| 1 | //! Web browsing tool with multi-command support (search/open/click/find/screenshot). |
| 2 | //! |
| 3 | //! This mirrors the Codex harness `web.run` interface so models can use a single |
| 4 | //! tool call to perform multiple web actions and cite sources with ref_ids. |
| 5 | |
| 6 | use super::spec::{ |
| 7 | ApprovalRequirement, ToolCapability, ToolContext, ToolError, ToolResult, ToolSpec, |
| 8 | optional_u64, required_str, |
| 9 | }; |
| 10 | use crate::network_policy::{Decision, host_from_url}; |
| 11 | use async_trait::async_trait; |
| 12 | use base64::{Engine as _, engine::general_purpose}; |
| 13 | use regex::Regex; |
| 14 | use serde::{Deserialize, Serialize}; |
| 15 | use serde_json::{Value, json}; |
| 16 | use std::collections::{HashMap, VecDeque}; |
| 17 | use std::hash::{Hash, Hasher}; |
| 18 | use std::sync::{Mutex, OnceLock}; |
| 19 | use std::time::{Duration, Instant}; |
| 20 | |
| 21 | const MAX_RESULTS: usize = 10; |
| 22 | const DEFAULT_TIMEOUT_MS: u64 = 15_000; |
| 23 | const DEFAULT_OPEN_TIMEOUT_MS: u64 = 20_000; |
| 24 | const MAX_WEB_RUN_SESSIONS: usize = 64; |
| 25 | const MAX_PAGES_PER_SESSION: usize = 256; |
| 26 | const WEB_RUN_SESSION_TTL: Duration = Duration::from_secs(30 * 60); |
| 27 | const USER_AGENT: &str = "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/17.0 Safari/605.1.15"; |
| 28 | |
| 29 | static WEB_RUN_STATE: OnceLock<Mutex<WebRunState>> = OnceLock::new(); |
| 30 | |
| 31 | #[derive(Default)] |
| 32 | struct WebRunState { |
| 33 | sessions: HashMap<String, WebRunSessionState>, |
| 34 | pages: HashMap<String, StoredWebPage>, |
| 35 | } |
| 36 | |
| 37 | struct WebRunSessionState { |
| 38 | next_turn: u64, |
| 39 | refs: VecDeque<String>, |
| 40 | last_access: Instant, |
| 41 | } |
| 42 | |
| 43 | impl Default for WebRunSessionState { |
| 44 | fn default() -> Self { |
| 45 | Self { |
| 46 | next_turn: 0, |
| 47 | refs: VecDeque::new(), |
| 48 | last_access: Instant::now(), |
| 49 | } |
| 50 | } |
| 51 | } |
| 52 | |
| 53 | #[derive(Debug, Clone)] |
| 54 | struct StoredWebPage { |
| 55 | namespace: String, |
| 56 | page: WebPage, |
| 57 | } |
| 58 | |
| 59 | impl WebRunState { |
| 60 | fn cleanup(&mut self) { |
| 61 | let now = Instant::now(); |
| 62 | let expired = self |
| 63 | .sessions |
| 64 | .iter() |
| 65 | .filter_map(|(namespace, session)| { |
| 66 | if now.duration_since(session.last_access) > WEB_RUN_SESSION_TTL { |
| 67 | Some(namespace.clone()) |
| 68 | } else { |
| 69 | None |
| 70 | } |
| 71 | }) |
| 72 | .collect::<Vec<_>>(); |
| 73 | for namespace in expired { |
| 74 | self.remove_session(&namespace); |
| 75 | } |
| 76 | |
| 77 | while self.sessions.len() > MAX_WEB_RUN_SESSIONS { |
| 78 | let Some(oldest_namespace) = self |
| 79 | .sessions |
| 80 | .iter() |
| 81 | .min_by_key(|(_, session)| session.last_access) |
| 82 | .map(|(namespace, _)| namespace.clone()) |
| 83 | else { |
| 84 | break; |
| 85 | }; |
| 86 | self.remove_session(&oldest_namespace); |
| 87 | } |
| 88 | } |
| 89 | |
| 90 | fn remove_session(&mut self, namespace: &str) { |
| 91 | if let Some(session) = self.sessions.remove(namespace) { |
| 92 | for ref_id in session.refs { |
| 93 | self.pages.remove(&ref_id); |
| 94 | } |
| 95 | } |
| 96 | } |
| 97 | |
| 98 | fn touch_session(&mut self, namespace: &str) { |
| 99 | self.cleanup(); |
| 100 | if !self.sessions.contains_key(namespace) |
| 101 | && self.sessions.len() >= MAX_WEB_RUN_SESSIONS |
| 102 | && let Some(oldest_namespace) = self |
| 103 | .sessions |
| 104 | .iter() |
| 105 | .min_by_key(|(_, session)| session.last_access) |
| 106 | .map(|(existing_namespace, _)| existing_namespace.clone()) |
| 107 | { |
| 108 | self.remove_session(&oldest_namespace); |
| 109 | } |
| 110 | |
| 111 | let session = self.sessions.entry(namespace.to_string()).or_default(); |
| 112 | session.last_access = Instant::now(); |
| 113 | } |
| 114 | |
| 115 | fn next_turn(&mut self, namespace: &str) -> u64 { |
| 116 | self.touch_session(namespace); |
| 117 | let session = self |
| 118 | .sessions |
| 119 | .get_mut(namespace) |
| 120 | .expect("session should exist after touch"); |
| 121 | let current = session.next_turn; |
| 122 | session.next_turn = session.next_turn.saturating_add(1); |
| 123 | current |
| 124 | } |
| 125 | |
| 126 | fn store_page(&mut self, namespace: &str, ref_id: &str, page: WebPage) { |
| 127 | self.touch_session(namespace); |
| 128 | let mut evicted_refs = Vec::new(); |
| 129 | { |
| 130 | let session = self |
| 131 | .sessions |
| 132 | .get_mut(namespace) |
| 133 | .expect("session should exist after touch"); |
| 134 | if let Some(existing_idx) = session.refs.iter().position(|existing| existing == ref_id) |
| 135 | { |
| 136 | session.refs.remove(existing_idx); |
| 137 | } |
| 138 | session.refs.push_back(ref_id.to_string()); |
| 139 | |
| 140 | while session.refs.len() > MAX_PAGES_PER_SESSION { |
| 141 | if let Some(evicted_ref) = session.refs.pop_front() { |
| 142 | evicted_refs.push(evicted_ref); |
| 143 | } |
| 144 | } |
| 145 | } |
| 146 | |
| 147 | self.pages.insert( |
| 148 | ref_id.to_string(), |
| 149 | StoredWebPage { |
| 150 | namespace: namespace.to_string(), |
| 151 | page, |
| 152 | }, |
| 153 | ); |
| 154 | for evicted_ref in evicted_refs { |
| 155 | self.pages.remove(&evicted_ref); |
| 156 | } |
| 157 | } |
| 158 | |
| 159 | fn get_page(&mut self, ref_id: &str) -> Option<WebPage> { |
| 160 | self.cleanup(); |
| 161 | let stored = self.pages.get(ref_id)?.clone(); |
| 162 | if let Some(session) = self.sessions.get_mut(&stored.namespace) { |
| 163 | session.last_access = Instant::now(); |
| 164 | } |
| 165 | Some(stored.page) |
| 166 | } |
| 167 | } |
| 168 | |
| 169 | #[derive(Debug, Clone, Serialize)] |
| 170 | struct WebLink { |
| 171 | id: usize, |
| 172 | url: String, |
| 173 | text: String, |
| 174 | } |
| 175 | |
| 176 | #[derive(Debug, Clone)] |
| 177 | struct WebPage { |
| 178 | url: String, |
| 179 | title: Option<String>, |
| 180 | content_type: Option<String>, |
| 181 | lines: Vec<String>, |
| 182 | links: Vec<WebLink>, |
| 183 | pdf_pages: Option<Vec<Vec<String>>>, |
| 184 | } |
| 185 | |
| 186 | #[derive(Debug, Clone, Copy)] |
| 187 | enum ResponseLength { |
| 188 | Short, |
| 189 | Medium, |
| 190 | Long, |
| 191 | } |
| 192 | |
| 193 | impl ResponseLength { |
| 194 | fn from_input(input: Option<&Value>) -> Self { |
| 195 | let raw = input.and_then(|v| v.as_str()).unwrap_or("medium"); |
| 196 | match raw.to_lowercase().as_str() { |
| 197 | "short" => Self::Short, |
| 198 | "long" => Self::Long, |
| 199 | _ => Self::Medium, |
| 200 | } |
| 201 | } |
| 202 | |
| 203 | fn view_lines(self) -> usize { |
| 204 | match self { |
| 205 | Self::Short => 40, |
| 206 | Self::Medium => 80, |
| 207 | Self::Long => 160, |
| 208 | } |
| 209 | } |
| 210 | |
| 211 | fn wrap_width(self) -> usize { |
| 212 | match self { |
| 213 | Self::Short => 88, |
| 214 | Self::Medium => 110, |
| 215 | Self::Long => 140, |
| 216 | } |
| 217 | } |
| 218 | |
| 219 | fn max_results(self) -> usize { |
| 220 | match self { |
| 221 | Self::Short => 5, |
| 222 | Self::Medium => 8, |
| 223 | Self::Long => 10, |
| 224 | } |
| 225 | } |
| 226 | |
| 227 | fn max_find_matches(self) -> usize { |
| 228 | match self { |
| 229 | Self::Short => 8, |
| 230 | Self::Medium => 15, |
| 231 | Self::Long => 30, |
| 232 | } |
| 233 | } |
| 234 | } |
| 235 | |
| 236 | #[derive(Debug, Clone, Serialize)] |
| 237 | struct SearchEntry { |
| 238 | title: String, |
| 239 | url: String, |
| 240 | snippet: Option<String>, |
| 241 | } |
| 242 | |
| 243 | #[derive(Debug, Clone, Serialize)] |
| 244 | struct SearchResult { |
| 245 | ref_id: String, |
| 246 | query: String, |
| 247 | source: String, |
| 248 | count: usize, |
| 249 | results: Vec<SearchEntry>, |
| 250 | #[serde(skip_serializing_if = "Option::is_none")] |
