@ai-agent-forge/plugin-search 0.0.0-stage → 0.88.15

package/dist/http.d.ts

@@ -0,0 +1,65 @@
+/**
+ * Bounded HTTP fetch for the search tool (搜索工具设计 §6): http(s) only,
+ * manual redirects, per-page timeout/size caps, no cookies or provider
+ * credentials. Domain/private-network policy is deliberately out of scope;
+ * deployments that need it install a policy plugin before the tool call.
+ */
+import type { NetworkRouteConfigV1 } from "@agent-forge/plugin-sdk";
+export interface FetchResponseLike {
+    readonly status: number;
+    readonly headers: {
+        get(name: string): string | null;
+    };
+    readonly body?: AsyncIterable<Uint8Array> | null;
+    text(): Promise<string>;
+}
+export type FetchImpl = (url: string, init: {
+    method?: string;
+    headers?: Record<string, string>;
+    redirect: "manual";
+    signal: AbortSignal;
+    dispatcher?: unknown;
+}) => Promise<FetchResponseLike>;
+export type PageErrorCodeV1 = "invalid_url" | "invalid_redirect" | "too_many_redirects" | "http_error" | "timeout" | "network_error" | "too_large" | "unsupported_content_type" | "empty_content" | "cancelled";
+export declare class PageError extends Error {
+    readonly code: PageErrorCodeV1;
+    readonly httpStatus?: number;
+    /** Raw bytes consumed before the failure, when the body read reached them. */
+    readonly bytesRead?: number;
+    constructor(code: PageErrorCodeV1, message: string, httpStatus?: number, bytesRead?: number);
+}
+export interface FetchedPageV1 {
+    readonly finalUrl: string;
+    readonly title?: string;
+    readonly contentType: string;
+    readonly text: string;
+    readonly truncated: boolean;
+    readonly bytesFetched: number;
+}
+export interface FetchPageOptionsV1 {
+    readonly url: string;
+    readonly maxTextBytes: number;
+    readonly maxRawBytes: number;
+    readonly timeoutMs: number;
+    readonly signal?: AbortSignal;
+    readonly network?: NetworkRouteConfigV1;
+    readonly fetchImpl?: FetchImpl;
+    readonly dispatchers?: DispatcherPool;
+}
+/**
+ * Lazily-created undici dispatchers; closed by the caller. Direct routes get an
+ * explicit Agent so the host's process-wide env-proxy dispatcher cannot
+ * silently re-route a settings `direct` mode or a bypassed host.
+ */
+export declare class DispatcherPool {
+    private readonly agents;
+    private directAgent;
+    private readonly network;
+    private readonly enabled;
+    constructor(network: NetworkRouteConfigV1 | undefined, enabled: boolean);
+    get(host: string): unknown | undefined;
+    close(): Promise<void>;
+}
+/** Fetch and extract one page; bounded and explicit about every failure. */
+export declare function fetchPage(options: FetchPageOptionsV1): Promise<FetchedPageV1>;
+//# sourceMappingURL=http.d.ts.map

