionsFor(c), 200, { \"content-type\": MD }));\napp.get(\"/source\", (c) => c.redirect(parseVal().links.self.val));\n\n// The zero-token upload CLI and its installer, templated to this origin.\nconst SH = \"text/x-shellscript; charset=utf-8\";\napp.get(\"/sharechat.sh\", (c) => c.text(CLI_SH.replaceAll(\"{{BASE}}\", originOf(c)), 200, { \"content-type\": SH }));\napp.get(\"/install\", (c) => c.text(INSTALL_SH.replaceAll(\"{{BASE}}\", originOf(c)), 200, { \"content-type\": SH }));\n\nasync function handleCreate(c: Context) {\n const origin = originOf(c);\n let body: unknown;\n try {\n body = await readJsonBody(c.req.raw);\n } catch {\n return c.json({ ok: false, error: `Body must be valid JSON. GET ${origin} for the format.` }, 400);\n }\n return createAndRespond(c, body);\n}\n\nasync function createAndRespond(c: Context, body: unknown) {\n const origin = originOf(c);\n const parsed = parseChatInput(body);\n if (!parsed.ok) return c.json({ ok: false, error: parsed.error, docs: origin }, 400);\n const { id, deleteToken } = await createChat(parsed.value);\n return c.json({\n ok: true,\n id,\n url: `${origin}/c/${id}`,\n markdownUrl: `${origin}/c/${id}.md`,\n jsonUrl: `${origin}/c/${id}.json`,\n deleteUrl: `${origin}/c/${id}?token=${deleteToken}`,\n note: \"Give your human the `url` — that's the shareable link. To delete: curl -X DELETE ''\",\n }, 201);\n}\n\napp.post(\"/api/chats\", handleCreate);\napp.post(\"/\", handleCreate);\n\n// Accepts a raw agent-CLI transcript (ideally gzipped) and converts it\n// server-side — no client-side transformation or model retyping needed.\n// Format is auto-detected: Claude Code .jsonl, Codex rollout .jsonl,\n// `opencode export` JSON, or pi session .jsonl.\nasync function handleImport(c: Context) {\n let text: string;\n try {\n text = await readTextBody(c.req.raw);\n } catch (e) {\n return c.json({ ok: false, error: `Could not read body: ${(e as Error).message}` }, 400);\n }\n const input = detectAndConvert(text);\n if (\"error\" in input) return c.json({ ok: false, error: input.error }, 400);\n if (input.messages.length === 0) {\n return c.json({ ok: false, error: \"Transcript recognized but no conversation turns found in it.\" }, 400);\n }\n const title = c.req.query(\"title\") || input.title;\n return createAndRespond(c, { ...input, title });\n}\n\napp.post(\"/api/chats/import\", handleImport);\napp.post(\"/api/chats/claude-code\", handleImport);\n\n// Unlisted means unlisted: even if a link leaks to a crawler, don't index it.\napp.use(\"/c/*\", async (c, next) => {\n await next();\n c.res.headers.set(\"X-Robots-Tag\", \"noindex\");\n});\n\n// Fence code so it survives content that itself contains ``` fences.\nfunction fenced(s: string): string {\n const f = s.includes(\"```\") ? \"````\" : \"```\";\n return `${f}\\n${s}\\n${f}`;\n}\n\nfunction transcriptMd(chat: Chat, origin: string): string {\n const turns = chat.messages.filter((m) => m.role !== \"tool\").length;\n const tools = chat.messages.length - turns;\n const meta = `${turns} message${turns === 1 ? \"\" : \"s\"}${tools ? ` · ${tools} tool call${tools === 1 ? \"\" : \"s\"}` : \"\"} · ${chat.createdAt.slice(0, 10)}` +\n `${chat.source ? ` · via ${chat.source}` : \"\"} · ${origin}/c/${chat.id}`;\n const body = chat.messages\n .map((m) => {\n if (m.role === \"tool\") {\n const io = [m.input && fenced(m.input), m.output && fenced(m.output)].filter(Boolean).join(\"\\n\\n\");\n return `**⏺ ${m.name}**\\n\\n${io}\\n`;\n }\n return `---\\n\\n**${m.role === \"user\" ? \"You\" : chat.source || \"Assistant\"}:**\\n\\n${m.content}\\n`;\n })\n .join(\"\\n\");\n