| 251 | warning: Option<String>, |
| 252 | } |
| 253 | |
| 254 | #[derive(Debug, Clone, Serialize)] |
| 255 | struct PageViewResult { |
| 256 | ref_id: String, |
| 257 | url: String, |
| 258 | #[serde(skip_serializing_if = "Option::is_none")] |
| 259 | title: Option<String>, |
| 260 | #[serde(skip_serializing_if = "Option::is_none")] |
| 261 | content_type: Option<String>, |
| 262 | line_start: usize, |
| 263 | line_end: usize, |
| 264 | total_lines: usize, |
| 265 | content: String, |
| 266 | links: Vec<WebLink>, |
| 267 | } |
| 268 | |
| 269 | #[derive(Debug, Clone, Serialize)] |
| 270 | struct FindMatch { |
| 271 | line: usize, |
| 272 | text: String, |
| 273 | } |
| 274 | |
| 275 | #[derive(Debug, Clone, Serialize)] |
| 276 | struct FindResult { |
| 277 | ref_id: String, |
| 278 | pattern: String, |
| 279 | count: usize, |
| 280 | matches: Vec<FindMatch>, |
| 281 | } |
| 282 | |
| 283 | #[derive(Debug, Clone, Serialize)] |
| 284 | struct ScreenshotResult { |
| 285 | ref_id: String, |
| 286 | pageno: usize, |
| 287 | total_pages: usize, |
| 288 | content: String, |
| 289 | } |
| 290 | |
| 291 | #[derive(Debug, Clone, Serialize)] |
| 292 | struct ImageResultEntry { |
| 293 | image: String, |
| 294 | #[serde(skip_serializing_if = "Option::is_none")] |
| 295 | thumbnail: Option<String>, |
| 296 | #[serde(skip_serializing_if = "Option::is_none")] |
| 297 | title: Option<String>, |
| 298 | #[serde(skip_serializing_if = "Option::is_none")] |
| 299 | url: Option<String>, |
| 300 | #[serde(skip_serializing_if = "Option::is_none")] |
| 301 | source: Option<String>, |
| 302 | #[serde(skip_serializing_if = "Option::is_none")] |
| 303 | width: Option<u32>, |
| 304 | #[serde(skip_serializing_if = "Option::is_none")] |
| 305 | height: Option<u32>, |
| 306 | } |
| 307 | |
| 308 | #[derive(Debug, Clone, Serialize)] |
| 309 | struct ImageQueryResult { |
| 310 | query: String, |
| 311 | source: String, |
| 312 | count: usize, |
| 313 | results: Vec<ImageResultEntry>, |
| 314 | #[serde(skip_serializing_if = "Option::is_none")] |
| 315 | warning: Option<String>, |
| 316 | } |
| 317 | |
| 318 | #[derive(Debug, Clone, Serialize, Default)] |
| 319 | struct WebRunOutput { |
| 320 | #[serde(skip_serializing_if = "Option::is_none")] |
| 321 | search_query: Option<Vec<SearchResult>>, |
| 322 | #[serde(skip_serializing_if = "Option::is_none")] |
| 323 | image_query: Option<Vec<ImageQueryResult>>, |
| 324 | #[serde(skip_serializing_if = "Option::is_none")] |
| 325 | open: Option<Vec<PageViewResult>>, |
| 326 | #[serde(skip_serializing_if = "Option::is_none")] |
| 327 | click: Option<Vec<PageViewResult>>, |
| 328 | #[serde(skip_serializing_if = "Option::is_none")] |
| 329 | find: Option<Vec<FindResult>>, |
| 330 | #[serde(skip_serializing_if = "Option::is_none")] |
| 331 | screenshot: Option<Vec<ScreenshotResult>>, |
| 332 | #[serde(skip_serializing_if = "Vec::is_empty", default)] |
| 333 | warnings: Vec<String>, |
| 334 | } |
| 335 | |
| 336 | pub struct WebRunTool; |
| 337 | |
| 338 | #[async_trait] |
| 339 | impl ToolSpec for WebRunTool { |
| 340 | fn name(&self) -> &'static str { |
| 341 | "web.run" |
| 342 | } |
| 343 | |
| 344 | fn description(&self) -> &'static str { |
| 345 | "Browse the web (search/open/click/find/screenshot/image_query) and return structured results with ref_ids for citations." |
| 346 | } |
| 347 | |
| 348 | fn input_schema(&self) -> Value { |
| 349 | json!({ |
| 350 | "type": "object", |
| 351 | "properties": { |
| 352 | "search_query": { |
| 353 | "type": "array", |
| 354 | "items": { |
| 355 | "type": "object", |
| 356 | "properties": { |
| 357 | "q": { "type": "string" }, |
| 358 | "recency": { "type": "integer" }, |
| 359 | "max_results": { "type": "integer" }, |
| 360 | "timeout_ms": { "type": "integer" }, |
| 361 | "domains": { "type": "array", "items": { "type": "string" } } |
| 362 | }, |
| 363 | "required": ["q"] |
| 364 | } |
| 365 | }, |
| 366 | "image_query": { |
| 367 | "type": "array", |
| 368 | "items": { |
| 369 | "type": "object", |
| 370 | "properties": { |
| 371 | "q": { "type": "string" }, |
| 372 | "recency": { "type": "integer" }, |
| 373 | "max_results": { "type": "integer" }, |
| 374 | "timeout_ms": { "type": "integer" }, |
| 375 | "domains": { "type": "array", "items": { "type": "string" } } |
| 376 | }, |
| 377 | "required": ["q"] |
| 378 | } |
| 379 | }, |
| 380 | "open": { |
| 381 | "type": "array", |
| 382 | "items": { |
| 383 | "type": "object", |
| 384 | "properties": { |
| 385 | "ref_id": { "type": "string" }, |
| 386 | "lineno": { "type": "integer" } |
| 387 | }, |
| 388 | "required": ["ref_id"] |
| 389 | } |
| 390 | }, |
| 391 | "click": { |
| 392 | "type": "array", |
| 393 | "items": { |
| 394 | "type": "object", |
| 395 | "properties": { |
| 396 | "ref_id": { "type": "string" }, |
| 397 | "id": { "type": "integer" } |
| 398 | }, |
| 399 | "required": ["ref_id", "id"] |
| 400 | } |
| 401 | }, |
| 402 | "find": { |
| 403 | "type": "array", |
| 404 | "items": { |
| 405 | "type": "object", |
| 406 | "properties": { |
| 407 | "ref_id": { "type": "string" }, |
| 408 | "pattern": { "type": "string" } |
| 409 | }, |
| 410 | "required": ["ref_id", "pattern"] |
| 411 | } |
| 412 | }, |
| 413 | "screenshot": { |
| 414 | "type": "array", |
| 415 | "items": { |
| 416 | "type": "object", |
| 417 | "properties": { |
| 418 | "ref_id": { "type": "string" }, |
| 419 | "pageno": { "type": "integer" } |
| 420 | }, |
| 421 | "required": ["ref_id", "pageno"] |
| 422 | } |
| 423 | }, |
| 424 | "response_length": { |
| 425 | "type": "string", |
| 426 | "enum": ["short", "medium", "long"], |
| 427 | "description": "Controls result verbosity" |
| 428 | } |
| 429 | } |
| 430 | }) |
| 431 | } |
| 432 | |
| 433 | fn capabilities(&self) -> Vec<ToolCapability> { |
| 434 | vec![ToolCapability::ReadOnly, ToolCapability::Network] |
| 435 | } |
| 436 | |
| 437 | fn approval_requirement(&self) -> ApprovalRequirement { |
| 438 | ApprovalRequirement::Auto |
| 439 | } |
| 440 | |
| 441 | async fn execute(&self, input: Value, context: &ToolContext) -> Result<ToolResult, ToolError> { |
| 442 | let response_length = ResponseLength::from_input(input.get("response_length")); |
| 443 | let mut output = WebRunOutput::default(); |
| 444 | let scope = scoped_ref_prefix(&context.state_namespace); |
| 445 | let turn = with_state(|state| state.next_turn(&context.state_namespace)); |
| 446 | |
| 447 | let mut search_counter = 0usize; |
| 448 | let mut view_counter = 0usize; |
| 449 | let mut click_counter = 0usize; |
| 450 | |
| 451 | if let Some(searches) = input.get("search_query").and_then(|v| v.as_array()) { |
| 452 | let mut results = Vec::new(); |
| 453 | for search in searches { |
| 454 | let query = required_str(search, "q")?.trim().to_string(); |
| 455 | if query.is_empty() { |
| 456 | continue; |
| 457 | } |
| 458 | let recency = optional_u64(search, "recency", 0); |
| 459 | let max_results = usize::try_from(optional_u64( |
| 460 | search, |
| 461 | "max_results", |
| 462 | response_length.max_results() as u64, |
| 463 | )) |
| 464 | .unwrap_or(response_length.max_results()) |
| 465 | .clamp(1, MAX_RESULTS); |
| 466 | let timeout_ms = optional_u64(search, "timeout_ms", DEFAULT_TIMEOUT_MS).min(60_000); |
| 467 | |
| 468 | let domains = search |
| 469 | .get("domains") |
| 470 | .and_then(|v| v.as_array()) |
| 471 | .map(|arr| { |
| 472 | arr.iter() |
| 473 | .filter_map(|v| v.as_str().map(|s| s.to_string())) |
| 474 | .collect::<Vec<_>>() |
| 475 | }) |
| 476 | .unwrap_or_default(); |
| 477 | |
| 478 | let (entries, source, warning) = |
| 479 | run_search(&query, max_results, timeout_ms, &domains).await?; |
| 480 | let mut warnings = Vec::new(); |
| 481 | if recency > 0 { |
| 482 | warnings.push(format!( |
| 483 | "Recency filter not enforced (requested last {recency} days)" |
| 484 | )); |