package/dist/http.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"http.d.ts","sourceRoot":"","sources":["../src/http.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,yBAAyB,CAAC;AAOpE,MAAM,WAAW,iBAAiB;IACjC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,OAAO,EAAE;QAAE,GAAG,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAAA;KAAE,CAAC;IACvD,QAAQ,CAAC,IAAI,CAAC,EAAE,aAAa,CAAC,UAAU,CAAC,GAAG,IAAI,CAAC;IACjD,IAAI,IAAI,OAAO,CAAC,MAAM,CAAC,CAAC;CACxB;AAED,MAAM,MAAM,SAAS,GAAG,CACvB,GAAG,EAAE,MAAM,EACX,IAAI,EAAE;IACL,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,QAAQ,EAAE,QAAQ,CAAC;IACnB,MAAM,EAAE,WAAW,CAAC;IACpB,UAAU,CAAC,EAAE,OAAO,CAAC;CACrB,KACG,OAAO,CAAC,iBAAiB,CAAC,CAAC;AAEhC,MAAM,MAAM,eAAe,GACxB,aAAa,GACb,kBAAkB,GAClB,oBAAoB,GACpB,YAAY,GACZ,SAAS,GACT,eAAe,GACf,WAAW,GACX,0BAA0B,GAC1B,eAAe,GACf,WAAW,CAAC;AAEf,qBAAa,SAAU,SAAQ,KAAK;IACnC,QAAQ,CAAC,IAAI,EAAE,eAAe,CAAC;IAC/B,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAC7B,8EAA8E;IAC9E,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAC;IAE5B,YAAY,IAAI,EAAE,eAAe,EAAE,OAAO,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,MAAM,EAM1F;CACD;AAED,MAAM,WAAW,aAAa;IAC7B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,SAAS,EAAE,OAAO,CAAC;IAC5B,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;CAC9B;AAED,MAAM,WAAW,kBAAkB;IAClC,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,MAAM,CAAC,EAAE,WAAW,CAAC;IAC9B,QAAQ,CAAC,OAAO,CAAC,EAAE,oBAAoB,CAAC;IACxC,QAAQ,CAAC,SAAS,CAAC,EAAE,SAAS,CAAC;IAC/B,QAAQ,CAAC,WAAW,CAAC,EAAE,cAAc,CAAC;CACtC;AAqBD;;;;GAIG;AACH,qBAAa,cAAc;IAC1B,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAiC;IACxD,OAAO,CAAC,WAAW,CAAoB;IACvC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAmC;IAC3D,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAU;IAElC,YAAY,OAAO,EAAE,oBAAoB,GAAG,SAAS,EAAE,OAAO,EAAE,OAAO,EAGtE;IAED,GAAG,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,GAAG,SAAS,CAsBrC;IAEK,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,CAO3B;CACD;AA2LD,4EAA4E;AAC5E,wBAAsB,SAAS,CAAC,OAAO,EAAE,kBAAkB,GAAG,OAAO,CAAC,aAAa,CAAC,CAqDnF","sourcesContent":["/**\n * Bounded HTTP fetch for the search tool (搜索工具设计 §6): http(s) only,\n * manual redirects, per-page timeout/size caps, no cookies or provider\n * credentials. Domain/private-network policy is deliberately out of scope;\n * deployments that need it install a policy plugin before the tool call.\n */\n\nimport type { NetworkRouteConfigV1 } from \"@agent-forge/plugin-sdk\";\nimport { resolveNetworkRouteV1 } from \"@agent-forge/plugin-sdk\";\nimport { Agent, ProxyAgent, fetch as undiciFetch } from \"undici\";\nimport { extractReadablePage } from \"./extract.ts\";\nimport { SEARCH_LIMITS } from \"./limits.ts\";\nimport { byteLength, truncateToBytes } from \"./text.ts\";\n\nexport interface FetchResponseLike {\n\treadonly status: number;\n\treadonly headers: { get(name: string): string | null };\n\treadonly body?: AsyncIterable<Uint8Array> | null;\n\ttext(): Promise<string>;\n}\n\nexport type FetchImpl = (\n\turl: string,\n\tinit: {\n\t\tmethod?: string;\n\t\theaders?: Record<string, string>;\n\t\tredirect: \"manual\";\n\t\tsignal: AbortSignal;\n\t\tdispatcher?: unknown;\n\t},\n) => Promise<FetchResponseLike>;\n\nexport type PageErrorCodeV1 =\n\t| \"invalid_url\"\n\t| \"invalid_redirect\"\n\t| \"too_many_redirects\"\n\t| \"http_error\"\n\t| \"timeout\"\n\t| \"network_error\"\n\t| \"too_large\"\n\t| \"unsupported_content_type\"\n\t| \"empty_content\"\n\t| \"cancelled\";\n\nexport class PageError extends Error {\n\treadonly code: PageErrorCodeV1;\n\treadonly httpStatus?: number;\n\t/** Raw bytes consumed before the failure, when the body read reached them. */\n\treadonly bytesRead?: number;\n\n\tconstructor(code: PageErrorCodeV1, message: string, httpStatus?: number, bytesRead?: number) {\n\t\tsuper(message);\n\t\tthis.name = \"PageError\";\n\t\tthis.code = code;\n\t\tif (httpStatus !== undefined) this.httpStatus = httpStatus;\n\t\tif (bytesRead !== undefined) this.bytesRead = bytesRead;\n\t}\n}\n\nexport interface FetchedPageV1 {\n\treadonly finalUrl: string;\n\treadonly title?: string;\n\treadonly contentType: string;\n\treadonly text: string;\n\treadonly truncated: boolean;\n\treadonly bytesFetched: number;\n}\n\nexport interface FetchPageOptionsV1 {\n\treadonly url: string;\n\treadonly maxTextBytes: number;\n\treadonly maxRawBytes: number;\n\treadonly timeoutMs: number;\n\treadonly signal?: AbortSignal;\n\treadonly network?: NetworkRouteConfigV1;\n\treadonly fetchImpl?: FetchImpl;\n\treadonly dispatchers?: DispatcherPool;\n}\n\nconst ALLOWED_CONTENT_TYPES: ReadonlySet<string> = new Set([\n\t\"text/html\",\n\t\"application/xhtml+xml\",\n\t\"text/plain\",\n\t\"application/json\",\n]);\n\nconst MIN_MEANINGFUL_TEXT_CHARS = 300;\nconst JS_SHELL_RAW_BYTES = 10_000;\nconst JS_SHELL_TEXT_RATIO = 0.01;\n\nconst DEFAULT_HEADERS: Readonly<Record<string, string>> = {\n\t\"user-agent\": \"agent-forge-search/1.0\",\n\taccept: \"text/html, application/xhtml+xml, text/plain, application/json;q=0.9\",\n\t// Keep zstd out of the negotiation: undici advertises it, but Node runtimes\n\t// without zlib zstd decoding fail the whole response stream.\n\t\"accept-encoding\": \"gzip, deflate, br\",\n};\n\n/**\n * Lazily-created undici dispatchers; closed by the caller. Direct routes get an\n * explicit Agent so the host's process-wide env-proxy dispatcher cannot\n * silently re-route a settings `direct` mode or a bypassed host.\n */\nexport class DispatcherPool {\n\tprivate readonly agents = new Map<string, ProxyAgent>();\n\tprivate directAgent: Agent | undefined;\n\tprivate readonly network: NetworkRouteConfigV1 | undefined;\n\tprivate readonly enabled: boolean;\n\n\tconstructor(network: NetworkRouteConfigV1 | undefined, enabled: boolean) {\n\t\tthis.network = network;\n\t\tthis.enabled = enabled;\n\t}\n\n\tget(host: string): unknown | undefined {\n\t\tif (!this.enabled || this.network === undefined) return undefined;\n\t\tconst resolution = resolveNetworkRouteV1(this.network, host);\n\t\tif (resolution.mode !== \"proxy\" || resolution.proxy === undefined) {\n\t\t\tthis.directAgent ??= new Agent();\n\t\t\treturn this.directAgent;\n\t\t}\n\t\tconst key = `${resolution.proxy.url}\\u0000${resolution.proxy.username ?? \"\"}\\u0000${resolution.proxy.password ?? \"\"}`;\n\t\tconst existing = this.agents.get(key);\n\t\tif (existing !== undefined) return existing;\n\t\tconst credentials =\n\t\t\tresolution.proxy.username === undefined && resolution.proxy.password === undefined\n\t\t\t\t? undefined\n\t\t\t\t: `${resolution.proxy.username ?? \"\"}:${resolution.proxy.password ?? \"\"}`;\n\t\tconst agent = new ProxyAgent({\n\t\t\turi: resolution.proxy.url,\n\t\t\t...(credentials === undefined\n\t\t\t\t? {}\n\t\t\t\t: { token: `Basic ${Buffer.from(credentials, \"utf8\").toString(\"base64\")}` }),\n\t\t});\n\t\tthis.agents.set(key, agent);\n\t\treturn agent;\n\t}\n\n\tasync close(): Promise<void> {\n\t\tawait Promise.all([...this.agents.values()].map((agent) => agent.close()));\n\t\tthis.agents.clear();\n\t\tif (this.directAgent !== undefined) {\n\t\t\tawait this.directAgent.close();\n\t\t\tthis.directAgent = undefined;\n\t\t}\n\t}\n}\n\nconst defaultFetch: FetchImpl = (url, init) =>\n\tundiciFetch(url, init as Parameters<typeof undiciFetch>[1]) as unknown as Promise<FetchResponseLike>;\n\nfunction parseHttpUrl(raw: string): URL {\n\tlet url: URL;\n\ttry {\n\t\turl = new URL(raw);\n\t} catch {\n\t\tthrow new PageError(\"invalid_url\", `unparseable URL: ${JSON.stringify(raw)}`);\n\t}\n\tif (url.protocol !== \"http:\" && url.protocol !== \"https:\") {\n\t\tthrow new PageError(\"invalid_url\", `only http(s) URLs are supported: ${JSON.stringify(raw)}`);\n\t}\n\treturn url;\n}\n\nasync function readBodyCapped(response: FetchResponseLike, maxBytes: number): Promise<string> {\n\tif (response.body === undefined || response.body === null) {\n\t\tconst text = await response.text();\n\t\tconst size = byteLength(text);\n\t\tif (size > maxBytes) {\n\t\t\tthrow new PageError(\"too_large\", `response body exceeds the ${maxBytes} byte limit`, undefined, size);\n\t\t}\n\t\treturn text;\n\t}\n\tconst chunks: Uint8Array[] = [];\n\tlet total = 0;\n\ttry {\n\t\tfor await (const chunk of response.body) {\n\t\t\ttotal += chunk.byteLength;\n\t\t\tif (total > maxBytes) {\n\t\t\t\tthrow new PageError(\"too_large\", `response body exceeds the ${maxBytes} byte limit`, undefined, total);\n\t\t\t}\n\t\t\tchunks.push(chunk);\n\t\t}\n\t} catch (error) {\n\t\tif (error instanceof PageError) throw error;\n\t\tthrow new PageError(\"network_error\", error instanceof Error ? error.message : String(error), undefined, total);\n\t}\n\treturn Buffer.concat(chunks).toString(\"utf8\");\n}\n\nfunction contentTypeOf(response: FetchResponseLike): string {\n\tconst header = response.headers.get(\"content-type\") ?? \"\";\n\treturn header.split(\";\")[0]?.trim().toLowerCase() ?? \"\";\n}\n\ninterface FetchOnceResultV1 {\n\treadonly finalUrl: string;\n\treadonly contentType: string;\n\treadonly bodyText: string;\n\treadonly bytesFetched: number;\n}\n\nasync function fetchOnce(options: FetchPageOptionsV1, deadlineAt: number): Promise<FetchOnceResultV1> {\n\tconst fetchImpl = options.fetchImpl ?? defaultFetch;\n\tconst start = parseHttpUrl(options.url);\n\tlet current = start;\n\tfor (let hop = 0; hop <= SEARCH_LIMITS.maxRedirects; hop += 1) {\n\t\tconst remaining = deadlineAt - Date.now();\n\t\tif (remaining <= 0) {\n\t\t\tthrow new PageError(\"timeout\", `request timed out after ${options.timeoutMs}ms`);\n\t\t}\n\t\tconst controller = new AbortController();\n\t\tconst timer = setTimeout(() => controller.abort(), remaining);\n\t\tconst abortFromCaller = (): void => controller.abort();\n\t\toptions.signal?.addEventListener(\"abort\", abortFromCaller, { once: true });\n\t\ttry {\n\t\t\tconst headers: Record<string, string> = { ...DEFAULT_HEADERS };\n\t\t\tconst dispatcher = options.dispatchers?.get(current.hostname);\n\t\t\tlet response: FetchResponseLike;\n\t\t\ttry {\n\t\t\t\tresponse = await fetchImpl(current.toString(), {\n\t\t\t\t\tmethod: \"GET\",\n\t\t\t\t\theaders,\n\t\t\t\t\tredirect: \"manual\",\n\t\t\t\t\tsignal: controller.signal,\n\t\t\t\t\t...(dispatcher === undefined ? {} : { dispatcher }),\n\t\t\t\t});\n\t\t\t} catch (error) {\n\t\t\t\tconst bytesRead = error instanceof PageError ? error.bytesRead : undefined;\n\t\t\t\tif (options.signal?.aborted === true) {\n\t\t\t\t\tthrow new PageError(\"cancelled\", \"request cancelled by caller\", undefined, bytesRead);\n\t\t\t\t}\n\t\t\t\tif (controller.signal.aborted) {\n\t\t\t\t\tthrow new PageError(\"timeout\", `request timed out after ${options.timeoutMs}ms`, undefined, bytesRead);\n\t\t\t\t}\n\t\t\t\tthrow new PageError(\n\t\t\t\t\t\"network_error\",\n\t\t\t\t\t`request failed: ${error instanceof Error ? error.message : String(error)}`,\n\t\t\t\t\tundefined,\n\t\t\t\t\tbytesRead,\n\t\t\t\t);\n\t\t\t}\n\n\t\t\tif (response.status >= 300 && response.status < 400) {\n\t\t\t\tconst location = response.headers.get(\"location\");\n\t\t\t\tif (location === null || location.trim() === \"\") {\n\t\t\t\t\tthrow new PageError(\"invalid_redirect\", `redirect ${response.status} without Location`);\n\t\t\t\t}\n\t\t\t\tif (hop === SEARCH_LIMITS.maxRedirects) {\n\t\t\t\t\tthrow new PageError(\"too_many_redirects\", `more than ${SEARCH_LIMITS.maxRedirects} redirects`);\n\t\t\t\t}\n\t\t\t\tlet next: URL;\n\t\t\t\ttry {\n\t\t\t\t\tnext = new URL(location, current);\n\t\t\t\t} catch {\n\t\t\t\t\tthrow new PageError(\"invalid_redirect\", `redirect target is not a URL: ${JSON.stringify(location)}`);\n\t\t\t\t}\n\t\t\t\tif (next.protocol !== \"http:\" && next.protocol !== \"https:\") {\n\t\t\t\t\tthrow new PageError(\"invalid_redirect\", `redirect target must be http(s): ${next.toString()}`);\n\t\t\t\t}\n\t\t\t\tcurrent = next;\n\t\t\t\tcontinue;\n\t\t\t}\n\n\t\t\tif (response.status < 200 || response.status >= 300) {\n\t\t\t\tthrow new PageError(\"http_error\", `HTTP ${response.status}`, response.status);\n\t\t\t}\n\n\t\t\tconst contentType = contentTypeOf(response);\n\t\t\tif (!ALLOWED_CONTENT_TYPES.has(contentType)) {\n\t\t\t\tthrow new PageError(\n\t\t\t\t\t\"unsupported_content_type\",\n\t\t\t\t\t`content type ${contentType === \"\" ? \"(missing)\" : contentType} is not supported`,\n\t\t\t\t);\n\t\t\t}\n\t\t\tconst declaredLength = response.headers.get(\"content-length\");\n\t\t\tif (\n\t\t\t\tdeclaredLength !== null &&\n\t\t\t\tNumber.isInteger(Number(declaredLength)) &&\n\t\t\t\tNumber(declaredLength) > options.maxRawBytes\n\t\t\t) {\n\t\t\t\tthrow new PageError(\"too_large\", `response declares ${declaredLength} bytes`);\n\t\t\t}\n\t\t\tlet bodyText: string;\n\t\t\ttry {\n\t\t\t\tbodyText = await readBodyCapped(response, options.maxRawBytes);\n\t\t\t} catch (error) {\n\t\t\t\tconst bytesRead = error instanceof PageError ? error.bytesRead : undefined;\n\t\t\t\tif (options.signal?.aborted === true) {\n\t\t\t\t\tthrow new PageError(\"cancelled\", \"request cancelled by caller\", undefined, bytesRead);\n\t\t\t\t}\n\t\t\t\tif (controller.signal.aborted) {\n\t\t\t\t\tthrow new PageError(\"timeout\", `request timed out after ${options.timeoutMs}ms`, undefined, bytesRead);\n\t\t\t\t}\n\t\t\t\tthrow error;\n\t\t\t}\n\t\t\treturn {\n\t\t\t\tfinalUrl: current.toString(),\n\t\t\t\tcontentType,\n\t\t\t\tbodyText,\n\t\t\t\tbytesFetched: byteLength(bodyText),\n\t\t\t};\n\t\t} finally {\n\t\t\tclearTimeout(timer);\n\t\t\toptions.signal?.removeEventListener(\"abort\", abortFromCaller);\n\t\t}\n\t}\n\tthrow new PageError(\"too_many_redirects\", `more than ${SEARCH_LIMITS.maxRedirects} redirects`);\n}\n\nfunction extractPage(contentType: string, bodyText: string): { title?: string; text: string } {\n\tif (contentType === \"text/html\" || contentType === \"application/xhtml+xml\") {\n\t\tconst page = extractReadablePage(bodyText);\n\t\tif (page === undefined || page.text.trim() === \"\") {\n\t\t\tthrow new PageError(\"empty_content\", \"no readable text extracted; the page may require JavaScript\");\n\t\t}\n\t\treturn { ...(page.title === undefined ? {} : { title: page.title }), text: page.text };\n\t}\n\tconst text = bodyText.trim();\n\tif (text === \"\") throw new PageError(\"empty_content\", \"empty response body\");\n\treturn { text };\n}\n\nfunction isRetryable(error: unknown): boolean {\n\tif (!(error instanceof PageError)) return false;\n\tif (error.code === \"network_error\") return true;\n\treturn (\n\t\terror.code === \"http_error\" &&\n\t\terror.httpStatus !== undefined &&\n\t\t(error.httpStatus === 429 || error.httpStatus >= 500)\n\t);\n}\n\n/** Fetch and extract one page; bounded and explicit about every failure. */\nexport async function fetchPage(options: FetchPageOptionsV1): Promise<FetchedPageV1> {\n\tif (options.signal?.aborted === true) {\n\t\tthrow new PageError(\"cancelled\", \"request cancelled by caller\");\n\t}\n\tconst deadlineAt = Date.now() + options.timeoutMs;\n\tlet bytesRead = 0;\n\tlet lastError: unknown;\n\tfor (let attempt = 0; attempt < 2; attempt += 1) {\n\t\ttry {\n\t\t\tconst remainingRawBytes = Math.max(0, options.maxRawBytes - bytesRead);\n\t\t\tconst result = await fetchOnce({ ...options, maxRawBytes: remainingRawBytes }, deadlineAt);\n\t\t\tlet extracted: { title?: string; text: string };\n\t\t\ttry {\n\t\t\t\textracted = extractPage(result.contentType, result.bodyText);\n\t\t\t} catch (error) {\n\t\t\t\tif (error instanceof PageError) {\n\t\t\t\t\tthrow new PageError(error.code, error.message, error.httpStatus, result.bytesFetched);\n\t\t\t\t}\n\t\t\t\tthrow error;\n\t\t\t}\n\t\t\tconst meaningfulChars = extracted.text.trim().length;\n\t\t\tif (\n\t\t\t\tmeaningfulChars < MIN_MEANINGFUL_TEXT_CHARS &&\n\t\t\t\tresult.bytesFetched > JS_SHELL_RAW_BYTES &&\n\t\t\t\tmeaningfulChars / result.bytesFetched < JS_SHELL_TEXT_RATIO\n\t\t\t) {\n\t\t\t\tthrow new PageError(\n\t\t\t\t\t\"empty_content\",\n\t\t\t\t\t\"no meaningful text extracted; the page may require JavaScript or client-side rendering\",\n\t\t\t\t\tundefined,\n\t\t\t\t\tresult.bytesFetched,\n\t\t\t\t);\n\t\t\t}\n\t\t\tconst bounded = truncateToBytes(extracted.text, options.maxTextBytes);\n\t\t\treturn {\n\t\t\t\tfinalUrl: result.finalUrl,\n\t\t\t\t...(extracted.title === undefined ? {} : { title: extracted.title }),\n\t\t\t\tcontentType: result.contentType,\n\t\t\t\ttext: bounded.text,\n\t\t\t\ttruncated: bounded.truncated,\n\t\t\t\tbytesFetched: result.bytesFetched + bytesRead,\n\t\t\t};\n\t\t} catch (error) {\n\t\t\tif (error instanceof PageError && error.bytesRead !== undefined) bytesRead += error.bytesRead;\n\t\t\tlastError = error;\n\t\t\tif (attempt === 0 && isRetryable(error) && options.maxRawBytes > bytesRead) continue;\n\t\t\tif (error instanceof PageError) {\n\t\t\t\tthrow new PageError(error.code, error.message, error.httpStatus, bytesRead);\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t}\n\tthrow lastError instanceof Error ? lastError : new PageError(\"network_error\", String(lastError));\n}\n"]}