return `# ${chat.title}\\n\\n${meta}\\n\\n${body}`;\n}\n\napp.get(\"/c/:id\", async (c) => {\n let id = c.req.param(\"id\");\n let format: \"html\" | \"md\" | \"json\" | null = null;\n if (id.endsWith(\".json\")) {\n id = id.slice(0, -5);\n format = \"json\";\n } else if (id.endsWith(\".md\")) {\n id = id.slice(0, -3);\n format = \"md\";\n }\n if (!format) {\n const accept = c.req.header(\"accept\") ?? \"\";\n format = accept.includes(\"text/html\") ? \"html\" : accept.includes(\"application/json\") ? \"json\" : \"md\";\n }\n const chat = await getChat(id);\n if (!chat) {\n return format === \"html\"\n ? c.html(, 404)\n : c.json({ ok: false, error: \"Chat not found. It may have been deleted.\" }, 404);\n }\n if (format === \"json\") {\n return c.json({\n id: chat.id,\n title: chat.title,\n source: chat.source,\n createdAt: chat.createdAt,\n messages: chat.messages,\n });\n }\n if (format === \"md\") return c.text(transcriptMd(chat, originOf(c)), 200, { \"content-type\": MD });\n return c.html();\n});\n\napp.delete(\"/c/:id\", async (c) => {\n const id = c.req.param(\"id\").replace(/\\.(md|json)$/, \"\");\n const token = c.req.query(\"token\") ?? \"\";\n if (!token) {\n return c.json({ ok: false, error: \"Missing ?token= — use the deleteUrl returned when the chat was created.\" }, 400);\n }\n const result = await deleteChat(id, token);\n if (result === \"not_found\") return c.json({ ok: false, error: \"Chat not found (already deleted?)\" }, 404);\n if (result === \"bad_token\") return c.json({ ok: false, error: \"Invalid delete token.\" }, 403);\n return c.json({ ok: true, deleted: id });\n});\n\napp.notFound((c) =>\n wantsHtml(c)\n ? c.html(, 404)\n : c.text(`Not found. GET ${originOf(c)} for usage.\\n`, 404));\n\nexport default app.fetch;\n","children",[],[365],"e0c82413-53e0-4e30-88da-2f298def8c5b","instructions.md","b047e6ec-7aee-11f1-9a7b-1607ee4eb77e",90,["D",1785513835155],18,"---\nname: sharechat\ndescription: Publish an AI coding session as a clean, unlisted, deletable link. One command, zero model tokens.\n---\n\n# sharechat\n\nPublish an AI conversation at an unlisted URL. No account. Viewers need no login. Every upload returns a delete link. Agents welcome — this page is the documentation.\n\n## Fast path — POST the transcript file (use this by default)\n\nThe session is already on disk. Upload the raw file gzipped; the server detects the format and converts it. Do not retype the conversation, do not open a browser.\n\nIf you are Claude Code:\n\n```bash\nFILE=$(ls -t ~/.claude/projects/$(pwd | sed 's/[^a-zA-Z0-9]/-/g')/*.jsonl | head -1)\ngzip -c \"$FILE\" | curl -fsS {{BASE}}/api/chats/import -H 'content-encoding: gzip' --data-binary @-\n```\n\nIf you are Codex:\n\n```bash\nFILE=$(ls -t ~/.codex/sessions/*/*/*/rollout-*.jsonl | head -1)\ngzip -c \"$FILE\" | curl -fsS {{BASE}}/api/chats/import -H 'content-encoding: gzip' --data-binary @-\n```\n\nIf you are pi (the pi coding agent):\n\n```bash\nFILE=$(ls -t ~/.pi/agent/sessions/--$(pwd | sed 's/[^a-zA-Z0-9]/-/g' | sed 's/^-//')--/*.jsonl | head -1)\ngzip -c \"$FILE\" | curl -fsS {{BASE}}/api/chats/import -H 'content-encoding: gzip' --data-binary @-\n```\n\nIf you are opencode: `opencode export | gzip -c | curl ...` same as above. (opencode also has native `/share` — use whichever the human prefers.)\n\nThe response:\n\n```json\n{\"ok\":true,\"url\":\"{{BASE}}/c/\",\"markdownUrl\":\"…\",\"jsonUrl\":\"…\",\"deleteUrl\":\"{{BASE}}/c/?token=\"}\n```\n\nGive the human BOTH `url` and `deleteUrl`. Server-side, thinking blocks, system prompts, injected context, and subagent sidechains are dropped; tool calls are kept with trimmed inputs/outputs. Add `?title=...