| 485 | } |
| 486 | if let Some(w) = warning { |
| 487 | warnings.push(w); |
| 488 | } |
| 489 | search_counter += 1; |
| 490 | let ref_id = format!("{scope}turn{turn}search{search_counter}"); |
| 491 | |
| 492 | let page = page_from_search(&query, &entries); |
| 493 | store_page(&context.state_namespace, &ref_id, page); |
| 494 | |
| 495 | results.push(SearchResult { |
| 496 | ref_id, |
| 497 | query, |
| 498 | source, |
| 499 | count: entries.len(), |
| 500 | results: entries, |
| 501 | warning: if warnings.is_empty() { |
| 502 | None |
| 503 | } else { |
| 504 | Some(warnings.join("; ")) |
| 505 | }, |
| 506 | }); |
| 507 | } |
| 508 | if !results.is_empty() { |
| 509 | output.search_query = Some(results); |
| 510 | } |
| 511 | } |
| 512 | |
| 513 | if let Some(images) = input.get("image_query").and_then(|v| v.as_array()) { |
| 514 | let mut results = Vec::new(); |
| 515 | for image in images { |
| 516 | let query = required_str(image, "q")?.trim().to_string(); |
| 517 | if query.is_empty() { |
| 518 | continue; |
| 519 | } |
| 520 | let recency = optional_u64(image, "recency", 0); |
| 521 | let max_results = usize::try_from(optional_u64( |
| 522 | image, |
| 523 | "max_results", |
| 524 | response_length.max_results() as u64, |
| 525 | )) |
| 526 | .unwrap_or(response_length.max_results()) |
| 527 | .clamp(1, MAX_RESULTS); |
| 528 | let timeout_ms = optional_u64(image, "timeout_ms", DEFAULT_TIMEOUT_MS).min(60_000); |
| 529 | |
| 530 | let domains = image |
| 531 | .get("domains") |
| 532 | .and_then(|v| v.as_array()) |
| 533 | .map(|arr| { |
| 534 | arr.iter() |
| 535 | .filter_map(|v| v.as_str().map(|s| s.to_string())) |
| 536 | .collect::<Vec<_>>() |
| 537 | }) |
| 538 | .unwrap_or_default(); |
| 539 | |
| 540 | let (entries, warning) = |
| 541 | run_image_search(&query, max_results, timeout_ms, &domains).await?; |
| 542 | |
| 543 | let mut warnings = Vec::new(); |
| 544 | if recency > 0 { |
| 545 | warnings.push(format!( |
| 546 | "Recency filter not enforced (requested last {recency} days)" |
| 547 | )); |
| 548 | } |
| 549 | if let Some(w) = warning { |
| 550 | warnings.push(w); |
| 551 | } |
| 552 | |
| 553 | results.push(ImageQueryResult { |
| 554 | query, |
| 555 | source: "duckduckgo_images".to_string(), |
| 556 | count: entries.len(), |
| 557 | results: entries, |
| 558 | warning: if warnings.is_empty() { |
| 559 | None |
| 560 | } else { |
| 561 | Some(warnings.join("; ")) |
| 562 | }, |
| 563 | }); |
| 564 | } |
| 565 | if !results.is_empty() { |
| 566 | output.image_query = Some(results); |
| 567 | } |
| 568 | } |
| 569 | |
| 570 | if let Some(opens) = input.get("open").and_then(|v| v.as_array()) { |
| 571 | let mut views = Vec::new(); |
| 572 | for open in opens { |
| 573 | let ref_id = required_str(open, "ref_id")?.to_string(); |
| 574 | let lineno = optional_u64(open, "lineno", 1).max(1) as usize; |
| 575 | |
| 576 | let page = resolve_or_fetch_page(&ref_id, DEFAULT_OPEN_TIMEOUT_MS, context).await?; |
| 577 | view_counter += 1; |
| 578 | let view_ref = format!("{scope}turn{turn}view{view_counter}"); |
| 579 | store_page(&context.state_namespace, &view_ref, page.clone()); |
| 580 | |
| 581 | let view = render_view(&view_ref, &page, lineno, response_length); |
| 582 | views.push(view); |
| 583 | } |
| 584 | if !views.is_empty() { |
| 585 | output.open = Some(views); |
| 586 | } |
| 587 | } |
| 588 | |
| 589 | if let Some(clicks) = input.get("click").and_then(|v| v.as_array()) { |
| 590 | let mut views = Vec::new(); |
| 591 | for click in clicks { |
| 592 | let ref_id = required_str(click, "ref_id")?.to_string(); |
| 593 | let link_id = optional_u64(click, "id", 0) as usize; |
| 594 | if link_id == 0 { |
| 595 | return Err(ToolError::invalid_input("click.id must be >= 1")); |
| 596 | } |
| 597 | let page = get_page(&ref_id).ok_or_else(|| { |
| 598 | ToolError::invalid_input(format!("Unknown ref_id '{ref_id}'")) |
| 599 | })?; |
| 600 | let link = page.links.iter().find(|l| l.id == link_id).ok_or_else(|| { |
| 601 | ToolError::invalid_input(format!( |
| 602 | "Link id {link_id} not found for ref_id '{ref_id}'" |
| 603 | )) |
| 604 | })?; |
| 605 | let target = link.url.clone(); |
| 606 | let fetched = |
| 607 | resolve_or_fetch_page(&target, DEFAULT_OPEN_TIMEOUT_MS, context).await?; |
| 608 | click_counter += 1; |
| 609 | let click_ref = format!("{scope}turn{turn}click{click_counter}"); |
| 610 | store_page(&context.state_namespace, &click_ref, fetched.clone()); |
| 611 | let view = render_view(&click_ref, &fetched, 1, response_length); |
| 612 | views.push(view); |
| 613 | } |
| 614 | if !views.is_empty() { |
| 615 | output.click = Some(views); |
| 616 | } |
| 617 | } |
| 618 | |
| 619 | if let Some(find_requests) = input.get("find").and_then(|v| v.as_array()) { |
| 620 | let mut finds = Vec::new(); |
| 621 | for find_req in find_requests { |
| 622 | let ref_id = required_str(find_req, "ref_id")?.to_string(); |
| 623 | let pattern = required_str(find_req, "pattern")?.to_string(); |
| 624 | let page = get_page(&ref_id).ok_or_else(|| { |
| 625 | ToolError::invalid_input(format!("Unknown ref_id '{ref_id}'")) |
| 626 | })?; |
| 627 | let find_result = find_in_page(&ref_id, &pattern, &page, response_length); |
| 628 | finds.push(find_result); |
| 629 | } |
| 630 | if !finds.is_empty() { |
| 631 | output.find = Some(finds); |
| 632 | } |
| 633 | } |
| 634 | |
| 635 | if let Some(shots) = input.get("screenshot").and_then(|v| v.as_array()) { |
| 636 | let mut screenshots = Vec::new(); |
| 637 | for shot in shots { |
| 638 | let ref_id = required_str(shot, "ref_id")?.to_string(); |
| 639 | let pageno = optional_u64(shot, "pageno", 0) as usize; |
| 640 | let page = get_page(&ref_id).ok_or_else(|| { |
| 641 | ToolError::invalid_input(format!("Unknown ref_id '{ref_id}'")) |
| 642 | })?; |
| 643 | let screenshot = screenshot_page(&ref_id, pageno, &page)?; |
| 644 | screenshots.push(screenshot); |
| 645 | } |
| 646 | if !screenshots.is_empty() { |
| 647 | output.screenshot = Some(screenshots); |
| 648 | } |
| 649 | } |
| 650 | |
| 651 | ToolResult::json(&output).map_err(|e| ToolError::execution_failed(e.to_string())) |
| 652 | } |
| 653 | } |
| 654 | |
| 655 | fn with_state<T>(f: impl FnOnce(&mut WebRunState) -> T) -> T { |
| 656 | let lock = WEB_RUN_STATE.get_or_init(|| Mutex::new(WebRunState::default())); |
| 657 | let mut state = lock |
| 658 | .lock() |
| 659 | .expect("web run state mutex should not be poisoned"); |
| 660 | state.cleanup(); |
| 661 | f(&mut state) |
| 662 | } |
| 663 | |
| 664 | fn scoped_ref_prefix(namespace: &str) -> String { |
| 665 | let mut hasher = std::collections::hash_map::DefaultHasher::new(); |
| 666 | namespace.hash(&mut hasher); |
| 667 | format!("s{:016x}_", hasher.finish()) |
| 668 | } |
| 669 | |
| 670 | fn store_page(namespace: &str, ref_id: &str, page: WebPage) { |
| 671 | with_state(|state| { |
| 672 | state.store_page(namespace, ref_id, page); |
| 673 | }); |
| 674 | } |
| 675 | |
| 676 | fn get_page(ref_id: &str) -> Option<WebPage> { |
| 677 | with_state(|state| state.get_page(ref_id)) |
| 678 | } |
| 679 | |
| 680 | #[cfg(test)] |
| 681 | fn reset_web_run_state() { |
| 682 | with_state(|state| { |
| 683 | *state = WebRunState::default(); |
| 684 | }); |
| 685 | } |
| 686 | |
| 687 | #[cfg(test)] |
| 688 | fn next_turn_for_namespace(namespace: &str) -> u64 { |
| 689 | with_state(|state| state.next_turn(namespace)) |
| 690 | } |
| 691 | |
| 692 | async fn resolve_or_fetch_page( |
| 693 | ref_id: &str, |
| 694 | timeout_ms: u64, |
| 695 | context: &ToolContext, |
| 696 | ) -> Result<WebPage, ToolError> { |
| 697 | if let Some(page) = get_page(ref_id) { |
| 698 | return Ok(page); |
| 699 | } |
| 700 | if looks_like_url(ref_id) { |
| 701 | check_network_policy(ref_id, context)?; |
| 702 | return fetch_page(ref_id, timeout_ms).await; |