package/dist/http.js

@@ -0,0 +1,307 @@
+/**
+ * Bounded HTTP fetch for the search tool (搜索工具设计 §6): http(s) only,
+ * manual redirects, per-page timeout/size caps, no cookies or provider
+ * credentials. Domain/private-network policy is deliberately out of scope;
+ * deployments that need it install a policy plugin before the tool call.
+ */
+import { resolveNetworkRouteV1 } from "@agent-forge/plugin-sdk";
+import { Agent, ProxyAgent, fetch as undiciFetch } from "undici";
+import { extractReadablePage } from "./extract.js";
+import { SEARCH_LIMITS } from "./limits.js";
+import { byteLength, truncateToBytes } from "./text.js";
+export class PageError extends Error {
+    code;
+    httpStatus;
+    /** Raw bytes consumed before the failure, when the body read reached them. */
+    bytesRead;
+    constructor(code, message, httpStatus, bytesRead) {
+        super(message);
+        this.name = "PageError";
+        this.code = code;
+        if (httpStatus !== undefined)
+            this.httpStatus = httpStatus;
+        if (bytesRead !== undefined)
+            this.bytesRead = bytesRead;
+    }
+}
+const ALLOWED_CONTENT_TYPES = new Set([
+    "text/html",
+    "application/xhtml+xml",
+    "text/plain",
+    "application/json",
+]);
+const MIN_MEANINGFUL_TEXT_CHARS = 300;
+const JS_SHELL_RAW_BYTES = 10_000;
+const JS_SHELL_TEXT_RATIO = 0.01;
+const DEFAULT_HEADERS = {
+    "user-agent": "agent-forge-search/1.0",
+    accept: "text/html, application/xhtml+xml, text/plain, application/json;q=0.9",
+    // Keep zstd out of the negotiation: undici advertises it, but Node runtimes
+    // without zlib zstd decoding fail the whole response stream.
+    "accept-encoding": "gzip, deflate, br",
+};
+/**
+ * Lazily-created undici dispatchers; closed by the caller. Direct routes get an
+ * explicit Agent so the host's process-wide env-proxy dispatcher cannot
+ * silently re-route a settings `direct` mode or a bypassed host.
+ */
+export class DispatcherPool {
+    agents = new Map();
+    directAgent;
+    network;
+    enabled;
+    constructor(network, enabled) {
+        this.network = network;
+        this.enabled = enabled;
+    }
+    get(host) {
+        if (!this.enabled || this.network === undefined)
+            return undefined;
+        const resolution = resolveNetworkRouteV1(this.network, host);
+        if (resolution.mode !== "proxy" || resolution.proxy === undefined) {
+            this.directAgent ??= new Agent();
+            return this.directAgent;
+        }
+        const key = `${resolution.proxy.url}\u0000${resolution.proxy.username ?? ""}\u0000${resolution.proxy.password ?? ""}`;
+        const existing = this.agents.get(key);
+        if (existing !== undefined)
+            return existing;
+        const credentials = resolution.proxy.username === undefined && resolution.proxy.password === undefined
+            ? undefined
+            : `${resolution.proxy.username ?? ""}:${resolution.proxy.password ?? ""}`;
+        const agent = new ProxyAgent({
+            uri: resolution.proxy.url,
+            ...(credentials === undefined
+                ? {}
+                : { token: `Basic ${Buffer.from(credentials, "utf8").toString("base64")}` }),
+        });
+        this.agents.set(key, agent);
+        return agent;
+    }
+    async close() {
+        await Promise.all([...this.agents.values()].map((agent) => agent.close()));
+        this.agents.clear();
+        if (this.directAgent !== undefined) {
+            await this.directAgent.close();
+            this.directAgent = undefined;
+        }
+    }
+}
+const defaultFetch = (url, init) => undiciFetch(url, init);
+function parseHttpUrl(raw) {
+    let url;
+    try {
+        url = new URL(raw);
+    }
+    catch {
+        throw new PageError("invalid_url", `unparseable URL: ${JSON.stringify(raw)}`);
+    }
+    if (url.protocol !== "http:" && url.protocol !== "https:") {
+        throw new PageError("invalid_url", `only http(s) URLs are supported: ${JSON.stringify(raw)}`);
+    }
+    return url;
+}
+async function readBodyCapped(response, maxBytes) {
+    if (response.body === undefined || response.body === null) {
+        const text = await response.text();
+        const size = byteLength(text);
+        if (size > maxBytes) {
+            throw new PageError("too_large", `response body exceeds the ${maxBytes} byte limit`, undefined, size);
+        }
+        return text;
+    }
+    const chunks = [];
+    let total = 0;
+    try {
+        for await (const chunk of response.body) {
+            total += chunk.byteLength;
+            if (total > maxBytes) {
+                throw new PageError("too_large", `response body exceeds the ${maxBytes} byte limit`, undefined, total);
+            }
+            chunks.push(chunk);
+        }
+    }
+    catch (error) {
+        if (error instanceof PageError)
+            throw error;
+        throw new PageError("network_error", error instanceof Error ? error.message : String(error), undefined, total);
+    }
+    return Buffer.concat(chunks).toString("utf8");
+}
+function contentTypeOf(response) {
+    const header = response.headers.get("content-type") ?? "";
+    return header.split(";")[0]?.trim().toLowerCase() ?? "";
+}
+async function fetchOnce(options, deadlineAt) {
+    const fetchImpl = options.fetchImpl ?? defaultFetch;
+    const start = parseHttpUrl(options.url);
+    let current = start;
+    for (let hop = 0; hop <= SEARCH_LIMITS.maxRedirects; hop += 1) {
+        const remaining = deadlineAt - Date.now();
+        if (remaining <= 0) {
+            throw new PageError("timeout", `request timed out after ${options.timeoutMs}ms`);
+        }
+        const controller = new AbortController();
+        const timer = setTimeout(() => controller.abort(), remaining);
+        const abortFromCaller = () => controller.abort();
+        options.signal?.addEventListener("abort", abortFromCaller, { once: true });
+        try {
+            const headers = { ...DEFAULT_HEADERS };
+            const dispatcher = options.dispatchers?.get(current.hostname);
+            let response;
+            try {
+                response = await fetchImpl(current.toString(), {
+                    method: "GET",
+                    headers,
+                    redirect: "manual",
+                    signal: controller.signal,
+                    ...(dispatcher === undefined ? {} : { dispatcher }),
+                });
+            }
+            catch (error) {
+                const bytesRead = error instanceof PageError ? error.bytesRead : undefined;
+                if (options.signal?.aborted === true) {
+                    throw new PageError("cancelled", "request cancelled by caller", undefined, bytesRead);
+                }
+                if (controller.signal.aborted) {
+                    throw new PageError("timeout", `request timed out after ${options.timeoutMs}ms`, undefined, bytesRead);
+                }
+                throw new PageError("network_error", `request failed: ${error instanceof Error ? error.message : String(error)}`, undefined, bytesRead);
+            }
+            if (response.status >= 300 && response.status < 400) {
+                const location = response.headers.get("location");
+                if (location === null || location.trim() === "") {
+                    throw new PageError("invalid_redirect", `redirect ${response.status} without Location`);
+                }
+                if (hop === SEARCH_LIMITS.maxRedirects) {
+                    throw new PageError("too_many_redirects", `more than ${SEARCH_LIMITS.maxRedirects} redirects`);
+                }
+                let next;
+                try {
+                    next = new URL(location, current);
+                }
+                catch {
+                    throw new PageError("invalid_redirect", `redirect target is not a URL: ${JSON.stringify(location)}`);
+                }
+                if (next.protocol !== "http:" && next.protocol !== "https:") {
+                    throw new PageError("invalid_redirect", `redirect target must be http(s): ${next.toString()}`);
+                }
+                current = next;
+                continue;
+            }
+            if (response.status < 200 || response.status >= 300) {
+                throw new PageError("http_error", `HTTP ${response.status}`, response.status);
+            }
+            const contentType = contentTypeOf(response);
+            if (!ALLOWED_CONTENT_TYPES.has(contentType)) {
+                throw new PageError("unsupported_content_type", `content type ${contentType === "" ? "(missing)" : contentType} is not supported`);
+            }
+            const declaredLength = response.headers.get("content-length");
+            if (declaredLength !== null &&
+                Number.isInteger(Number(declaredLength)) &&
+                Number(declaredLength) > options.maxRawBytes) {
+                throw new PageError("too_large", `response declares ${declaredLength} bytes`);
+            }
+            let bodyText;
+            try {
+                bodyText = await readBodyCapped(response, options.maxRawBytes);
+            }
+            catch (error) {
+                const bytesRead = error instanceof PageError ? error.bytesRead : undefined;
+                if (options.signal?.aborted === true) {
+                    throw new PageError("cancelled", "request cancelled by caller", undefined, bytesRead);
+                }
+                if (controller.signal.aborted) {
+                    throw new PageError("timeout", `request timed out after ${options.timeoutMs}ms`, undefined, bytesRead);
+                }
+                throw error;
+            }
+            return {
+                finalUrl: current.toString(),
+                contentType,
+                bodyText,
+                bytesFetched: byteLength(bodyText),
+            };
+        }
+        finally {
+            clearTimeout(timer);
+            options.signal?.removeEventListener("abort", abortFromCaller);
+        }
+    }
+    throw new PageError("too_many_redirects", `more than ${SEARCH_LIMITS.maxRedirects} redirects`);
+}
+function extractPage(contentType, bodyText) {
+    if (contentType === "text/html" || contentType === "application/xhtml+xml") {
+        const page = extractReadablePage(bodyText);
+        if (page === undefined || page.text.trim() === "") {
+            throw new PageError("empty_content", "no readable text extracted; the page may require JavaScript");
+        }
+        return { ...(page.title === undefined ? {} : { title: page.title }), text: page.text };
+    }
+    const text = bodyText.trim();
+    if (text === "")
+        throw new PageError("empty_content", "empty response body");
+    return { text };
+}
+function isRetryable(error) {
+    if (!(error instanceof PageError))
+        return false;
+    if (error.code === "network_error")
+        return true;
+    return (error.code === "http_error" &&
+        error.httpStatus !== undefined &&
+        (error.httpStatus === 429 || error.httpStatus >= 500));
+}
+/** Fetch and extract one page; bounded and explicit about every failure. */
+export async function fetchPage(options) {
+    if (options.signal?.aborted === true) {
+        throw new PageError("cancelled", "request cancelled by caller");
+    }
+    const deadlineAt = Date.now() + options.timeoutMs;
+    let bytesRead = 0;
+    let lastError;
+    for (let attempt = 0; attempt < 2; attempt += 1) {
+        try {
+            const remainingRawBytes = Math.max(0, options.maxRawBytes - bytesRead);
+            const result = await fetchOnce({ ...options, maxRawBytes: remainingRawBytes }, deadlineAt);
+            let extracted;
+            try {
+                extracted = extractPage(result.contentType, result.bodyText);
+            }
+            catch (error) {
+                if (error instanceof PageError) {
+                    throw new PageError(error.code, error.message, error.httpStatus, result.bytesFetched);
+                }
+                throw error;
+            }
+            const meaningfulChars = extracted.text.trim().length;
+            if (meaningfulChars < MIN_MEANINGFUL_TEXT_CHARS &&
+                result.bytesFetched > JS_SHELL_RAW_BYTES &&
+                meaningfulChars / result.bytesFetched < JS_SHELL_TEXT_RATIO) {
+                throw new PageError("empty_content", "no meaningful text extracted; the page may require JavaScript or client-side rendering", undefined, result.bytesFetched);
+            }
+            const bounded = truncateToBytes(extracted.text, options.maxTextBytes);
+            return {
+                finalUrl: result.finalUrl,
+                ...(extracted.title === undefined ? {} : { title: extracted.title }),
+                contentType: result.contentType,
+                text: bounded.text,
+                truncated: bounded.truncated,
+                bytesFetched: result.bytesFetched + bytesRead,
+            };
+        }
+        catch (error) {
+            if (error instanceof PageError && error.bytesRead !== undefined)
+                bytesRead += error.bytesRead;
+            lastError = error;
+            if (attempt === 0 && isRetryable(error) && options.maxRawBytes > bytesRead)
+                continue;
+            if (error instanceof PageError) {
+                throw new PageError(error.code, error.message, error.httpStatus, bytesRead);
+            }
+            throw error;
+        }
+    }
+    throw lastError instanceof Error ? lastError : new PageError("network_error", String(lastError));
+}
+//# sourceMappingURL=http.js.map