` to override the title. Gzip is required — the hosting layer may reject code-heavy uncompressed bodies.\n\n## No transcript file? Build the JSON\n\nFor web chats, or when the human asks for a curated or partial share, POST the messages yourself:\n\n```bash\ngzip -c <<'JSON' | curl -fsS {{BASE}}/api/chats \\\n -H 'content-type: application/json' -H 'content-encoding: gzip' --data-binary @-\n{\"title\":\"Optional\",\"source\":\"claude-code\",\"messages\":[{\"role\":\"user\",\"content\":\"Exact user message\"},{\"role\":\"assistant\",\"content\":\"Exact final answer\"}]}\nJSON\n```\n\nMessages verbatim, in order. Roles: `user`, `assistant`, and optionally `{\"role\":\"tool\",\"name\":\"Bash\",\"input\":\"plain text\",\"output\":\"plain text\"}` interleaved where they happened. Omit system prompts and hidden reasoning. Trim enormous tool outputs to the part that mattered. Never include secrets or credentials.\n\n## Links\n\n- View: `/c/` (HTML), `/c/.md`, `/c/.json`\n- Delete: `curl -X DELETE ''`\n- Links are unlisted snapshots: 128-bit random IDs, noindex, listed nowhere. Anyone who has the URL can view it — unlisted is not private. Messages after upload are never included; share again for an update. Only what is posted is published — files the conversation touched stay on your machine.\n\n## Humans\n\n- Skill — after this, \"share this chat\" just works in Claude Code, Codex, and Cursor: `npx skills add stevekrouse/sharechat`\n- CLI — `sharechat` uploads your newest session, zero model tokens: fetch `{{BASE}}/install` and pipe to `sh`. Script source: `{{BASE}}/sharechat.sh`\n- In Claude Code, type `! sharechat` to run it without spending a model turn.\n",[],[375],"3a1e16f4-ab7e-490f-b03d-7f7c9a1e4936","README.md","2e6a93a8-7aef-11f1-91b0-1607ee4eb77e",54,["D",1784215478729],"# sharechat.val.run\n\nShare an AI conversation as a clean, **unlisted** web page — like a ChatGPT/Claude share link, but for any agent (Claude Code, Codex, Cursor, …).\n\n**The homepage is the API doc.** Its copied prompt contains the browser-free `gzip | curl` fast path. `curl https://sharechat.val.run` returns agent instructions as markdown; browsers get the styled version. Agents read it, POST the dialog, and hand their human back a URL.\n\n## Endpoints\n\n- `POST /api/chats` (or `POST /`) — `{title?, source?, messages: [{role: \"user\"|\"assistant\", content} | {role: \"tool\", name, input?, output?}]}` → `{url, deleteUrl, …}`. Tool calls render as collapsed `⏺ Bash(…)` rows, like Claude Code's `/export`.\n- `GET /c/` — web page (browsers) or markdown transcript (everyone else)\n- `GET /c/.md` / `GET /c/.json` — explicit formats\n- `DELETE /c/?token=…` — permanent delete\n\n## Design\n\n- **Storage**: metadata in val-scoped SQLite (`chats_v1`), message bodies in val-scoped blob storage (`chat:`). Neither is public even though this code is.\n- **Unlisted**: 128-bit random IDs, no list endpoint, `noindex` on chat pages, `no-referrer` on outbound links. No auth — treat links as secrets, don't post credentials.\n- **Rendering**: server-side only, zero client JS. Markdown goes through an escape-first renderer ([lib/markdown.ts](lib/markdown.ts)) so posted content can't inject markup; images render as links so pages can't beacon viewers.\n- **UI**: follows the `valtown-ui` skill (IBM Plex Sans, black-on-white, slate + sky, 8px radius).