| 703 | } |
| 704 | Err(ToolError::invalid_input(format!( |
| 705 | "Unknown ref_id '{ref_id}'" |
| 706 | ))) |
| 707 | } |
| 708 | |
| 709 | fn looks_like_url(value: &str) -> bool { |
| 710 | value.starts_with("http://") || value.starts_with("https://") |
| 711 | } |
| 712 | |
| 713 | async fn run_search( |
| 714 | query: &str, |
| 715 | max_results: usize, |
| 716 | timeout_ms: u64, |
| 717 | domains: &[String], |
| 718 | ) -> Result<(Vec<SearchEntry>, String, Option<String>), ToolError> { |
| 719 | let client = reqwest::Client::builder() |
| 720 | .timeout(Duration::from_millis(timeout_ms)) |
| 721 | .user_agent(USER_AGENT) |
| 722 | .build() |
| 723 | .map_err(|e| ToolError::execution_failed(format!("Failed to build HTTP client: {e}")))?; |
| 724 | |
| 725 | let encoded = url_encode(query); |
| 726 | let url = format!("https://html.duckduckgo.com/html/?q={encoded}"); |
| 727 | let resp = client |
| 728 | .get(&url) |
| 729 | .header( |
| 730 | "Accept", |
| 731 | "text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8", |
| 732 | ) |
| 733 | .header("Accept-Language", "en-US,en;q=0.5") |
| 734 | .send() |
| 735 | .await |
| 736 | .map_err(|e| ToolError::execution_failed(format!("Web search request failed: {e}")))?; |
| 737 | |
| 738 | let status = resp.status(); |
| 739 | let body = resp |
| 740 | .text() |
| 741 | .await |
| 742 | .map_err(|e| ToolError::execution_failed(format!("Failed to read response: {e}")))?; |
| 743 | |
| 744 | if !status.is_success() { |
| 745 | return Err(ToolError::execution_failed(format!( |
| 746 | "Web search failed: HTTP {}", |
| 747 | status.as_u16() |
| 748 | ))); |
| 749 | } |
| 750 | |
| 751 | let mut results = parse_duckduckgo_results(&body, max_results); |
| 752 | let mut source = "duckduckgo".to_string(); |
| 753 | let mut warnings = Vec::new(); |
| 754 | |
| 755 | if results.is_empty() { |
| 756 | let duckduckgo_blocked = is_duckduckgo_challenge(&body); |
| 757 | match run_bing_search(&client, query, max_results).await { |
| 758 | Ok(fallback_results) if !fallback_results.is_empty() => { |
| 759 | results = fallback_results; |
| 760 | source = "bing".to_string(); |
| 761 | warnings.push(if duckduckgo_blocked { |
| 762 | "DuckDuckGo returned a bot challenge; used Bing fallback".to_string() |
| 763 | } else { |
| 764 | "DuckDuckGo returned no parseable results; used Bing fallback".to_string() |
| 765 | }); |
| 766 | } |
| 767 | Ok(_) if duckduckgo_blocked => { |
| 768 | return Err(ToolError::execution_failed( |
| 769 | "DuckDuckGo returned a bot challenge and Bing fallback returned no results", |
| 770 | )); |
| 771 | } |
| 772 | Err(err) if duckduckgo_blocked => { |
| 773 | return Err(ToolError::execution_failed(format!( |
| 774 | "DuckDuckGo returned a bot challenge and Bing fallback failed: {err}" |
| 775 | ))); |
| 776 | } |
| 777 | Ok(_) | Err(_) => {} |
| 778 | } |
| 779 | } |
| 780 | |
| 781 | if !domains.is_empty() { |
| 782 | let before = results.len(); |
| 783 | results.retain(|entry| domain_matches(&entry.url, domains)); |
| 784 | if before != results.len() { |
| 785 | warnings.push("Filtered search results by domain list".to_string()); |
| 786 | } |
| 787 | } |
| 788 | |
| 789 | Ok(( |
| 790 | results, |
| 791 | source, |
| 792 | if warnings.is_empty() { |
| 793 | None |
| 794 | } else { |
| 795 | Some(warnings.join("; ")) |
| 796 | }, |
| 797 | )) |
| 798 | } |
| 799 | |
| 800 | async fn run_bing_search( |
| 801 | client: &reqwest::Client, |
| 802 | query: &str, |
| 803 | max_results: usize, |
| 804 | ) -> Result<Vec<SearchEntry>, ToolError> { |
| 805 | let encoded = url_encode(query); |
| 806 | let url = format!("https://www.bing.com/search?q={encoded}"); |
| 807 | let resp = client |
| 808 | .get(&url) |
| 809 | .header( |
| 810 | "Accept", |
| 811 | "text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8", |
| 812 | ) |
| 813 | .header("Accept-Language", "en-US,en;q=0.9") |
| 814 | .send() |
| 815 | .await |
| 816 | .map_err(|e| ToolError::execution_failed(format!("Bing fallback request failed: {e}")))?; |
| 817 | |
| 818 | let status = resp.status(); |
| 819 | let body = resp.text().await.map_err(|e| { |
| 820 | ToolError::execution_failed(format!("Failed to read Bing fallback response: {e}")) |
| 821 | })?; |
| 822 | |
| 823 | if !status.is_success() { |
| 824 | return Err(ToolError::execution_failed(format!( |
| 825 | "Bing fallback failed: HTTP {}", |
| 826 | status.as_u16() |
| 827 | ))); |
| 828 | } |
| 829 | |
| 830 | Ok(parse_bing_results(&body, max_results)) |
| 831 | } |
| 832 | |
| 833 | fn domain_matches(url: &str, domains: &[String]) -> bool { |
| 834 | if domains.is_empty() { |
| 835 | return true; |
| 836 | } |
| 837 | let Ok(parsed) = reqwest::Url::parse(url) else { |
| 838 | return false; |
| 839 | }; |
| 840 | let Some(host) = parsed.host_str() else { |
| 841 | return false; |
| 842 | }; |
| 843 | domains.iter().any(|domain| { |
| 844 | let domain = domain.trim_start_matches("www."); |
| 845 | host == domain || host.ends_with(&format!(".{domain}")) |
| 846 | }) |
| 847 | } |
| 848 | |
| 849 | #[derive(Debug, Clone, Deserialize)] |
| 850 | struct DuckDuckGoImageResponse { |
| 851 | #[serde(default)] |
| 852 | results: Vec<DuckDuckGoImageResult>, |
| 853 | } |
| 854 | |
| 855 | #[derive(Debug, Clone, Deserialize)] |
| 856 | struct DuckDuckGoImageResult { |
| 857 | image: String, |
| 858 | #[serde(default)] |
| 859 | thumbnail: Option<String>, |
| 860 | #[serde(default)] |
| 861 | title: Option<String>, |
| 862 | #[serde(default)] |
| 863 | url: Option<String>, |
| 864 | #[serde(default)] |
| 865 | source: Option<String>, |
| 866 | #[serde(default)] |
| 867 | width: Option<u32>, |
| 868 | #[serde(default)] |
| 869 | height: Option<u32>, |
| 870 | } |
| 871 | |
| 872 | fn extract_duckduckgo_vqd(html: &str) -> Option<String> { |
| 873 | let html = html.trim(); |
| 874 | if html.is_empty() { |
| 875 | return None; |
| 876 | } |
| 877 | |
| 878 | for (prefix, suffix) in [("vqd='", "'"), ("vqd=\"", "\"")] { |
| 879 | if let Some(start) = html.find(prefix) { |
| 880 | let rest = &html[start + prefix.len()..]; |
| 881 | if let Some(end) = rest.find(suffix) { |
| 882 | let token = rest[..end].trim(); |
| 883 | if !token.is_empty() { |
| 884 | return Some(token.to_string()); |
| 885 | } |
| 886 | } |
| 887 | } |
| 888 | } |
| 889 | |
| 890 | // Fallback: look for `vqd=` and accept a conservative token charset. |
| 891 | if let Some(start) = html.find("vqd=") { |
| 892 | let rest = &html[start + 4..]; |
| 893 | let mut token = String::new(); |
| 894 | for ch in rest.chars() { |
| 895 | if ch.is_ascii_alphanumeric() || ch == '-' || ch == '_' { |
| 896 | token.push(ch); |
| 897 | } else { |
| 898 | break; |
| 899 | } |
| 900 | } |
| 901 | if !token.is_empty() { |
| 902 | return Some(token); |
| 903 | } |
| 904 | } |
| 905 | |
| 906 | None |
| 907 | } |
| 908 | |
| 909 | async fn run_image_search( |
| 910 | query: &str, |
| 911 | max_results: usize, |
| 912 | timeout_ms: u64, |
| 913 | domains: &[String], |
| 914 | ) -> Result<(Vec<ImageResultEntry>, Option<String>), ToolError> { |
| 915 | let client = reqwest::Client::builder() |
| 916 | .timeout(Duration::from_millis(timeout_ms)) |
| 917 | .user_agent(USER_AGENT) |
| 918 | .build() |
| 919 | .map_err(|e| ToolError::execution_failed(format!("Failed to build HTTP client: {e}")))?; |
| 920 | |
| 921 | // Step 1: fetch the HTML page to obtain the `vqd` token used by the images API. |
| 922 | let encoded = url_encode(query); |
| 923 | let seed_url = format!("https://duckduckgo.com/?q={encoded}&iax=images&ia=images"); |
| 924 | let seed_resp = client |
| 925 | .get(&seed_url) |
| 926 | .header( |
| 927 | "Accept", |
| 928 | "text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8", |
| 929 | ) |
| 930 | .header("Accept-Language", "en-US,en;q=0.5") |