package/dist/http.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"http.js","sourceRoot":"","sources":["../src/http.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,EAAE,qBAAqB,EAAE,MAAM,yBAAyB,CAAC;AAChE,OAAO,EAAE,KAAK,EAAE,UAAU,EAAE,KAAK,IAAI,WAAW,EAAE,MAAM,QAAQ,CAAC;AACjE,OAAO,EAAE,mBAAmB,EAAE,MAAM,cAAc,CAAC;AACnD,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAC5C,OAAO,EAAE,UAAU,EAAE,eAAe,EAAE,MAAM,WAAW,CAAC;AAgCxD,MAAM,OAAO,SAAU,SAAQ,KAAK;IAC1B,IAAI,CAAkB;IACtB,UAAU,CAAU;IAC7B,8EAA8E;IACrE,SAAS,CAAU;IAE5B,YAAY,IAAqB,EAAE,OAAe,EAAE,UAAmB,EAAE,SAAkB,EAAE;QAC5F,KAAK,CAAC,OAAO,CAAC,CAAC;QACf,IAAI,CAAC,IAAI,GAAG,WAAW,CAAC;QACxB,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;QACjB,IAAI,UAAU,KAAK,SAAS;YAAE,IAAI,CAAC,UAAU,GAAG,UAAU,CAAC;QAC3D,IAAI,SAAS,KAAK,SAAS;YAAE,IAAI,CAAC,SAAS,GAAG,SAAS,CAAC;IAAA,CACxD;CACD;AAsBD,MAAM,qBAAqB,GAAwB,IAAI,GAAG,CAAC;IAC1D,WAAW;IACX,uBAAuB;IACvB,YAAY;IACZ,kBAAkB;CAClB,CAAC,CAAC;AAEH,MAAM,yBAAyB,GAAG,GAAG,CAAC;AACtC,MAAM,kBAAkB,GAAG,MAAM,CAAC;AAClC,MAAM,mBAAmB,GAAG,IAAI,CAAC;AAEjC,MAAM,eAAe,GAAqC;IACzD,YAAY,EAAE,wBAAwB;IACtC,MAAM,EAAE,sEAAsE;IAC9E,4EAA4E;IAC5E,6DAA6D;IAC7D,iBAAiB,EAAE,mBAAmB;CACtC,CAAC;AAEF;;;;GAIG;AACH,MAAM,OAAO,cAAc;IACT,MAAM,GAAG,IAAI,GAAG,EAAsB,CAAC;IAChD,WAAW,CAAoB;IACtB,OAAO,CAAmC;IAC1C,OAAO,CAAU;IAElC,YAAY,OAAyC,EAAE,OAAgB,EAAE;QACxE,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;QACvB,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;IAAA,CACvB;IAED,GAAG,CAAC,IAAY,EAAuB;QACtC,IAAI,CAAC,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC,OAAO,KAAK,SAAS;YAAE,OAAO,SAAS,CAAC;QAClE,MAAM,UAAU,GAAG,qBAAqB,CAAC,IAAI,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC;QAC7D,IAAI,UAAU,CAAC,IAAI,KAAK,OAAO,IAAI,UAAU,CAAC,KAAK,KAAK,SAAS,EAAE,CAAC;YACnE,IAAI,CAAC,WAAW,KAAK,IAAI,KAAK,EAAE,CAAC;YACjC,OAAO,IAAI,CAAC,WAAW,CAAC;QACzB,CAAC;QACD,MAAM,GAAG,GAAG,GAAG,UAAU,CAAC,KAAK,CAAC,GAAG,SAAS,UAAU,CAAC,KAAK,CAAC,QAAQ,IAAI,EAAE,SAAS,UAAU,CAAC,KAAK,CAAC,QAAQ,IAAI,EAAE,EAAE,CAAC;QACtH,MAAM,QAAQ,GAAG,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACtC,IAAI,QAAQ,KAAK,SAAS;YAAE,OAAO,QAAQ,CAAC;QAC5C,MAAM,WAAW,GAChB,UAAU,CAAC,KAAK,CAAC,QAAQ,KAAK,SAAS,IAAI,UAAU,CAAC,KAAK,CAAC,QAAQ,KAAK,SAAS;YACjF,CAAC,CAAC,SAAS;YACX,CAAC,CAAC,GAAG,UAAU,CAAC,KAAK,CAAC,QAAQ,IAAI,EAAE,IAAI,UAAU,CAAC,KAAK,CAAC,QAAQ,IAAI,EAAE,EAAE,CAAC;QAC5E,MAAM,KAAK,GAAG,IAAI,UAAU,CAAC;YAC5B,GAAG,EAAE,UAAU,CAAC,KAAK,CAAC,GAAG;YACzB,GAAG,CAAC,WAAW,KAAK,SAAS;gBAC5B,CAAC,CAAC,EAAE;gBACJ,CAAC,CAAC,EAAE,KAAK,EAAE,SAAS,MAAM,CAAC,IAAI,CAAC,WAAW,EAAE,MAAM,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;SAC7E,CAAC,CAAC;QACH,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;QAC5B,OAAO,KAAK,CAAC;IAAA,CACb;IAED,KAAK,CAAC,KAAK,GAAkB;QAC5B,MAAM,OAAO,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;QAC3E,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC;QACpB,IAAI,IAAI,CAAC,WAAW,KAAK,SAAS,EAAE,CAAC;YACpC,MAAM,IAAI,CAAC,WAAW,CAAC,KAAK,EAAE,CAAC;YAC/B,IAAI,CAAC,WAAW,GAAG,SAAS,CAAC;QAC9B,CAAC;IAAA,CACD;CACD;AAED,MAAM,YAAY,GAAc,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CAC7C,WAAW,CAAC,GAAG,EAAE,IAAyC,CAA0C,CAAC;AAEtG,SAAS,YAAY,CAAC,GAAW,EAAO;IACvC,IAAI,GAAQ,CAAC;IACb,IAAI,CAAC;QACJ,GAAG,GAAG,IAAI,GAAG,CAAC,GAAG,CAAC,CAAC;IACpB,CAAC;IAAC,MAAM,CAAC;QACR,MAAM,IAAI,SAAS,CAAC,aAAa,EAAE,oBAAoB,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;IAC/E,CAAC;IACD,IAAI,GAAG,CAAC,QAAQ,KAAK,OAAO,IAAI,GAAG,CAAC,QAAQ,KAAK,QAAQ,EAAE,CAAC;QAC3D,MAAM,IAAI,SAAS,CAAC,aAAa,EAAE,oCAAoC,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;IAC/F,CAAC;IACD,OAAO,GAAG,CAAC;AAAA,CACX;AAED,KAAK,UAAU,cAAc,CAAC,QAA2B,EAAE,QAAgB,EAAmB;IAC7F,IAAI,QAAQ,CAAC,IAAI,KAAK,SAAS,IAAI,QAAQ,CAAC,IAAI,KAAK,IAAI,EAAE,CAAC;QAC3D,MAAM,IAAI,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;QACnC,MAAM,IAAI,GAAG,UAAU,CAAC,IAAI,CAAC,CAAC;QAC9B,IAAI,IAAI,GAAG,QAAQ,EAAE,CAAC;YACrB,MAAM,IAAI,SAAS,CAAC,WAAW,EAAE,6BAA6B,QAAQ,aAAa,EAAE,SAAS,EAAE,IAAI,CAAC,CAAC;QACvG,CAAC;QACD,OAAO,IAAI,CAAC;IACb,CAAC;IACD,MAAM,MAAM,GAAiB,EAAE,CAAC;IAChC,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,IAAI,CAAC;QACJ,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,QAAQ,CAAC,IAAI,EAAE,CAAC;YACzC,KAAK,IAAI,KAAK,CAAC,UAAU,CAAC;YAC1B,IAAI,KAAK,GAAG,QAAQ,EAAE,CAAC;gBACtB,MAAM,IAAI,SAAS,CAAC,WAAW,EAAE,6BAA6B,QAAQ,aAAa,EAAE,SAAS,EAAE,KAAK,CAAC,CAAC;YACxG,CAAC;YACD,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACpB,CAAC;IACF,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QAChB,IAAI,KAAK,YAAY,SAAS;YAAE,MAAM,KAAK,CAAC;QAC5C,MAAM,IAAI,SAAS,CAAC,eAAe,EAAE,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,SAAS,EAAE,KAAK,CAAC,CAAC;IAChH,CAAC;IACD,OAAO,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;AAAA,CAC9C;AAED,SAAS,aAAa,CAAC,QAA2B,EAAU;IAC3D,MAAM,MAAM,GAAG,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,IAAI,EAAE,CAAC;IAC1D,OAAO,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,WAAW,EAAE,IAAI,EAAE,CAAC;AAAA,CACxD;AASD,KAAK,UAAU,SAAS,CAAC,OAA2B,EAAE,UAAkB,EAA8B;IACrG,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,IAAI,YAAY,CAAC;IACpD,MAAM,KAAK,GAAG,YAAY,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IACxC,IAAI,OAAO,GAAG,KAAK,CAAC;IACpB,KAAK,IAAI,GAAG,GAAG,CAAC,EAAE,GAAG,IAAI,aAAa,CAAC,YAAY,EAAE,GAAG,IAAI,CAAC,EAAE,CAAC;QAC/D,MAAM,SAAS,GAAG,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC1C,IAAI,SAAS,IAAI,CAAC,EAAE,CAAC;YACpB,MAAM,IAAI,SAAS,CAAC,SAAS,EAAE,2BAA2B,OAAO,CAAC,SAAS,IAAI,CAAC,CAAC;QAClF,CAAC;QACD,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;QACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,SAAS,CAAC,CAAC;QAC9D,MAAM,eAAe,GAAG,GAAS,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC;QACvD,OAAO,CAAC,MAAM,EAAE,gBAAgB,CAAC,OAAO,EAAE,eAAe,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;QAC3E,IAAI,CAAC;YACJ,MAAM,OAAO,GAA2B,EAAE,GAAG,eAAe,EAAE,CAAC;YAC/D,MAAM,UAAU,GAAG,OAAO,CAAC,WAAW,EAAE,GAAG,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;YAC9D,IAAI,QAA2B,CAAC;YAChC,IAAI,CAAC;gBACJ,QAAQ,GAAG,MAAM,SAAS,CAAC,OAAO,CAAC,QAAQ,EAAE,EAAE;oBAC9C,MAAM,EAAE,KAAK;oBACb,OAAO;oBACP,QAAQ,EAAE,QAAQ;oBAClB,MAAM,EAAE,UAAU,CAAC,MAAM;oBACzB,GAAG,CAAC,UAAU,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC;iBACnD,CAAC,CAAC;YACJ,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBAChB,MAAM,SAAS,GAAG,KAAK,YAAY,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC;gBAC3E,IAAI,OAAO,CAAC,MAAM,EAAE,OAAO,KAAK,IAAI,EAAE,CAAC;oBACtC,MAAM,IAAI,SAAS,CAAC,WAAW,EAAE,6BAA6B,EAAE,SAAS,EAAE,SAAS,CAAC,CAAC;gBACvF,CAAC;gBACD,IAAI,UAAU,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;oBAC/B,MAAM,IAAI,SAAS,CAAC,SAAS,EAAE,2BAA2B,OAAO,CAAC,SAAS,IAAI,EAAE,SAAS,EAAE,SAAS,CAAC,CAAC;gBACxG,CAAC;gBACD,MAAM,IAAI,SAAS,CAClB,eAAe,EACf,mBAAmB,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,EAC3E,SAAS,EACT,SAAS,CACT,CAAC;YACH,CAAC;YAED,IAAI,QAAQ,CAAC,MAAM,IAAI,GAAG,IAAI,QAAQ,CAAC,MAAM,GAAG,GAAG,EAAE,CAAC;gBACrD,MAAM,QAAQ,GAAG,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;gBAClD,IAAI,QAAQ,KAAK,IAAI,IAAI,QAAQ,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;oBACjD,MAAM,IAAI,SAAS,CAAC,kBAAkB,EAAE,YAAY,QAAQ,CAAC,MAAM,mBAAmB,CAAC,CAAC;gBACzF,CAAC;gBACD,IAAI,GAAG,KAAK,aAAa,CAAC,YAAY,EAAE,CAAC;oBACxC,MAAM,IAAI,SAAS,CAAC,oBAAoB,EAAE,aAAa,aAAa,CAAC,YAAY,YAAY,CAAC,CAAC;gBAChG,CAAC;gBACD,IAAI,IAAS,CAAC;gBACd,IAAI,CAAC;oBACJ,IAAI,GAAG,IAAI,GAAG,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;gBACnC,CAAC;gBAAC,MAAM,CAAC;oBACR,MAAM,IAAI,SAAS,CAAC,kBAAkB,EAAE,iCAAiC,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC;gBACtG,CAAC;gBACD,IAAI,IAAI,CAAC,QAAQ,KAAK,OAAO,IAAI,IAAI,CAAC,QAAQ,KAAK,QAAQ,EAAE,CAAC;oBAC7D,MAAM,IAAI,SAAS,CAAC,kBAAkB,EAAE,oCAAoC,IAAI,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;gBAChG,CAAC;gBACD,OAAO,GAAG,IAAI,CAAC;gBACf,SAAS;YACV,CAAC;YAED,IAAI,QAAQ,CAAC,MAAM,GAAG,GAAG,IAAI,QAAQ,CAAC,MAAM,IAAI,GAAG,EAAE,CAAC;gBACrD,MAAM,IAAI,SAAS,CAAC,YAAY,EAAE,QAAQ,QAAQ,CAAC,MAAM,EAAE,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;YAC/E,CAAC;YAED,MAAM,WAAW,GAAG,aAAa,CAAC,QAAQ,CAAC,CAAC;YAC5C,IAAI,CAAC,qBAAqB,CAAC,GAAG,CAAC,WAAW,CAAC,EAAE,CAAC;gBAC7C,MAAM,IAAI,SAAS,CAClB,0BAA0B,EAC1B,gBAAgB,WAAW,KAAK,EAAE,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,WAAW,mBAAmB,CACjF,CAAC;YACH,CAAC;YACD,MAAM,cAAc,GAAG,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC;YAC9D,IACC,cAAc,KAAK,IAAI;gBACvB,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,cAAc,CAAC,CAAC;gBACxC,MAAM,CAAC,cAAc,CAAC,GAAG,OAAO,CAAC,WAAW,EAC3C,CAAC;gBACF,MAAM,IAAI,SAAS,CAAC,WAAW,EAAE,qBAAqB,cAAc,QAAQ,CAAC,CAAC;YAC/E,CAAC;YACD,IAAI,QAAgB,CAAC;YACrB,IAAI,CAAC;gBACJ,QAAQ,GAAG,MAAM,cAAc,CAAC,QAAQ,EAAE,OAAO,CAAC,WAAW,CAAC,CAAC;YAChE,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBAChB,MAAM,SAAS,GAAG,KAAK,YAAY,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC;gBAC3E,IAAI,OAAO,CAAC,MAAM,EAAE,OAAO,KAAK,IAAI,EAAE,CAAC;oBACtC,MAAM,IAAI,SAAS,CAAC,WAAW,EAAE,6BAA6B,EAAE,SAAS,EAAE,SAAS,CAAC,CAAC;gBACvF,CAAC;gBACD,IAAI,UAAU,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;oBAC/B,MAAM,IAAI,SAAS,CAAC,SAAS,EAAE,2BAA2B,OAAO,CAAC,SAAS,IAAI,EAAE,SAAS,EAAE,SAAS,CAAC,CAAC;gBACxG,CAAC;gBACD,MAAM,KAAK,CAAC;YACb,CAAC;YACD,OAAO;gBACN,QAAQ,EAAE,OAAO,CAAC,QAAQ,EAAE;gBAC5B,WAAW;gBACX