\n",[],[384],"60ac5193-d8ab-4be1-a957-3bf83ff7619b","layout.tsx","1986b5ac-7aef-11f1-959b-1607ee4eb77e","157f0e0e-2693-4209-9101-95e8dff9b640",80,["D",1784824896703],11,"/** @jsxImportSource npm:hono@4/jsx */\n// HTML shell + all styling. Themed with CSS variables that flip on\n// prefers-color-scheme (system light/dark). Chat pages are built to read like a\n// ChatGPT/Claude share page: a near-invisible header, then the conversation as a\n// spacious document — user = soft right-aligned bubble, assistant = full-width\n// plain prose, no role labels. Base type is IBM Plex Sans (valtown-ui).\n\nconst FAVICON = \"data:image/svg+xml,\" + encodeURIComponent(\n ``,\n);\n\nconst CSS = `\n:root {\n /* --bubble is the user message; code surfaces are their own tokens so a user\n bubble never reads as the same \"thing\" as an inline-code chip / code block. */\n --bg: #ffffff;\n --text: #1c1c1e;\n --muted: #6b7280;\n --bubble: #e7ecf4; /* user bubble — subtly cool, distinct from code grays */\n --border: #e6e6e8;\n --card: #ffffff;\n --code-bg: #f5f6f8; /* fenced code blocks + table headers */\n --chip: #edeef1; /* inline code */\n --hover: #f0f1f3; /* interactive hover fill */\n --link: #0b62b8;\n}\n@media (prefers-color-scheme: dark) {\n :root {\n --bg: #1a1a1a;\n --text: #ececec;\n --muted: #9a9a9d;\n --bubble: #2b3040; /* cool blue-gray, clearly not code */\n --border: #343436;\n --card: #202021;\n --code-bg: #0e0e10; /* near-black code blocks */\n --chip: #33343a; /* inline code — lighter than page so it pops */\n --hover: #2a2a2c;\n --link: #7cc0f5;\n }\n}\n\n* { box-sizing: border-box; }\nbody { margin: 0; background: var(--bg); color: var(--text); font: 16px/1.6 \"IBM Plex Sans\", -apple-system, BlinkMacSystemFont, sans-serif; -webkit-font-smoothing: antialiased; }\n.wrap { max-width: 44rem; margin: 0 auto; padding: 40px 20px 120px; }\n.eyebrow { font-family: \"IBM Plex Mono\", ui-monospace, monospace; font-size: 10px; font-weight: 600; letter-spacing: .14em; text-transform: uppercase; color: var(--muted); }\na { color: var(--link); text-decoration: none; }\na:hover { text-decoration: underline; text-underline-offset: 2px; }\n.icon { width: 16px; height: 16px; flex: none; }\n\n/* ---- chat page ---- */\n.chat-head { margin-bottom: 36px; }\n.chat-title { font-size: 22px; font-weight: 600; line-height: 1.3; margin: 0 0 4px; letter-spacing: -0.01em; }\n.chat-meta { font-size: 13.5px; color: var(--muted); }\n\n.turn { margin: 30px 0; }\n.turn.user { display: flex; justify-content: flex-end; }\n.turn.user .bubble { background: var(--bubble); border-radius: 22px; padding: 10px 18px; max-width: 72%; }\n.turn.assistant .prose { max-width: 100%; font-size: 17px; line-height: 1.75; }\n.turn.assistant .prose p { margin: 16px 0; }\n\n/* ---- tool calls (collapsed /export-style rows) ---- */\n.turn.tool { margin: 10px 0; }\n.turn.tool + .turn.tool { margin-top: 2px; }\n.turn:not(.tool) + .turn.tool { margin-top: 22px; }\n.tool-call { font-family: \"IBM Plex Mono\", ui-monospace, monospace; font-size: 13px; }\n.tool-call summary { cursor: pointer; color: var(--muted); white-space: nowrap; overflow: hidden; text-overflow: ellipsis; list-style: none; border-radius: 8px; padding: 3px 8px; margin: 0 -8px; }\n.tool-call summary::-webkit-details-marker { display: none; }\n.tool-call summary:hover { background: var(--hover); }\n.tool-dot { font-size: 10px; vertical-align: 1px; }\n.tool-call[open] .tool-dot { color: var(--link); }\n.tool-name { color: var(--text); font-weight: 600; }\n.tool-arg { color: var(--muted); }\n.tool-io { background: var(--code-bg); border: 1px solid var(--border); border-radius: 8px; padding: 10px 12px; margin: 8px 0 8px 16px; overflow: auto; max-height: 420px; font-size: 12.5px; line-height: 