| 931 | .send() |
| 932 | .await |
| 933 | .map_err(|e| { |
| 934 | ToolError::execution_failed(format!("Image search seed request failed: {e}")) |
| 935 | })?; |
| 936 | |
| 937 | let seed_status = seed_resp.status(); |
| 938 | let seed_body = seed_resp.text().await.map_err(|e| { |
| 939 | ToolError::execution_failed(format!("Failed to read image seed response: {e}")) |
| 940 | })?; |
| 941 | |
| 942 | if !seed_status.is_success() { |
| 943 | return Err(ToolError::execution_failed(format!( |
| 944 | "Image search seed request failed: HTTP {}", |
| 945 | seed_status.as_u16() |
| 946 | ))); |
| 947 | } |
| 948 | |
| 949 | let vqd = extract_duckduckgo_vqd(&seed_body).ok_or_else(|| { |
| 950 | ToolError::execution_failed("Failed to extract DuckDuckGo image token (vqd)") |
| 951 | })?; |
| 952 | |
| 953 | // Step 2: query the DuckDuckGo images JSON endpoint. |
| 954 | let api_url = format!("https://duckduckgo.com/i.js?l=us-en&o=json&q={encoded}&vqd={vqd}&p=1"); |
| 955 | let api_resp = client |
| 956 | .get(&api_url) |
| 957 | .header("Accept", "application/json") |
| 958 | .header("Referer", "https://duckduckgo.com/") |
| 959 | .send() |
| 960 | .await |
| 961 | .map_err(|e| ToolError::execution_failed(format!("Image search request failed: {e}")))?; |
| 962 | |
| 963 | let api_status = api_resp.status(); |
| 964 | let api_body = api_resp |
| 965 | .text() |
| 966 | .await |
| 967 | .map_err(|e| ToolError::execution_failed(format!("Failed to read image response: {e}")))?; |
| 968 | |
| 969 | if !api_status.is_success() { |
| 970 | return Err(ToolError::execution_failed(format!( |
| 971 | "Image search failed: HTTP {}", |
| 972 | api_status.as_u16() |
| 973 | ))); |
| 974 | } |
| 975 | |
| 976 | let parsed: DuckDuckGoImageResponse = serde_json::from_str(&api_body).map_err(|e| { |
| 977 | ToolError::execution_failed(format!("Failed to parse image search JSON: {e}")) |
| 978 | })?; |
| 979 | |
| 980 | let mut results = parsed |
| 981 | .results |
| 982 | .into_iter() |
| 983 | .filter(|item| !item.image.trim().is_empty()) |
| 984 | .map(|item| ImageResultEntry { |
| 985 | image: item.image, |
| 986 | thumbnail: item.thumbnail, |
| 987 | title: item.title, |
| 988 | url: item.url, |
| 989 | source: item.source, |
| 990 | width: item.width, |
| 991 | height: item.height, |
| 992 | }) |
| 993 | .collect::<Vec<_>>(); |
| 994 | |
| 995 | // Domain filter is applied to the source page URL when available. |
| 996 | let warning = if !domains.is_empty() { |
| 997 | let before = results.len(); |
| 998 | results.retain(|entry| match entry.url.as_deref() { |
| 999 | Some(url) => domain_matches(url, domains), |
| 1000 | None => true, |
| 1001 | }); |
| 1002 | if before != results.len() { |
| 1003 | Some("Filtered image results by domain list".to_string()) |
| 1004 | } else { |
| 1005 | None |
| 1006 | } |
| 1007 | } else { |
| 1008 | None |
| 1009 | }; |
| 1010 | |
| 1011 | results.truncate(max_results); |
| 1012 | Ok((results, warning)) |
| 1013 | } |
| 1014 | |
| 1015 | fn page_from_search(query: &str, results: &[SearchEntry]) -> WebPage { |
| 1016 | let mut lines = Vec::new(); |
| 1017 | let mut links = Vec::new(); |
| 1018 | |
| 1019 | lines.push(format!("Search results for: {query}")); |
| 1020 | for (idx, entry) in results.iter().enumerate() { |
| 1021 | let id = idx + 1; |
| 1022 | links.push(WebLink { |
| 1023 | id, |
| 1024 | url: entry.url.clone(), |
| 1025 | text: entry.title.clone(), |
| 1026 | }); |
| 1027 | lines.push(format!("{}. [{}] {}", id, id, entry.title)); |
| 1028 | if let Some(snippet) = entry.snippet.as_ref() |
| 1029 | && !snippet.trim().is_empty() |
| 1030 | { |
| 1031 | lines.push(format!(" {snippet}")); |
| 1032 | } |
| 1033 | lines.push(format!(" {url}", url = entry.url)); |
| 1034 | } |
| 1035 | |
| 1036 | WebPage { |
| 1037 | url: "https://html.duckduckgo.com/html/".to_string(), |
| 1038 | title: Some("Search Results".to_string()), |
| 1039 | content_type: Some("text/html".to_string()), |
| 1040 | lines, |
| 1041 | links, |
| 1042 | pdf_pages: None, |
| 1043 | } |
| 1044 | } |
| 1045 | |
| 1046 | /// Check network policy for a URL before fetching. |
| 1047 | /// Returns an error if the policy denies access. |
| 1048 | fn check_network_policy(url: &str, context: &ToolContext) -> Result<(), ToolError> { |
| 1049 | let Some(decider) = context.network_policy.as_ref() else { |
| 1050 | return Ok(()); |
| 1051 | }; |
| 1052 | let Some(host) = host_from_url(url) else { |
| 1053 | return Ok(()); |
| 1054 | }; |
| 1055 | match decider.evaluate(&host, "web_run") { |
| 1056 | Decision::Allow => Ok(()), |
| 1057 | Decision::Deny => Err(ToolError::permission_denied(format!( |
| 1058 | "network call to '{host}' blocked by network policy" |
| 1059 | ))), |
| 1060 | Decision::Prompt => Err(ToolError::permission_denied(format!( |
| 1061 | "network call to '{host}' requires approval; \ |
| 1062 | re-run after `/network allow {host}` or set network.default = \"allow\" in config" |
| 1063 | ))), |
| 1064 | } |
| 1065 | } |
| 1066 | |
| 1067 | async fn fetch_page(url: &str, timeout_ms: u64) -> Result<WebPage, ToolError> { |
| 1068 | let client = reqwest::Client::builder() |
| 1069 | .timeout(Duration::from_millis(timeout_ms)) |
| 1070 | .user_agent(USER_AGENT) |
| 1071 | .build() |
| 1072 | .map_err(|e| ToolError::execution_failed(format!("Failed to build HTTP client: {e}")))?; |
| 1073 | |
| 1074 | let resp = client |
| 1075 | .get(url) |
| 1076 | .header( |
| 1077 | "Accept", |
| 1078 | "text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8", |
| 1079 | ) |
| 1080 | .header("Accept-Language", "en-US,en;q=0.5") |
| 1081 | .send() |
| 1082 | .await |
| 1083 | .map_err(|e| ToolError::execution_failed(format!("Web request failed: {e}")))?; |
| 1084 | |
| 1085 | let status = resp.status(); |
| 1086 | let content_type = resp |
| 1087 | .headers() |
| 1088 | .get(reqwest::header::CONTENT_TYPE) |
| 1089 | .and_then(|v| v.to_str().ok()) |
| 1090 | .map(|s| s.to_string()); |
| 1091 | let bytes = resp |
| 1092 | .bytes() |
| 1093 | .await |
| 1094 | .map_err(|e| ToolError::execution_failed(format!("Failed to read response: {e}")))?; |
| 1095 | |
| 1096 | if !status.is_success() { |
| 1097 | return Err(ToolError::execution_failed(format!( |
| 1098 | "Web request failed: HTTP {}", |
| 1099 | status.as_u16() |
| 1100 | ))); |
| 1101 | } |
| 1102 | |
| 1103 | if is_pdf(&content_type, url) { |
| 1104 | return parse_pdf_page(url, content_type, &bytes); |
| 1105 | } |
| 1106 | |
| 1107 | let body = String::from_utf8_lossy(&bytes).to_string(); |
| 1108 | let (lines, links, title) = parse_html(&body, url); |
| 1109 | |
| 1110 | Ok(WebPage { |
| 1111 | url: url.to_string(), |
| 1112 | title, |
| 1113 | content_type, |
| 1114 | lines, |
| 1115 | links, |
| 1116 | pdf_pages: None, |
| 1117 | }) |
| 1118 | } |
| 1119 | |
| 1120 | fn is_pdf(content_type: &Option<String>, url: &str) -> bool { |
| 1121 | if let Some(ct) = content_type |
| 1122 | && ct.to_lowercase().contains("application/pdf") |
| 1123 | { |
| 1124 | return true; |
| 1125 | } |
| 1126 | url.to_lowercase().ends_with(".pdf") |
| 1127 | } |
| 1128 | |
| 1129 | fn parse_pdf_page( |
| 1130 | url: &str, |
| 1131 | content_type: Option<String>, |
| 1132 | bytes: &[u8], |
| 1133 | ) -> Result<WebPage, ToolError> { |
| 1134 | let text = pdf_extract_text(bytes)?; |
| 1135 | let pages = split_pdf_pages(&text); |
| 1136 | let lines = pages.first().cloned().unwrap_or_default(); |
| 1137 | |
| 1138 | Ok(WebPage { |
| 1139 | url: url.to_string(), |
| 1140 | title: Some("PDF Document".to_string()), |
| 1141 | content_type, |
| 1142 | lines, |
| 1143 | links: Vec::new(), |
| 1144 | pdf_pages: Some(pages), |
| 1145 | }) |
| 1146 | } |
| 1147 | |
| 1148 | fn pdf_extract_text(bytes: &[u8]) -> Result<String, ToolError> { |
| 1149 | pdf_extract::extract_text_from_mem(bytes) |