,QAAQ;gBACR,YAAY,EAAE,UAAU,CAAC,QAAQ,CAAC;aAClC,CAAC;QACH,CAAC;gBAAS,CAAC;YACV,YAAY,CAAC,KAAK,CAAC,CAAC;YACpB,OAAO,CAAC,MAAM,EAAE,mBAAmB,CAAC,OAAO,EAAE,eAAe,CAAC,CAAC;QAC/D,CAAC;IACF,CAAC;IACD,MAAM,IAAI,SAAS,CAAC,oBAAoB,EAAE,aAAa,aAAa,CAAC,YAAY,YAAY,CAAC,CAAC;AAAA,CAC/F;AAED,SAAS,WAAW,CAAC,WAAmB,EAAE,QAAgB,EAAoC;IAC7F,IAAI,WAAW,KAAK,WAAW,IAAI,WAAW,KAAK,uBAAuB,EAAE,CAAC;QAC5E,MAAM,IAAI,GAAG,mBAAmB,CAAC,QAAQ,CAAC,CAAC;QAC3C,IAAI,IAAI,KAAK,SAAS,IAAI,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;YACnD,MAAM,IAAI,SAAS,CAAC,eAAe,EAAE,6DAA6D,CAAC,CAAC;QACrG,CAAC;QACD,OAAO,EAAE,GAAG,CAAC,IAAI,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,CAAC;IACxF,CAAC;IACD,MAAM,IAAI,GAAG,QAAQ,CAAC,IAAI,EAAE,CAAC;IAC7B,IAAI,IAAI,KAAK,EAAE;QAAE,MAAM,IAAI,SAAS,CAAC,eAAe,EAAE,qBAAqB,CAAC,CAAC;IAC7E,OAAO,EAAE,IAAI,EAAE,CAAC;AAAA,CAChB;AAED,SAAS,WAAW,CAAC,KAAc,EAAW;IAC7C,IAAI,CAAC,CAAC,KAAK,YAAY,SAAS,CAAC;QAAE,OAAO,KAAK,CAAC;IAChD,IAAI,KAAK,CAAC,IAAI,KAAK,eAAe;QAAE,OAAO,IAAI,CAAC;IAChD,OAAO,CACN,KAAK,CAAC,IAAI,KAAK,YAAY;QAC3B,KAAK,CAAC,UAAU,KAAK,SAAS;QAC9B,CAAC,KAAK,CAAC,UAAU,KAAK,GAAG,IAAI,KAAK,CAAC,UAAU,IAAI,GAAG,CAAC,CACrD,CAAC;AAAA,CACF;AAED,4EAA4E;AAC5E,MAAM,CAAC,KAAK,UAAU,SAAS,CAAC,OAA2B,EAA0B;IACpF,IAAI,OAAO,CAAC,MAAM,EAAE,OAAO,KAAK,IAAI,EAAE,CAAC;QACtC,MAAM,IAAI,SAAS,CAAC,WAAW,EAAE,6BAA6B,CAAC,CAAC;IACjE,CAAC;IACD,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,OAAO,CAAC,SAAS,CAAC;IAClD,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,SAAkB,CAAC;IACvB,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,CAAC,EAAE,CAAC;QACjD,IAAI,CAAC;YACJ,MAAM,iBAAiB,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,WAAW,GAAG,SAAS,CAAC,CAAC;YACvE,MAAM,MAAM,GAAG,MAAM,SAAS,CAAC,EAAE,GAAG,OAAO,EAAE,WAAW,EAAE,iBAAiB,EAAE,EAAE,UAAU,CAAC,CAAC;YAC3F,IAAI,SAA2C,CAAC;YAChD,IAAI,CAAC;gBACJ,SAAS,GAAG,WAAW,CAAC,MAAM,CAAC,WAAW,EAAE,MAAM,CAAC,QAAQ,CAAC,CAAC;YAC9D,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBAChB,IAAI,KAAK,YAAY,SAAS,EAAE,CAAC;oBAChC,MAAM,IAAI,SAAS,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,UAAU,EAAE,MAAM,CAAC,YAAY,CAAC,CAAC;gBACvF,CAAC;gBACD,MAAM,KAAK,CAAC;YACb,CAAC;YACD,MAAM,eAAe,GAAG,SAAS,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,MAAM,CAAC;YACrD,IACC,eAAe,GAAG,yBAAyB;gBAC3C,MAAM,CAAC,YAAY,GAAG,kBAAkB;gBACxC,eAAe,GAAG,MAAM,CAAC,YAAY,GAAG,mBAAmB,EAC1D,CAAC;gBACF,MAAM,IAAI,SAAS,CAClB,eAAe,EACf,wFAAwF,EACxF,SAAS,EACT,MAAM,CAAC,YAAY,CACnB,CAAC;YACH,CAAC;YACD,MAAM,OAAO,GAAG,eAAe,CAAC,SAAS,CAAC,IAAI,EAAE,OAAO,CAAC,YAAY,CAAC,CAAC;YACtE,OAAO;gBACN,QAAQ,EAAE,MAAM,CAAC,QAAQ;gBACzB,GAAG,CAAC,SAAS,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,SAAS,CAAC,KAAK,EAAE,CAAC;gBACpE,WAAW,EAAE,MAAM,CAAC,WAAW;gBAC/B,IAAI,EAAE,OAAO,CAAC,IAAI;gBAClB,SAAS,EAAE,OAAO,CAAC,SAAS;gBAC5B,YAAY,EAAE,MAAM,CAAC,YAAY,GAAG,SAAS;aAC7C,CAAC;QACH,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,IAAI,KAAK,YAAY,SAAS,IAAI,KAAK,CAAC,SAAS,KAAK,SAAS;gBAAE,SAAS,IAAI,KAAK,CAAC,SAAS,CAAC;YAC9F,SAAS,GAAG,KAAK,CAAC;YAClB,IAAI,OAAO,KAAK,CAAC,IAAI,WAAW,CAAC,KAAK,CAAC,IAAI,OAAO,CAAC,WAAW,GAAG,SAAS;gBAAE,SAAS;YACrF,IAAI,KAAK,YAAY,SAAS,EAAE,CAAC;gBAChC,MAAM,IAAI,SAAS,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,UAAU,EAAE,SAAS,CAAC,CAAC;YAC7E,CAAC;YACD,MAAM,KAAK,CAAC;QACb,CAAC;IACF,CAAC;IACD,MAAM,SAAS,YAAY,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,SAAS,CAAC,eAAe,EAAE,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC;AAAA,CACjG","sourcesContent":["/**\n * Bounded HTTP fetch for the search tool (搜索工具设计 §6): http(s) only,\n * manual redirects, per-page timeout/size caps, no cookies or provider\n * credentials. Domain/private-network policy is deliberately out of scope;\n * deployments that need it install a policy plugin before the tool call.\n */\n\nimport type { NetworkRouteConfigV1 } from \"@agent-forge/plugin-sdk\";\nimport { resolveNetworkRouteV1 } from \"@agent-forge/plugin-sdk\";\nimport { Agent, ProxyAgent, fetch as undiciFetch } from \"undici\";\nimport { extractReadablePage } from \"./extract.ts\";\nimport { SEARCH_LIMITS } from \"./limits.ts\";\nimport { byteLength, truncateToBytes } from \"./text.ts\";\n\nexport interface FetchResponseLike {\n\treadonly status: number;\n\treadonly headers: { get(name: string): string | null };\n\treadonly body?: AsyncIterable<Uint8Array> | null;\n\ttext(): Promise<string>;\n}\n\nexport type FetchImpl = (\n\turl: string,\n\tinit: {\n\t\tmethod?: string;\n\t\theaders?: Record<string, string>;\n\t\tredirect: \"manual\";\n\t\tsignal: AbortSignal;\n\t\tdispatcher?: unknown;\n\t},\n) => Promise<FetchResponseLike>;\n\nexport type PageErrorCodeV1 =\n\t| \"invalid_url\"\n\t| \"invalid_redirect\"\n\t| \"too_many_redirects\"\n\t| \"http_error\"\n\t| \"timeout\"\n\t| \"network_error\"\n\t| \"too_large\"\n\t| \"unsupported_content_type\"\n\t| \"empty_content\"\n\t| \"cancelled\";\n\nexport class PageError extends Error {\n\treadonly code: PageErrorCodeV1;\n\treadonly httpStatus?: number;\n\t/** Raw bytes consumed before the failure, when the body read reached them. */\n\treadonly bytesRead?: number;\n\n\tconstructor(code: PageErrorCodeV1, message: string, httpStatus?: number, bytesRead?: number) {\n\t\tsuper(message);\n\t\tthis.name = \"PageError\";\n\t\tthis.code = code;\n\t\tif (httpStatus !== undefined) this.httpStatus = httpStatus;\n\t\tif (bytesRead !== undefined) this.bytesRead = bytesRead;\n\t}\n}\n\nexport interface FetchedPageV1 {\n\treadonly finalUrl: string;\n\treadonly title?: string;\n\treadonly contentType: string;\n\treadonly text: string;\n\treadonly truncated: boolean;\n\treadonly bytesFetched: number;\n}\n\nexport interface FetchPageOptionsV1 {\n\treadonly url: string;\n\treadonly maxTextBytes: number;\n\treadonly maxRawBytes: number;\n\treadonly timeoutMs: number;\n\treadonly signal?: AbortSignal;\n\treadonly network?: NetworkRouteConfigV1;\n\treadonly fetchImpl?: FetchImpl;\n\treadonly dispatchers?: DispatcherPool;\n}\n\nconst ALLOWED_CONTENT_TYPES: ReadonlySet<string> = new Set([\n\t\"text/html\",\n\t\"application/xhtml+xml\",\n\t\"text/plain\",\n\t\"application/json\",\n]);\n\nconst MIN_MEANINGFUL_TEXT_CHARS = 300;\nconst JS_SHELL_RAW_BYTES = 10_000;\nconst JS_SHELL_TEXT_RATIO = 0.01;\n\nconst DEFAULT_HEADERS: Readonly<Record<string, string>> = {\n\t\"user-agent\": \"agent-forge-search/1.0\",\n\taccept: \"text/html, application/xhtml+xml, text/plain, application/json;q=0.9\",\n\t// Keep zstd out of the negotiation: undici advertises it, but Node runtimes\n\t// without zlib zstd decoding fail the whole response stream.\n\t\"accept-encoding\": \"gzip, deflate, br\",\n};\n\n/**\n * Lazily-created undici dispatchers; closed by the caller. Direct routes get an\n * explicit Agent so the host's process-wide env-proxy dispatcher cannot\n * silently re-route a settings `direct` mode or a bypassed host.\n */\nexport class DispatcherPool {\n\tprivate readonly agents = new Map<string, ProxyAgent>();\n\tprivate directAgent: Agent | undefined;\n\tprivate readonly network: NetworkRouteConfigV1 | undefined;\n\tprivate readonly enabled: boolean;\n\n\tconstructor(network: NetworkRouteConfigV1 | undefined, enabled: boolean) {\n\t\tthis.network = network;\n\t\tthis.enabled = enabled;\n\t}\n\n\tget(host: string): unknown | undefined {\n\t\tif (!this.enabled || this.network === undefined) return undefined;\n\t\tconst resolution = resolveNetworkRouteV1(this.network, host);\n\t\tif (resolution.mode !== \"proxy\" || resolution.proxy === undefined) {\n\t\t\tthis.directAgent ??= new Agent();\n\t\t\treturn this.directAgent;\n\t\t}\n\t\tconst key = `${resolution.proxy.url}\\u0000${resolution.proxy.username ?? \"\"}\\u0000${resolution.proxy.password ?? \"\"}`;\n\t\tconst existing = this.agents.get(key);\n\t\tif (existing !== undefined) return existing;\n\t\tconst credentials =\n\t\t\tresolution.proxy.username === undefined && resolution.proxy.password === undefined\n\t\t\t\t? undefined\n\t\t\t\t: `${resolution.proxy.username ?? \"\"}:${resolution.proxy.password ?? \"\"}`;\n\t\tconst agent = new ProxyAgent({\n\t\t\turi: resolution.proxy.url,\n\t\t\t...(credentials === undefined\n\t\t\t\t? {}\n\t\t\t\t: { token: `Basic ${Buffer.from(credentials, \"utf8\").toString(\"base64\")}` }),\n\t\t});\n\t\tthis.agents.set(key, agent);\n\t\treturn agent;\n\t}\n\n\tasync close(): Promise<void> {\n\t\tawait Promise.all([...this.agents.values()].map((agent) => agent.close()));\n\t\tthis.agents.clear();\n\t\tif (this.directAgent !== undefined) {\n\t\t\tawait this.directAgent.close();\n\t\t\tthis.directAgent = undefined;\n\t\t}\n\t}\n}\n\nconst defaultFetch: FetchImpl = (url, init) =>\n\tundiciFetch(url, init as Parameters<typeof undiciFetch>[1]) as unknown as Promise<FetchResponseLike>;\n\nfunction parseHttpUrl(raw: string): URL {\n\tlet url: URL;\n\ttry {\n\t\turl = new URL(raw);\n\t} catch {\n\t\tthrow new PageError(\"invalid_url\", `unparseable URL: ${JSON.stringify(raw)}`);\n\t}\n\tif (url.protocol !== \"http:\" && url.protocol !