1.55; white-space: pre-wrap; word-break: break-word; }\n.tool-io::before { content: attr(data-label); display: block; font-size: 9.5px; font-weight: 600; letter-spacing: .13em; text-transform: uppercase; color: var(--muted); margin-bottom: 6px; }\n\n/* ---- prose (rendered markdown) ---- */\n.prose > :first-child { margin-top: 0; }\n.prose > :last-child { margin-bottom: 0; }\n.prose p { margin: 12px 0; }\n.prose h1, .prose h2, .prose h3, .prose h4, .prose h5, .prose h6 { margin: 26px 0 8px; line-height: 1.3; }\n.prose h1 { font-size: 1.35em; }\n.prose h2 { font-size: 1.2em; }\n.prose h3 { font-size: 1.08em; }\n.prose h4, .prose h5, .prose h6 { font-size: 1em; }\n.prose pre { background: var(--code-bg); border: 1px solid var(--border); border-radius: 10px; padding: 14px 16px; overflow-x: auto; margin: 16px 0; }\n.prose pre code { background: none; border: none; padding: 0; font-size: 13.5px; line-height: 1.55; }\n.prose code { font-family: \"IBM Plex Mono\", ui-monospace, monospace; font-size: 0.85em; background: var(--chip); border-radius: 5px; padding: 1.5px 5px; }\n.prose blockquote { border-left: 3px solid var(--border); margin: 16px 0; padding: 0 0 0 16px; color: var(--muted); }\n.prose ul, .prose ol { margin: 14px 0; padding-left: 26px; }\n.prose li { margin: 6px 0; }\n.prose table { border-collapse: collapse; margin: 16px 0; font-size: 0.9em; display: block; overflow-x: auto; max-width: 100%; }\n.prose th, .prose td { border: 1px solid var(--border); padding: 7px 12px; text-align: left; vertical-align: top; color: var(--text); }\n.prose th { background: var(--code-bg); font-weight: 600; }\n.prose hr { border: none; border-top: 1px solid var(--border); margin: 28px 0; }\n\nfooter { margin-top: 72px; padding-top: 16px; border-top: 1px solid var(--border); font-size: 13px; color: var(--muted); display: flex; justify-content: space-between; gap: 16px; flex-wrap: wrap; }\n\n/* ---- homepage (document style) ---- */\n.doc-head { display: flex; align-items: baseline; gap: 20px; margin-top: 18px; }\n.doc-name { font-size: 24px; font-weight: 700; letter-spacing: -0.01em; margin: 0; }\n.doc-nav { display: flex; gap: 14px; font-size: 13.5px; }\n.doc-nav a { color: var(--muted); }\n.doc-nav a:hover { color: var(--link); }\n.doc-lede { color: var(--muted); font-size: 15.5px; margin: 10px 0 0; }\n\n.doc-section { margin-top: 34px; }\n.doc-section h2 { font-size: 15.5px; font-weight: 600; margin: 0 0 10px; }\n.doc-section p { font-size: 15px; line-height: 1.65; margin: 10px 0; }\n.doc-section code { font-family: \"IBM Plex Mono\", ui-monospace, monospace; font-size: 13px; background: var(--chip); border-radius: 4px; padding: 1.5px 5px; }\n.doc-between { color: var(--muted); font-size: 13.5px; margin: 8px 2px; }\n\n.code-row { display: flex; align-items: center; gap: 12px; background: var(--code-bg); border-radius: 8px; padding: 10px 8px 10px 14px; margin: 10px 0; }\n.code-row .code-text { flex: 1; font-family: \"IBM Plex Mono\", ui-monospace, monospace; font-size: 13px; overflow-x: auto; white-space: nowrap; }\n.code-copy { display: inline-flex; align-items: center; justify-content: center; background: none; border: none; border-radius: 6px; padding: 6px; color: var(--muted); cursor: pointer; }\n.code-copy:hover { background: var(--hover); color: var(--text); }\n`;\n\nexport function Layout(props: { title: string; noindex?: boolean; children?: unknown }) {\n return (\n \n \n \n \n \n {/* Chat URLs are the secret — never leak them via Referer on outbound links. */}\n \n {props.noindex ? : null}\n {props.title}\n \n \n \n \n

History · lib