| 1150 | .map_err(|e| ToolError::execution_failed(format!("PDF extract failed: {e}"))) |
| 1151 | } |
| 1152 | |
| 1153 | fn split_pdf_pages(text: &str) -> Vec<Vec<String>> { |
| 1154 | let raw_pages: Vec<&str> = text.split('\x0C').collect(); |
| 1155 | raw_pages |
| 1156 | .iter() |
| 1157 | .map(|page| { |
| 1158 | page.lines() |
| 1159 | .map(|line| line.trim()) |
| 1160 | .filter(|line| !line.is_empty()) |
| 1161 | .map(|line| line.to_string()) |
| 1162 | .collect::<Vec<_>>() |
| 1163 | }) |
| 1164 | .collect() |
| 1165 | } |
| 1166 | |
| 1167 | fn render_view( |
| 1168 | ref_id: &str, |
| 1169 | page: &WebPage, |
| 1170 | lineno: usize, |
| 1171 | response: ResponseLength, |
| 1172 | ) -> PageViewResult { |
| 1173 | let total = page.lines.len(); |
| 1174 | let view_lines = response.view_lines(); |
| 1175 | let start = if total == 0 { |
| 1176 | 1 |
| 1177 | } else if lineno > total { |
| 1178 | total.saturating_sub(view_lines.saturating_sub(1)).max(1) |
| 1179 | } else { |
| 1180 | lineno |
| 1181 | }; |
| 1182 | let end = if total == 0 { |
| 1183 | 0 |
| 1184 | } else { |
| 1185 | (start + view_lines - 1).min(total) |
| 1186 | }; |
| 1187 | |
| 1188 | let content = if total == 0 { |
| 1189 | "(no content)".to_string() |
| 1190 | } else { |
| 1191 | render_lines(&page.lines, start, end) |
| 1192 | }; |
| 1193 | |
| 1194 | PageViewResult { |
| 1195 | ref_id: ref_id.to_string(), |
| 1196 | url: page.url.clone(), |
| 1197 | title: page.title.clone(), |
| 1198 | content_type: page.content_type.clone(), |
| 1199 | line_start: start, |
| 1200 | line_end: end, |
| 1201 | total_lines: total, |
| 1202 | content, |
| 1203 | links: page.links.clone(), |
| 1204 | } |
| 1205 | } |
| 1206 | |
| 1207 | fn render_lines(lines: &[String], start: usize, end: usize) -> String { |
| 1208 | lines |
| 1209 | .iter() |
| 1210 | .enumerate() |
| 1211 | .filter_map(|(idx, line)| { |
| 1212 | let line_no = idx + 1; |
| 1213 | if line_no < start || line_no > end { |
| 1214 | return None; |
| 1215 | } |
| 1216 | Some(format!("{:>4} {}", line_no, line)) |
| 1217 | }) |
| 1218 | .collect::<Vec<_>>() |
| 1219 | .join("\n") |
| 1220 | } |
| 1221 | |
| 1222 | fn find_in_page( |
| 1223 | ref_id: &str, |
| 1224 | pattern: &str, |
| 1225 | page: &WebPage, |
| 1226 | response: ResponseLength, |
| 1227 | ) -> FindResult { |
| 1228 | let needle = pattern.to_lowercase(); |
| 1229 | let mut matches = Vec::new(); |
| 1230 | for (idx, line) in page.lines.iter().enumerate() { |
| 1231 | if line.to_lowercase().contains(&needle) { |
| 1232 | matches.push(FindMatch { |
| 1233 | line: idx + 1, |
| 1234 | text: line.clone(), |
| 1235 | }); |
| 1236 | } |
| 1237 | if matches.len() >= response.max_find_matches() { |
| 1238 | break; |
| 1239 | } |
| 1240 | } |
| 1241 | |
| 1242 | FindResult { |
| 1243 | ref_id: ref_id.to_string(), |
| 1244 | pattern: pattern.to_string(), |
| 1245 | count: matches.len(), |
| 1246 | matches, |
| 1247 | } |
| 1248 | } |
| 1249 | |
| 1250 | fn screenshot_page( |
| 1251 | ref_id: &str, |
| 1252 | pageno: usize, |
| 1253 | page: &WebPage, |
| 1254 | ) -> Result<ScreenshotResult, ToolError> { |
| 1255 | let pages = page |
| 1256 | .pdf_pages |
| 1257 | .as_ref() |
| 1258 | .ok_or_else(|| ToolError::invalid_input("screenshot is only supported for PDF pages"))?; |
| 1259 | if pages.is_empty() { |
| 1260 | return Err(ToolError::execution_failed("PDF has no pages")); |
| 1261 | } |
| 1262 | if pageno >= pages.len() { |
| 1263 | return Err(ToolError::invalid_input(format!( |
| 1264 | "pageno {pageno} out of range (0..{max})", |
| 1265 | max = pages.len().saturating_sub(1) |
| 1266 | ))); |
| 1267 | } |
| 1268 | let content = pages[pageno].join("\n"); |
| 1269 | Ok(ScreenshotResult { |
| 1270 | ref_id: ref_id.to_string(), |
| 1271 | pageno, |
| 1272 | total_pages: pages.len(), |
| 1273 | content, |
| 1274 | }) |
| 1275 | } |
| 1276 | |
| 1277 | // === HTML Parsing === |
| 1278 | |
| 1279 | static ANCHOR_RE: OnceLock<Regex> = OnceLock::new(); |
| 1280 | static TAG_RE: OnceLock<Regex> = OnceLock::new(); |
| 1281 | static BLOCK_RE: OnceLock<Regex> = OnceLock::new(); |
| 1282 | static SCRIPT_RE: OnceLock<Regex> = OnceLock::new(); |
| 1283 | static STYLE_RE: OnceLock<Regex> = OnceLock::new(); |
| 1284 | static TITLE_RE: OnceLock<Regex> = OnceLock::new(); |
| 1285 | static SNIPPET_RE: OnceLock<Regex> = OnceLock::new(); |
| 1286 | static SEARCH_TITLE_RE: OnceLock<Regex> = OnceLock::new(); |
| 1287 | static BING_RESULT_RE: OnceLock<Regex> = OnceLock::new(); |
| 1288 | static BING_TITLE_RE: OnceLock<Regex> = OnceLock::new(); |
| 1289 | static BING_SNIPPET_RE: OnceLock<Regex> = OnceLock::new(); |
| 1290 | |
| 1291 | fn get_anchor_re() -> &'static Regex { |
| 1292 | ANCHOR_RE.get_or_init(|| { |
| 1293 | Regex::new(r#"(?is)<a\s+[^>]*href\s*=\s*['\"]([^'\"]+)['\"][^>]*>(.*?)</a>"#) |
| 1294 | .expect("anchor regex") |
| 1295 | }) |
| 1296 | } |
| 1297 | |
| 1298 | fn get_tag_re() -> &'static Regex { |
| 1299 | TAG_RE.get_or_init(|| Regex::new(r"<[^>]+>").expect("tag regex")) |
| 1300 | } |
| 1301 | |
| 1302 | fn get_block_re() -> &'static Regex { |
| 1303 | BLOCK_RE.get_or_init(|| { |
| 1304 | Regex::new(r"(?is)</?(p|div|li|ul|ol|br|h[1-6]|tr|td|th|table|section|article)[^>]*>") |
| 1305 | .expect("block regex") |
| 1306 | }) |
| 1307 | } |
| 1308 | |
| 1309 | fn get_script_re() -> &'static Regex { |
| 1310 | SCRIPT_RE.get_or_init(|| Regex::new(r"(?is)<script[^>]*>.*?</script>").unwrap()) |
| 1311 | } |
| 1312 | |
| 1313 | fn get_style_re() -> &'static Regex { |
| 1314 | STYLE_RE.get_or_init(|| Regex::new(r"(?is)<style[^>]*>.*?</style>").unwrap()) |
| 1315 | } |
| 1316 | |
| 1317 | fn get_title_re() -> &'static Regex { |
| 1318 | TITLE_RE.get_or_init(|| Regex::new(r"(?is)<title[^>]*>(.*?)</title>").unwrap()) |
| 1319 | } |
| 1320 | |
| 1321 | fn get_search_title_re() -> &'static Regex { |
| 1322 | SEARCH_TITLE_RE.get_or_init(|| { |
| 1323 | Regex::new(r#"<a[^>]*class=\"result__a\"[^>]*href=\"([^\"]+)\"[^>]*>(.*?)</a>"#) |
| 1324 | .expect("title regex pattern is valid") |
| 1325 | }) |
| 1326 | } |
| 1327 | |
| 1328 | fn get_search_snippet_re() -> &'static Regex { |
| 1329 | SNIPPET_RE.get_or_init(|| { |
| 1330 | Regex::new( |
| 1331 | r#"<a[^>]*class=\"result__snippet\"[^>]*>(.*?)</a>|<div[^>]*class=\"result__snippet\"[^>]*>(.*?)</div>"#, |
| 1332 | ) |
| 1333 | .expect("snippet regex pattern is valid") |
| 1334 | }) |
| 1335 | } |
| 1336 | |
| 1337 | fn get_bing_result_re() -> &'static Regex { |
| 1338 | BING_RESULT_RE.get_or_init(|| { |
| 1339 | Regex::new(r#"(?is)<li[^>]*class=\"[^\"]*\bb_algo\b[^\"]*\"[^>]*>(.*?)</li>"#) |
| 1340 | .expect("bing result regex pattern is valid") |
| 1341 | }) |
| 1342 | } |
| 1343 | |
| 1344 | fn get_bing_title_re() -> &'static Regex { |
| 1345 | BING_TITLE_RE.get_or_init(|| { |
| 1346 | Regex::new(r#"(?is)<h2[^>]*>.*?<a[^>]*href=\"([^\"]+)\"[^>]*>(.*?)</a>"#) |
| 1347 | .expect("bing title regex pattern is valid") |
| 1348 | }) |
| 1349 | } |
| 1350 | |
| 1351 | fn get_bing_snippet_re() -> &'static Regex { |
| 1352 | BING_SNIPPET_RE.get_or_init(|| { |
| 1353 | Regex::new(r#"(?is)<div[^>]*class=\"[^\"]*\bb_caption\b[^\"]*\"[^>]*>.*?<p[^>]*>(.*?)</p>"#) |
| 1354 | .expect("bing snippet regex pattern is valid") |
| 1355 | }) |
| 1356 | } |
| 1357 | |
| 1358 | fn parse_html(html: &str, base_url: &str) -> (Vec<String>, Vec<WebLink>, Option<String>) { |
| 1359 | let title = extract_title(html); |
| 1360 | let without_scripts = get_script_re().replace_all(html, "").to_string(); |
| 1361 | let without_styles = get_style_re().replace_all(&without_scripts, "").to_string(); |
| 1362 | |
| 1363 | let (with_links, links) = replace_links(&without_styles, base_url); |
| 1364 | let with_breaks = get_block_re().replace_all(&with_links, "\n").to_string(); |