== \"https:\") {\n\t\tthrow new PageError(\"invalid_url\", `only http(s) URLs are supported: ${JSON.stringify(raw)}`);\n\t}\n\treturn url;\n}\n\nasync function readBodyCapped(response: FetchResponseLike, maxBytes: number): Promise<string> {\n\tif (response.body === undefined || response.body === null) {\n\t\tconst text = await response.text();\n\t\tconst size = byteLength(text);\n\t\tif (size > maxBytes) {\n\t\t\tthrow new PageError(\"too_large\", `response body exceeds the ${maxBytes} byte limit`, undefined, size);\n\t\t}\n\t\treturn text;\n\t}\n\tconst chunks: Uint8Array[] = [];\n\tlet total = 0;\n\ttry {\n\t\tfor await (const chunk of response.body) {\n\t\t\ttotal += chunk.byteLength;\n\t\t\tif (total > maxBytes) {\n\t\t\t\tthrow new PageError(\"too_large\", `response body exceeds the ${maxBytes} byte limit`, undefined, total);\n\t\t\t}\n\t\t\tchunks.push(chunk);\n\t\t}\n\t} catch (error) {\n\t\tif (error instanceof PageError) throw error;\n\t\tthrow new PageError(\"network_error\", error instanceof Error ? error.message : String(error), undefined, total);\n\t}\n\treturn Buffer.concat(chunks).toString(\"utf8\");\n}\n\nfunction contentTypeOf(response: FetchResponseLike): string {\n\tconst header = response.headers.get(\"content-type\") ?? \"\";\n\treturn header.split(\";\")[0]?.trim().toLowerCase() ?? \"\";\n}\n\ninterface FetchOnceResultV1 {\n\treadonly finalUrl: string;\n\treadonly contentType: string;\n\treadonly bodyText: string;\n\treadonly bytesFetched: number;\n}\n\nasync function fetchOnce(options: FetchPageOptionsV1, deadlineAt: number): Promise<FetchOnceResultV1> {\n\tconst fetchImpl = options.fetchImpl ?? defaultFetch;\n\tconst start = parseHttpUrl(options.url);\n\tlet current = start;\n\tfor (let hop = 0; hop <= SEARCH_LIMITS.maxRedirects; hop += 1) {\n\t\tconst remaining = deadlineAt - Date.now();\n\t\tif (remaining <= 0) {\n\t\t\tthrow new PageError(\"timeout\", `request timed out after ${options.timeoutMs}ms`);\n\t\t}\n\t\tconst controller = new AbortController();\n\t\tconst timer = setTimeout(() => controller.abort(), remaining);\n\t\tconst abortFromCaller = (): void => controller.abort();\n\t\toptions.signal?.addEventListener(\"abort\", abortFromCaller, { once: true });\n\t\ttry {\n\t\t\tconst headers: Record<string, string> = { ...DEFAULT_HEADERS };\n\t\t\tconst dispatcher = options.dispatchers?.get(current.hostname);\n\t\t\tlet response: FetchResponseLike;\n\t\t\ttry {\n\t\t\t\tresponse = await fetchImpl(current.toString(), {\n\t\t\t\t\tmethod: \"GET\",\n\t\t\t\t\theaders,\n\t\t\t\t\tredirect: \"manual\",\n\t\t\t\t\tsignal: controller.signal,\n\t\t\t\t\t...(dispatcher === undefined ? {} : { dispatcher }),\n\t\t\t\t});\n\t\t\t} catch (error) {\n\t\t\t\tconst bytesRead = error instanceof PageError ? error.bytesRead : undefined;\n\t\t\t\tif (options.signal?.aborted === true) {\n\t\t\t\t\tthrow new PageError(\"cancelled\", \"request cancelled by caller\", undefined, bytesRead);\n\t\t\t\t}\n\t\t\t\tif (controller.signal.aborted) {\n\t\t\t\t\tthrow new PageError(\"timeout\", `request timed out after ${options.timeoutMs}ms`, undefined, bytesRead);\n\t\t\t\t}\n\t\t\t\tthrow new PageError(\n\t\t\t\t\t\"network_error\",\n\t\t\t\t\t`request failed: ${error instanceof Error ? error.message : String(error)}`,\n\t\t\t\t\tundefined,\n\t\t\t\t\tbytesRead,\n\t\t\t\t);\n\t\t\t}\n\n\t\t\tif (response.status >= 300 && response.status < 400) {\n\t\t\t\tconst location = response.headers.get(\"location\");\n\t\t\t\tif (location === null || location.trim() === \"\") {\n\t\t\t\t\tthrow new PageError(\"invalid_redirect\", `redirect ${response.status} without Location`);\n\t\t\t\t}\n\t\t\t\tif (hop === SEARCH_LIMITS.maxRedirects) {\n\t\t\t\t\tthrow new PageError(\"too_many_redirects\", `more than ${SEARCH_LIMITS.maxRedirects} redirects`);\n\t\t\t\t}\n\t\t\t\tlet next: URL;\n\t\t\t\ttry {\n\t\t\t\t\tnext = new URL(location, current);\n\t\t\t\t} catch {\n\t\t\t\t\tthrow new PageError(\"invalid_redirect\", `redirect target is not a URL: ${JSON.stringify(location)}`);\n\t\t\t\t}\n\t\t\t\tif (next.protocol !== \"http:\" && next.protocol !== \"https:\") {\n\t\t\t\t\tthrow new PageError(\"invalid_redirect\", `redirect target must be http(s): ${next.toString()}`);\n\t\t\t\t}\n\t\t\t\tcurrent = next;\n\t\t\t\tcontinue;\n\t\t\t}\n\n\t\t\tif (response.status < 200 || response.status >= 300) {\n\t\t\t\tthrow new PageError(\"http_error\", `HTTP ${response.status}`, response.status);\n\t\t\t}\n\n\t\t\tconst contentType = contentTypeOf(response);\n\t\t\tif (!ALLOWED_CONTENT_TYPES.has(contentType)) {\n\t\t\t\tthrow new PageError(\n\t\t\t\t\t\"unsupported_content_type\",\n\t\t\t\t\t`content type ${contentType === \"\" ? \"(missing)\" : contentType} is not supported`,\n\t\t\t\t);\n\t\t\t}\n\t\t\tconst declaredLength = response.headers.get(\"content-length\");\n\t\t\tif (\n\t\t\t\tdeclaredLength !== null &&\n\t\t\t\tNumber.isInteger(Number(declaredLength)) &&\n\t\t\t\tNumber(declaredLength) > options.maxRawBytes\n\t\t\t) {\n\t\t\t\tthrow new PageError(\"too_large\", `response declares ${declaredLength} bytes`);\n\t\t\t}\n\t\t\tlet bodyText: string;\n\t\t\ttry {\n\t\t\t\tbodyText = await readBodyCapped(response, options.maxRawBytes);\n\t\t\t} catch (error) {\n\t\t\t\tconst bytesRead = error instanceof PageError ? error.bytesRead : undefined;\n\t\t\t\tif (options.signal?.aborted === true) {\n\t\t\t\t\tthrow new PageError(\"cancelled\", \"request cancelled by caller\", undefined, bytesRead);\n\t\t\t\t}\n\t\t\t\tif (controller.signal.aborted) {\n\t\t\t\t\tthrow new PageError(\"timeout\", `request timed out after ${options.timeoutMs}ms`, undefined, bytesRead);\n\t\t\t\t}\n\t\t\t\tthrow error;\n\t\t\t}\n\t\t\treturn {\n\t\t\t\tfinalUrl: current.toString(),\n\t\t\t\tcontentType,\n\t\t\t\tbodyText,\n\t\t\t\tbytesFetched: byteLength(bodyText),\n\t\t\t};\n\t\t} finally {\n\t\t\tclearTimeout(timer);\n\t\t\toptions.signal?.removeEventListener(\"abort\", abortFromCaller);\n\t\t}\n\t}\n\tthrow new PageError(\"too_many_redirects\", `more than ${SEARCH_LIMITS.maxRedirects} redirects`);\n}\n\nfunction extractPage(contentType: string, bodyText: string): { title?: string; text: string } {\n\tif (contentType === \"text/html\" || contentType === \"application/xhtml+xml\") {\n\t\tconst page = extractReadablePage(bodyText);\n\t\tif (page === undefined || page.text.trim() === \"\") {\n\t\t\tthrow new PageError(\"empty_content\", \"no readable text extracted; the page may require JavaScript\");\n\t\t}\n\t\treturn { ...(page.title === undefined ? {} : { title: page.title }), text: page.text };\n\t}\n\tconst text = bodyText.trim();\n\tif (text === \"\") throw new PageError(\"empty_content\", \"empty response body\");\n\treturn { text };\n}\n\nfunction isRetryable(error: unknown): boolean {\n\tif (!(error instanceof PageError)) return false;\n\tif (error.code === \"network_error\") return true;\n\treturn (\n\t\terror.code === \"http_error\" &&\n\t\terror.httpStatus !== undefined &&\n\t\t(error.httpStatus === 429 || error.httpStatus >= 500)\n\t);\n}\n\n/** Fetch and extract one page; bounded and explicit about every failure. */\nexport async function fetchPage(options: FetchPageOptionsV1): Promise<FetchedPageV1> {\n\tif (options.signal?.aborted === true) {\n\t\tthrow new PageError(\"cancelled\", \"request cancelled by caller\");\n\t}\n\tconst deadlineAt = Date.now() + options.timeoutMs;\n\tlet bytesRead = 0;\n\tlet lastError: unknown;\n\tfor (let attempt = 0; attempt < 2; attempt += 1) {\n\t\ttry {\n\t\t\tconst remainingRawBytes = Math.max(0, options.maxRawBytes - bytesRead);\n\t\t\tconst result = await fetchOnce({ ...options, maxRawBytes: remainingRawBytes }, deadlineAt);\n\t\t\tlet extracted: { title?: string; text: string };\n\t\t\ttry {\n\t\t\t\textracted = extractPage(result.contentType, result.bodyText);\n\t\t\t} catch (error) {\n\t\t\t\tif (error instanceof PageError) {\n\t\t\t\t\tthrow new PageError(error.code, error.message, error.httpStatus, result.bytesFetched);\n\t\t\t\t}\n\t\t\t\tthrow error;\n\t\t\t}\n\t\t\tconst meaningfulChars = extracted.text.trim().length;\n\t\t\tif (\n\t\t\t\tmeaningfulChars < MIN_MEANINGFUL_TEXT_CHARS &&\n\t\t\t\tresult.bytesFetched > JS_SHELL_RAW_BYTES &&\n\t\t\t\tmeaningfulChars / result.bytesFetched < JS_SHELL_TEXT_RATIO\n\t\t\t) {\n\t\t\t\tthrow new PageError(\n\t\t\t\t\t\"empty_content\",\n\t\t\t\t\t\"no meaningful text extracted; the page may require JavaScript or client-side rendering\",\n\t\t\t\t\tundefined,\n\t\t\t\t\tresult.bytesFetched,\n\t\t\t\t);\n\t\t\t}\n\t\t\tconst bounded = truncateToBytes(extracted.text, options.maxTextBytes);\n\t\t\treturn {\n\t\t\t\tfinalUrl: result.finalUrl,\n\t\t\t\t...(extracted.title === undefined ? {} : { title: extracted.title }),\n\t\t\t\tcontentType: result.contentType,\n\t\t\t\ttext: bounded.text,\n\t\t\t\ttruncated: bounded.truncated,\n\t\t\t\tbytesFetched: result.bytesFetched + bytesRead,\n\t\t\t};\n\t\t} catch (error) {\n\t\t\tif (error instanceof PageError && error.bytesRead !== undefined) bytesRead += error.bytesRead;\n\t\t\tlastError = error;\n\t\t\tif (attempt === 0 && isRetryable(error) && options.maxRawBytes > bytesRead) continue;\n\t\t\tif (error instanceof PageError) {\n\t\t\t\tthrow new PageError(error.code, error.message, error.httpStatus, bytesRead);\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t}\n\tthrow lastError instanceof Error ? lastError : new PageError(\"network_error\", String(lastError));\n}\n"]}