| 1365 | let without_tags = get_tag_re().replace_all(&with_breaks, "").to_string(); |
| 1366 | let decoded = decode_html_entities(&without_tags); |
| 1367 | |
| 1368 | let mut lines = Vec::new(); |
| 1369 | for line in decoded.lines() { |
| 1370 | let trimmed = normalize_whitespace(line); |
| 1371 | if trimmed.is_empty() { |
| 1372 | continue; |
| 1373 | } |
| 1374 | for wrapped in wrap_line(&trimmed, ResponseLength::Medium.wrap_width()) { |
| 1375 | lines.push(wrapped); |
| 1376 | } |
| 1377 | } |
| 1378 | |
| 1379 | (lines, links, title) |
| 1380 | } |
| 1381 | |
| 1382 | fn extract_title(html: &str) -> Option<String> { |
| 1383 | let re = get_title_re(); |
| 1384 | let cap = re.captures(html)?; |
| 1385 | let raw = cap.get(1)?.as_str(); |
| 1386 | let cleaned = normalize_whitespace(&decode_html_entities(raw)); |
| 1387 | if cleaned.is_empty() { |
| 1388 | None |
| 1389 | } else { |
| 1390 | Some(cleaned) |
| 1391 | } |
| 1392 | } |
| 1393 | |
| 1394 | fn replace_links(html: &str, base_url: &str) -> (String, Vec<WebLink>) { |
| 1395 | let re = get_anchor_re(); |
| 1396 | let mut links = Vec::new(); |
| 1397 | let mut output = String::with_capacity(html.len()); |
| 1398 | let mut last = 0; |
| 1399 | |
| 1400 | for cap in re.captures_iter(html) { |
| 1401 | let Some(full) = cap.get(0) else { continue }; |
| 1402 | let Some(href) = cap.get(1) else { continue }; |
| 1403 | let Some(text_match) = cap.get(2) else { |
| 1404 | continue; |
| 1405 | }; |
| 1406 | |
| 1407 | output.push_str(&html[last..full.start()]); |
| 1408 | let text = normalize_whitespace(&strip_tags(text_match.as_str())); |
| 1409 | let resolved = resolve_url(base_url, href.as_str()); |
| 1410 | if !text.is_empty() { |
| 1411 | let id = links.len() + 1; |
| 1412 | links.push(WebLink { |
| 1413 | id, |
| 1414 | url: resolved.clone(), |
| 1415 | text: text.clone(), |
| 1416 | }); |
| 1417 | output.push_str(&format!("[{}] {}", id, text)); |
| 1418 | } else { |
| 1419 | output.push_str(&resolved); |
| 1420 | } |
| 1421 | last = full.end(); |
| 1422 | } |
| 1423 | |
| 1424 | output.push_str(&html[last..]); |
| 1425 | (output, links) |
| 1426 | } |
| 1427 | |
| 1428 | fn resolve_url(base: &str, href: &str) -> String { |
| 1429 | if href.starts_with("http://") || href.starts_with("https://") { |
| 1430 | return href.to_string(); |
| 1431 | } |
| 1432 | if href.starts_with("//") { |
| 1433 | return format!("https:{href}"); |
| 1434 | } |
| 1435 | if let Ok(base_url) = reqwest::Url::parse(base) |
| 1436 | && let Ok(joined) = base_url.join(href) |
| 1437 | { |
| 1438 | return joined.to_string(); |
| 1439 | } |
| 1440 | href.to_string() |
| 1441 | } |
| 1442 | |
| 1443 | fn strip_tags(text: &str) -> String { |
| 1444 | get_tag_re().replace_all(text, "").to_string() |
| 1445 | } |
| 1446 | |
| 1447 | fn normalize_whitespace(text: &str) -> String { |
| 1448 | text.split_whitespace().collect::<Vec<_>>().join(" ") |
| 1449 | } |
| 1450 | |
| 1451 | fn wrap_line(text: &str, width: usize) -> Vec<String> { |
| 1452 | if text.len() <= width { |
| 1453 | return vec![text.to_string()]; |
| 1454 | } |
| 1455 | let mut lines = Vec::new(); |
| 1456 | let mut current = String::new(); |
| 1457 | for word in text.split_whitespace() { |
| 1458 | if current.is_empty() { |
| 1459 | current.push_str(word); |
| 1460 | } else if current.len() + word.len() < width { |
| 1461 | current.push(' '); |
| 1462 | current.push_str(word); |
| 1463 | } else { |
| 1464 | lines.push(current); |
| 1465 | current = word.to_string(); |
| 1466 | } |
| 1467 | } |
| 1468 | if !current.is_empty() { |
| 1469 | lines.push(current); |
| 1470 | } |
| 1471 | lines |
| 1472 | } |
| 1473 | |
| 1474 | fn decode_html_entities(text: &str) -> String { |
| 1475 | text.replace("&", "&") |
| 1476 | .replace(""", "\"") |
| 1477 | .replace("'", "'") |
| 1478 | .replace("'", "'") |
| 1479 | .replace("<", "<") |
| 1480 | .replace(">", ">") |
| 1481 | .replace(" ", " ") |
| 1482 | } |
| 1483 | |
| 1484 | fn parse_duckduckgo_results(html: &str, max_results: usize) -> Vec<SearchEntry> { |
| 1485 | let title_re = get_search_title_re(); |
| 1486 | let snippet_re = get_search_snippet_re(); |
| 1487 | let snippets: Vec<String> = snippet_re |
| 1488 | .captures_iter(html) |
| 1489 | .filter_map(|cap| cap.get(1).or_else(|| cap.get(2))) |
| 1490 | .map(|m| normalize_whitespace(&decode_html_entities(&strip_tags(m.as_str())))) |
| 1491 | .collect(); |
| 1492 | |
| 1493 | let mut results = Vec::new(); |
| 1494 | for (idx, cap) in title_re.captures_iter(html).enumerate() { |
| 1495 | if results.len() >= max_results { |
| 1496 | break; |
| 1497 | } |
| 1498 | let href = cap.get(1).map(|m| m.as_str()).unwrap_or(""); |
| 1499 | let title_raw = cap.get(2).map(|m| m.as_str()).unwrap_or(""); |
| 1500 | let title = normalize_whitespace(&decode_html_entities(&strip_tags(title_raw))); |
| 1501 | if title.is_empty() { |
| 1502 | continue; |
| 1503 | } |
| 1504 | let url = normalize_search_url(href); |
| 1505 | let snippet = snippets |
| 1506 | .get(idx) |
| 1507 | .map(|s| s.to_string()) |
| 1508 | .filter(|s| !s.is_empty()); |
| 1509 | |
| 1510 | results.push(SearchEntry { |
| 1511 | title, |
| 1512 | url, |
| 1513 | snippet, |
| 1514 | }); |
| 1515 | } |
| 1516 | |
| 1517 | results |
| 1518 | } |
| 1519 | |
| 1520 | fn is_duckduckgo_challenge(html: &str) -> bool { |
| 1521 | html.contains("anomaly-modal") || html.contains("Unfortunately, bots use DuckDuckGo too") |
| 1522 | } |
| 1523 | |
| 1524 | fn parse_bing_results(html: &str, max_results: usize) -> Vec<SearchEntry> { |
| 1525 | let mut results = Vec::new(); |
| 1526 | for cap in get_bing_result_re().captures_iter(html) { |
| 1527 | if results.len() >= max_results { |
| 1528 | break; |
| 1529 | } |
| 1530 | let Some(block) = cap.get(1).map(|m| m.as_str()) else { |
| 1531 | continue; |
| 1532 | }; |
| 1533 | let Some(title_cap) = get_bing_title_re().captures(block) else { |
| 1534 | continue; |
| 1535 | }; |
| 1536 | let href = title_cap.get(1).map(|m| m.as_str()).unwrap_or(""); |
| 1537 | let title_raw = title_cap.get(2).map(|m| m.as_str()).unwrap_or(""); |
| 1538 | let title = normalize_whitespace(&decode_html_entities(&strip_tags(title_raw))); |
| 1539 | if title.is_empty() { |
| 1540 | continue; |
| 1541 | } |
| 1542 | let snippet = get_bing_snippet_re() |
| 1543 | .captures(block) |
| 1544 | .and_then(|snippet_cap| snippet_cap.get(1)) |
| 1545 | .map(|m| normalize_whitespace(&decode_html_entities(&strip_tags(m.as_str())))) |
| 1546 | .filter(|s| !s.is_empty()); |
| 1547 | |
| 1548 | results.push(SearchEntry { |
| 1549 | title, |
| 1550 | url: normalize_bing_url(href), |
| 1551 | snippet, |
| 1552 | }); |
| 1553 | } |
| 1554 | |
| 1555 | results |
| 1556 | } |
| 1557 | |
| 1558 | fn normalize_search_url(href: &str) -> String { |
| 1559 | if let Some(uddg) = extract_query_param(href, "uddg") { |
| 1560 | let decoded = percent_decode(&uddg); |
| 1561 | if !decoded.is_empty() { |
| 1562 | return decoded; |
| 1563 | } |
| 1564 | } |
| 1565 | if href.starts_with("//") { |
| 1566 | return format!("https:{href}"); |
| 1567 | } |
| 1568 | if href.starts_with('/') { |
| 1569 | return format!("https://duckduckgo.com{href}"); |
| 1570 | } |
| 1571 | href.to_string() |
| 1572 | } |
| 1573 | |
| 1574 | fn normalize_bing_url(href: &str) -> String { |
| 1575 | if let Some(encoded) = extract_query_param(href, "u") { |
| 1576 | let decoded = percent_decode(&encoded); |
| 1577 | let token = decoded.strip_prefix("a1").unwrap_or(&decoded); |
| 1578 | let mut padded = token.replace('-', "+").replace('_', "/"); |
| 1579 | while !padded.len().is_multiple_of(4) { |
| 1580 | padded.push('='); |
| 1581 | } |
| 1582 | if let Ok(bytes) = general_purpose::STANDARD.decode(padded) |
| 1583 | && let Ok(url) = String::from_utf8(bytes) |
| 1584 | && looks_like_url(&url) |
| 1585 | { |
| 1586 | return url; |