package/dist/index.d.ts

@@ -0,0 +1,5 @@
+export { PageCache, SummaryCache } from "./cache.ts";
+export { createSearchCapabilityFactoryV1, type SearchPluginOptionsV1 } from "./capability.ts";
+export { extractReadablePage, selectPassages } from "./extract.ts";
+export { SEARCH_LIMITS } from "./limits.ts";
+//# sourceMappingURL=index.d.ts.map

package/dist/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,SAAS,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AACrD,OAAO,EAAE,+BAA+B,EAAE,KAAK,qBAAqB,EAAE,MAAM,iBAAiB,CAAC;AAC9F,OAAO,EAAE,mBAAmB,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AACnE,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC","sourcesContent":["export { PageCache, SummaryCache } from \"./cache.ts\";\nexport { createSearchCapabilityFactoryV1, type SearchPluginOptionsV1 } from \"./capability.ts\";\nexport { extractReadablePage, selectPassages } from \"./extract.ts\";\nexport { SEARCH_LIMITS } from \"./limits.ts\";\n"]}

package/dist/index.js

@@ -0,0 +1,5 @@
+export { PageCache, SummaryCache } from "./cache.js";
+export { createSearchCapabilityFactoryV1 } from "./capability.js";
+export { extractReadablePage, selectPassages } from "./extract.js";
+export { SEARCH_LIMITS } from "./limits.js";
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,SAAS,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AACrD,OAAO,EAAE,+BAA+B,EAA8B,MAAM,iBAAiB,CAAC;AAC9F,OAAO,EAAE,mBAAmB,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AACnE,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC","sourcesContent":["export { PageCache, SummaryCache } from \"./cache.ts\";\nexport { createSearchCapabilityFactoryV1, type SearchPluginOptionsV1 } from \"./capability.ts\";\nexport { extractReadablePage, selectPassages } from \"./extract.ts\";\nexport { SEARCH_LIMITS } from \"./limits.ts\";\n"]}

package/dist/limits.d.ts

@@ -0,0 +1,39 @@
+/**
+ * Search tool resource ceilings (搜索工具设计 §8). These are resource bounds,
+ * not content policy: the model chooses URLs and no domain/scope restrictions
+ * are applied here.
+ *
+ * The page/extraction limits are coupled to the 5 MiB disk-cache entry size:
+ * a cached page is at most 5 MiB, so a fresh extraction keeps up to 4 MiB of
+ * text and reads at most 8 MiB of raw response bytes.
+ */
+export declare const SEARCH_LIMITS: Readonly<{
+    maxUrls: 32;
+    maxUrlLength: 2048;
+    maxQueryLength: 2000;
+    maxRawBytesPerPage: number;
+    maxTotalRawBytes: number;
+    maxExtractedBytesPerPage: number;
+    maxSummaryInputBytes: number;
+    maxSummaryOutputBytes: number;
+    maxSummaryOutputTokens: 8192;
+    maxPassagesPerPage: 24;
+    fetchConcurrency: 8;
+    urlTimeoutMs: 20000;
+    totalFetchTimeoutMs: 120000;
+    modelTimeoutMs: 120000;
+    maxRedirects: 5;
+    pageCacheBytes: number;
+    pageCacheEntries: 5000;
+    cacheMaxEntryBytes: number;
+    cacheBaseTtlMs: number;
+    cacheTtlTiersMs: readonly number[];
+    cacheMetaFlushMs: 5000;
+    cachePruneIntervalMs: number;
+    cacheWriteQueueEntries: 8;
+    cacheWriteQueueBytes: number;
+    summaryCacheEntries: 32;
+    summaryCacheTtlMs: number;
+}>;
+export type SearchMode = "model" | "url";
+//# sourceMappingURL=limits.d.ts.map

package/dist/limits.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"limits.d.ts","sourceRoot":"","sources":["../src/limits.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AACH,eAAO,MAAM,aAAa;;;;;;;;;;;;;;;;;;;;;;;;;;;EAgCxB,CAAC;AAEH,MAAM,MAAM,UAAU,GAAG,OAAO,GAAG,KAAK,CAAC","sourcesContent":["/**\n * Search tool resource ceilings (搜索工具设计 §8). These are resource bounds,\n * not content policy: the model chooses URLs and no domain/scope restrictions\n * are applied here.\n *\n * The page/extraction limits are coupled to the 5 MiB disk-cache entry size:\n * a cached page is at most 5 MiB, so a fresh extraction keeps up to 4 MiB of\n * text and reads at most 8 MiB of raw response bytes.\n */\nexport const SEARCH_LIMITS = Object.freeze({\n\tmaxUrls: 32,\n\tmaxUrlLength: 2048,\n\tmaxQueryLength: 2000,\n\tmaxRawBytesPerPage: 8 * 1024 * 1024,\n\tmaxTotalRawBytes: 128 * 1024 * 1024,\n\tmaxExtractedBytesPerPage: 4 * 1024 * 1024,\n\tmaxSummaryInputBytes: 512 * 1024,\n\tmaxSummaryOutputBytes: 16 * 1024,\n\tmaxSummaryOutputTokens: 8192,\n\tmaxPassagesPerPage: 24,\n\tfetchConcurrency: 8,\n\turlTimeoutMs: 20_000,\n\ttotalFetchTimeoutMs: 120_000,\n\tmodelTimeoutMs: 120_000,\n\tmaxRedirects: 5,\n\tpageCacheBytes: 2 * 1024 * 1024 * 1024,\n\tpageCacheEntries: 5000,\n\tcacheMaxEntryBytes: 5 * 1024 * 1024,\n\tcacheBaseTtlMs: 14 * 24 * 60 * 60 * 1000,\n\tcacheTtlTiersMs: Object.freeze([\n\t\t30 * 24 * 60 * 60 * 1000,\n\t\t60 * 24 * 60 * 60 * 1000,\n\t\t120 * 24 * 60 * 60 * 1000,\n\t\t365 * 24 * 60 * 60 * 1000,\n\t]),\n\tcacheMetaFlushMs: 5_000,\n\tcachePruneIntervalMs: 5 * 60 * 1000,\n\tcacheWriteQueueEntries: 8,\n\tcacheWriteQueueBytes: 32 * 1024 * 1024,\n\tsummaryCacheEntries: 32,\n\tsummaryCacheTtlMs: 10 * 60 * 1000,\n});\n\nexport type SearchMode = \"model\" | \"url\";\n"]}

package/dist/limits.js

@@ -0,0 +1,43 @@
+/**
+ * Search tool resource ceilings (搜索工具设计 §8). These are resource bounds,
+ * not content policy: the model chooses URLs and no domain/scope restrictions
+ * are applied here.
+ *
+ * The page/extraction limits are coupled to the 5 MiB disk-cache entry size:
+ * a cached page is at most 5 MiB, so a fresh extraction keeps up to 4 MiB of
+ * text and reads at most 8 MiB of raw response bytes.
+ */
+export const SEARCH_LIMITS = Object.freeze({
+    maxUrls: 32,
+    maxUrlLength: 2048,
+    maxQueryLength: 2000,
+    maxRawBytesPerPage: 8 * 1024 * 1024,
+    maxTotalRawBytes: 128 * 1024 * 1024,
+    maxExtractedBytesPerPage: 4 * 1024 * 1024,
+    maxSummaryInputBytes: 512 * 1024,
+    maxSummaryOutputBytes: 16 * 1024,
+    maxSummaryOutputTokens: 8192,
+    maxPassagesPerPage: 24,
+    fetchConcurrency: 8,
+    urlTimeoutMs: 20_000,
+    totalFetchTimeoutMs: 120_000,
+    modelTimeoutMs: 120_000,
+    maxRedirects: 5,
+    pageCacheBytes: 2 * 1024 * 1024 * 1024,
+    pageCacheEntries: 5000,
+    cacheMaxEntryBytes: 5 * 1024 * 1024,
+    cacheBaseTtlMs: 14 * 24 * 60 * 60 * 1000,
+    cacheTtlTiersMs: Object.freeze([
+        30 * 24 * 60 * 60 * 1000,
+        60 * 24 * 60 * 60 * 1000,
+        120 * 24 * 60 * 60 * 1000,
+        365 * 24 * 60 * 60 * 1000,
+    ]),
+    cacheMetaFlushMs: 5_000,
+    cachePruneIntervalMs: 5 * 60 * 1000,
+    cacheWriteQueueEntries: 8,
+    cacheWriteQueueBytes: 32 * 1024 * 1024,
+    summaryCacheEntries: 32,
+    summaryCacheTtlMs: 10 * 60 * 1000,
+});
+//# sourceMappingURL=limits.js.map

package/dist/limits.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"limits.js","sourceRoot":"","sources":["../src/limits.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AACH,MAAM,CAAC,MAAM,aAAa,GAAG,MAAM,CAAC,MAAM,CAAC;IAC1C,OAAO,EAAE,EAAE;IACX,YAAY,EAAE,IAAI;IAClB,cAAc,EAAE,IAAI;IACpB,kBAAkB,EAAE,CAAC,GAAG,IAAI,GAAG,IAAI;IACnC,gBAAgB,EAAE,GAAG,GAAG,IAAI,GAAG,IAAI;IACnC,wBAAwB,EAAE,CAAC,GAAG,IAAI,GAAG,IAAI;IACzC,oBAAoB,EAAE,GAAG,GAAG,IAAI;IAChC,qBAAqB,EAAE,EAAE,GAAG,IAAI;IAChC,sBAAsB,EAAE,IAAI;IAC5B,kBAAkB,EAAE,EAAE;IACtB,gBAAgB,EAAE,CAAC;IACnB,YAAY,EAAE,MAAM;IACpB,mBAAmB,EAAE,OAAO;IAC5B,cAAc,EAAE,OAAO;IACvB,YAAY,EAAE,CAAC;IACf,cAAc,EAAE,CAAC,GAAG,IAAI,GAAG,IAAI,GAAG,IAAI;IACtC,gBAAgB,EAAE,IAAI;IACtB,kBAAkB,EAAE,CAAC,GAAG,IAAI,GAAG,IAAI;IACnC,cAAc,EAAE,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI;IACxC,eAAe,EAAE,MAAM,CAAC,MAAM,CAAC;QAC9B,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI;QACxB,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI;QACxB,GAAG,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI;QACzB,GAAG,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI;KACzB,CAAC;IACF,gBAAgB,EAAE,KAAK;IACvB,oBAAoB,EAAE,CAAC,GAAG,EAAE,GAAG,IAAI;IACnC,sBAAsB,EAAE,CAAC;IACzB,oBAAoB,EAAE,EAAE,GAAG,IAAI,GAAG,IAAI;IACtC,mBAAmB,EAAE,EAAE;IACvB,iBAAiB,EAAE,EAAE,GAAG,EAAE,GAAG,IAAI;CACjC,CAAC,CAAC","sourcesContent":["/**\n * Search tool resource ceilings (搜索工具设计 §8). These are resource bounds,\n * not content policy: the model chooses URLs and no domain/scope restrictions\n * are applied here.\n *\n * The page/extraction limits are coupled to the 5 MiB disk-cache entry size:\n * a cached page is at most 5 MiB, so a fresh extraction keeps up to 4 MiB of\n * text and reads at most 8 MiB of raw response bytes.\n */\nexport const SEARCH_LIMITS = Object.freeze({\n\tmaxUrls: 32,\n\tmaxUrlLength: 2048,\n\tmaxQueryLength: 2000,\n\tmaxRawBytesPerPage: 8 * 1024 * 1024,\n\tmaxTotalRawBytes: 128 * 1024 * 1024,\n\tmaxExtractedBytesPerPage: 4 * 1024 * 1024,\n\tmaxSummaryInputBytes: 512 * 1024,\n\tmaxSummaryOutputBytes: 16 * 1024,\n\tmaxSummaryOutputTokens: 8192,\n\tmaxPassagesPerPage: 24,\n\tfetchConcurrency: 8,\n\turlTimeoutMs: 20_000,\n\ttotalFetchTimeoutMs: 120_000,\n\tmodelTimeoutMs: 120_000,\n\tmaxRedirects: 5,\n\tpageCacheBytes: 2 * 1024 * 1024 * 1024,\n\tpageCacheEntries: 5000,\n\tcacheMaxEntryBytes: 5 * 1024 * 1024,\n\tcacheBaseTtlMs: 14 * 24 * 60 * 60 * 1000,\n\tcacheTtlTiersMs: Object.freeze([\n\t\t30 * 24 * 60 * 60 * 1000,\n\t\t60 * 24 * 60 * 60 * 1000,\n\t\t120 * 24 * 60 * 60 * 1000,\n\t\t365 * 24 * 60 * 60 * 1000,\n\t]),\n\tcacheMetaFlushMs: 5_000,\n\tcachePruneIntervalMs: 5 * 60 * 1000,\n\tcacheWriteQueueEntries: 8,\n\tcacheWriteQueueBytes: 32 * 1024 * 1024,\n\tsummaryCacheEntries: 32,\n\tsummaryCacheTtlMs: 10 * 60 * 1000,\n});\n\nexport type SearchMode = \"model\" | \"url\";\n"]}

package/dist/model-call.d.ts

@@ -0,0 +1,10 @@
+export interface ModelCallSignalV1 {
+    readonly signal: AbortSignal;
+    timedOut(): boolean;
+    dispose(): void;
+}
+/** Combine a caller signal with the model-call deadline without leaking timers. */
+export declare function createModelCallSignal(caller: AbortSignal | undefined): ModelCallSignalV1;
+/** Combine the per-call signal with the overall fetch budget. */
+export declare function createFetchBudgetSignal(caller: AbortSignal | undefined): ModelCallSignalV1;
+//# sourceMappingURL=model-call.d.ts.map

package/dist/model-call.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"model-call.d.ts","sourceRoot":"","sources":["../src/model-call.ts"],"names":[],"mappings":"AAEA,MAAM,WAAW,iBAAiB;IACjC,QAAQ,CAAC,MAAM,EAAE,WAAW,CAAC;IAC7B,QAAQ,IAAI,OAAO,CAAC;IACpB,OAAO,IAAI,IAAI,CAAC;CAChB;AAED,mFAAmF;AACnF,wBAAgB,qBAAqB,CAAC,MAAM,EAAE,WAAW,GAAG,SAAS,GAAG,iBAAiB,CAkBxF;AAED,iEAAiE;AACjE,wBAAgB,uBAAuB,CAAC,MAAM,EAAE,WAAW,GAAG,SAAS,GAAG,iBAAiB,CAkB1F","sourcesContent":["import { SEARCH_LIMITS } from \"./limits.ts\";\n\nexport interface ModelCallSignalV1 {\n\treadonly signal: AbortSignal;\n\ttimedOut(): boolean;\n\tdispose(): void;\n}\n\n/** Combine a caller signal with the model-call deadline without leaking timers. */\nexport function createModelCallSignal(caller: AbortSignal | undefined): ModelCallSignalV1 {\n\tconst controller = new AbortController();\n\tlet timedOut = false;\n\tconst timer = setTimeout(() => {\n\t\ttimedOut = true;\n\t\tcontroller.abort();\n\t}, SEARCH_LIMITS.modelTimeoutMs);\n\tconst abortFromCaller = (): void => controller.abort();\n\tif (caller?.aborted === true) controller.abort();\n\telse caller?.addEventListener(\"abort\", abortFromCaller, { once: true });\n\treturn {\n\t\tsignal: controller.signal,\n\t\ttimedOut: () => timedOut,\n\t\tdispose: () => {\n\t\t\tclearTimeout(timer);\n\t\t\tcaller?.removeEventListener(\"abort\", abortFromCaller);\n\t\t},\n\t};\n}\n\n/** Combine the per-call signal with the overall fetch budget. */\nexport function createFetchBudgetSignal(caller: AbortSignal | undefined): ModelCallSignalV1 {\n\tconst controller = new AbortController();\n\tlet timedOut = false;\n\tconst timer = setTimeout(() => {\n\t\ttimedOut = true;\n\t\tcontroller.abort();\n\t}, SEARCH_LIMITS.totalFetchTimeoutMs);\n\tconst abortFromCaller = (): void => controller.abort();\n\tif (caller?.aborted === true) controller.abort();\n\telse caller?.addEventListener(\"abort\", abortFromCaller, { once: true });\n\treturn {\n\t\tsignal: controller.signal,\n\t\ttimedOut: () => timedOut,\n\t\tdispose: () => {\n\t\t\tclearTimeout(timer);\n\t\t\tcaller?.removeEventListener(\"abort\", abortFromCaller);\n\t\t},\n\t};\n}\n"]}