| 1587 | } |
| 1588 | } |
| 1589 | if href.starts_with("//") { |
| 1590 | return format!("https:{href}"); |
| 1591 | } |
| 1592 | if href.starts_with('/') { |
| 1593 | return format!("https://www.bing.com{href}"); |
| 1594 | } |
| 1595 | href.to_string() |
| 1596 | } |
| 1597 | |
| 1598 | fn extract_query_param(url: &str, key: &str) -> Option<String> { |
| 1599 | let query_start = url.find('?')?; |
| 1600 | let query = &url[query_start + 1..]; |
| 1601 | for part in query.split('&') { |
| 1602 | let (k, v) = part.split_once('=')?; |
| 1603 | if k == key { |
| 1604 | return Some(v.to_string()); |
| 1605 | } |
| 1606 | } |
| 1607 | None |
| 1608 | } |
| 1609 | |
| 1610 | fn percent_decode(input: &str) -> String { |
| 1611 | let mut out = Vec::with_capacity(input.len()); |
| 1612 | let bytes = input.as_bytes(); |
| 1613 | let mut idx = 0; |
| 1614 | while idx < bytes.len() { |
| 1615 | if bytes[idx] == b'%' |
| 1616 | && idx + 2 < bytes.len() |
| 1617 | && let Ok(hex) = std::str::from_utf8(&bytes[idx + 1..idx + 3]) |
| 1618 | && let Ok(val) = u8::from_str_radix(hex, 16) |
| 1619 | { |
| 1620 | out.push(val); |
| 1621 | idx += 3; |
| 1622 | continue; |
| 1623 | } |
| 1624 | out.push(bytes[idx]); |
| 1625 | idx += 1; |
| 1626 | } |
| 1627 | String::from_utf8_lossy(&out).into_owned() |
| 1628 | } |
| 1629 | |
| 1630 | fn url_encode(input: &str) -> String { |
| 1631 | crate::utils::url_encode(input) |
| 1632 | } |
| 1633 | |
| 1634 | // === Tests === |
| 1635 | |
| 1636 | #[cfg(test)] |
| 1637 | mod tests { |
| 1638 | use super::*; |
| 1639 | use std::path::PathBuf; |
| 1640 | |
| 1641 | fn sample_page(url: &str) -> WebPage { |
| 1642 | WebPage { |
| 1643 | url: url.to_string(), |
| 1644 | title: Some("Example".to_string()), |
| 1645 | content_type: Some("text/html".to_string()), |
| 1646 | lines: vec!["example line".to_string()], |
| 1647 | links: Vec::new(), |
| 1648 | pdf_pages: None, |
| 1649 | } |
| 1650 | } |
| 1651 | |
| 1652 | #[test] |
| 1653 | fn html_link_parsing_extracts_links() { |
| 1654 | let html = r#" |
| 1655 | <html><body> |
| 1656 | <p>Hello <a href="https://example.com">Example</a> world.</p> |
| 1657 | </body></html> |
| 1658 | "#; |
| 1659 | let (lines, links, title) = parse_html(html, "https://example.com"); |
| 1660 | assert!(title.is_none()); |
| 1661 | assert_eq!(links.len(), 1); |
| 1662 | assert_eq!(links[0].url, "https://example.com"); |
| 1663 | assert!(lines.iter().any(|line| line.contains("Example"))); |
| 1664 | } |
| 1665 | |
| 1666 | #[test] |
| 1667 | fn wrap_line_splits_long_lines() { |
| 1668 | let line = "This is a long line that should wrap cleanly at word boundaries"; |
| 1669 | let wrapped = wrap_line(line, 20); |
| 1670 | assert!(wrapped.len() > 1); |
| 1671 | assert!(wrapped.iter().all(|l| l.len() <= 20)); |
| 1672 | } |
| 1673 | |
| 1674 | #[test] |
| 1675 | fn extracts_duckduckgo_vqd_token() { |
| 1676 | let html_single = "<script>var x = {vqd='3-1234567890'};</script>"; |
| 1677 | assert_eq!( |
| 1678 | extract_duckduckgo_vqd(html_single), |
| 1679 | Some("3-1234567890".to_string()) |
| 1680 | ); |
| 1681 | |
| 1682 | let html_double = "<script>var x = {vqd=\"3-abcdef\"};</script>"; |
| 1683 | assert_eq!( |
| 1684 | extract_duckduckgo_vqd(html_double), |
| 1685 | Some("3-abcdef".to_string()) |
| 1686 | ); |
| 1687 | |
| 1688 | let html_plain = "https://duckduckgo.com/?q=test&vqd=3-xyz_123&ia=images"; |
| 1689 | assert_eq!( |
| 1690 | extract_duckduckgo_vqd(html_plain), |
| 1691 | Some("3-xyz_123".to_string()) |
| 1692 | ); |
| 1693 | } |
| 1694 | |
| 1695 | #[test] |
| 1696 | fn parses_bing_results_and_decodes_redirect_url() { |
| 1697 | let html = r#" |
| 1698 | <ol> |
| 1699 | <li class="b_algo"> |
| 1700 | <h2><a href="https://www.bing.com/ck/a?u=a1aHR0cHM6Ly9leGFtcGxlLmNvbS9wYXRoP3E9MQ">Example & Result</a></h2> |
| 1701 | <div class="b_caption"><p>A <strong>useful</strong> snippet.</p></div> |
| 1702 | </li> |
| 1703 | </ol> |
| 1704 | "#; |
| 1705 | |
| 1706 | let results = parse_bing_results(html, 5); |
| 1707 | |
| 1708 | assert_eq!(results.len(), 1); |
| 1709 | assert_eq!(results[0].title, "Example & Result"); |
| 1710 | assert_eq!(results[0].url, "https://example.com/path?q=1"); |
| 1711 | assert_eq!(results[0].snippet.as_deref(), Some("A useful snippet.")); |
| 1712 | } |
| 1713 | |
| 1714 | #[test] |
| 1715 | fn percent_decode_handles_utf8_multibyte_sequences() { |
| 1716 | // Percent-encoded CJK: %E4%B8%AA%E4%BA%BA = 个人 (each glyph is 3 UTF-8 bytes). |
| 1717 | assert_eq!(percent_decode("Hello %E4%B8%AA%E4%BA%BA"), "Hello 个人"); |
| 1718 | assert_eq!(percent_decode("%E7%B4%A0%E6%9D%90"), "素材"); |
| 1719 | // Percent-encoded UTF-8 inside a URL path (DuckDuckGo `uddg=` redirect shape). |
| 1720 | assert_eq!( |
| 1721 | percent_decode("https://example.com/%E9%A1%B5%E9%9D%A2"), |
| 1722 | "https://example.com/页面" |
| 1723 | ); |
| 1724 | // Raw UTF-8 in the input passes through unchanged. |
| 1725 | assert_eq!(percent_decode("查询 keyword"), "查询 keyword"); |
| 1726 | // ASCII-only inputs preserve existing behavior; `+` stays literal. |
| 1727 | assert_eq!(percent_decode("foo+bar%20baz"), "foo+bar baz"); |
| 1728 | } |
| 1729 | |
| 1730 | #[test] |
| 1731 | fn scoped_ref_prefix_is_session_specific() { |
| 1732 | reset_web_run_state(); |
| 1733 | let alpha = scoped_ref_prefix("session-alpha"); |
| 1734 | let beta = scoped_ref_prefix("session-beta"); |
| 1735 | |
| 1736 | assert_ne!(alpha, beta); |
| 1737 | assert!(alpha.starts_with('s')); |
| 1738 | assert!(alpha.ends_with('_')); |
| 1739 | assert_eq!(alpha.len(), 18); |
| 1740 | } |
| 1741 | |
| 1742 | #[test] |
| 1743 | fn stored_pages_do_not_cross_scoped_sessions() { |
| 1744 | reset_web_run_state(); |
| 1745 | let shared_suffix = "turn1search1"; |
| 1746 | let ref_alpha = format!("{}{}", scoped_ref_prefix("session-alpha"), shared_suffix); |
| 1747 | let ref_beta = format!("{}{}", scoped_ref_prefix("session-beta"), shared_suffix); |
| 1748 | |
| 1749 | store_page( |
| 1750 | "session-alpha", |
| 1751 | &ref_alpha, |
| 1752 | sample_page("https://example.com/alpha"), |
| 1753 | ); |
| 1754 | |
| 1755 | assert!(get_page(&ref_alpha).is_some()); |
| 1756 | assert!(get_page(&ref_beta).is_none()); |
| 1757 | } |
| 1758 | |
| 1759 | #[test] |
| 1760 | fn turn_counters_are_scoped_per_session() { |
| 1761 | reset_web_run_state(); |
| 1762 | |
| 1763 | assert_eq!(next_turn_for_namespace("session-alpha"), 0); |
| 1764 | assert_eq!(next_turn_for_namespace("session-alpha"), 1); |
| 1765 | assert_eq!(next_turn_for_namespace("session-beta"), 0); |
| 1766 | } |
| 1767 | |
| 1768 | #[test] |
| 1769 | fn stale_session_pages_are_evicted() { |
| 1770 | reset_web_run_state(); |
| 1771 | let namespace = "session-alpha"; |
| 1772 | let ref_id = format!("{}turn0search1", scoped_ref_prefix(namespace)); |
| 1773 | store_page(namespace, &ref_id, sample_page("https://example.com/alpha")); |
| 1774 | |
| 1775 | // On Windows, Instant's epoch is system boot. If the CI runner has |
| 1776 | // been up for less than WEB_RUN_SESSION_TTL the subtraction would |
| 1777 | // underflow, so we skip the test in that case. |
| 1778 | let stale = WEB_RUN_SESSION_TTL + Duration::from_secs(1); |
| 1779 | let can_test = with_state(|state| { |
| 1780 | let session = state |
| 1781 | .sessions |
| 1782 | .get_mut(namespace) |
| 1783 | .expect("session should exist"); |
| 1784 | match Instant::now().checked_sub(stale) { |
| 1785 | Some(past) => { |
| 1786 | session.last_access = past; |
| 1787 | true |
| 1788 | } |
| 1789 | None => false, |
| 1790 | } |
| 1791 | }); |
| 1792 | if !can_test { |
| 1793 | // System uptime shorter than session TTL; can't test eviction. |
| 1794 | return; |
| 1795 | } |
| 1796 | |
| 1797 | let _ = next_turn_for_namespace("session-beta"); |
| 1798 | |
| 1799 | assert!(get_page(&ref_id).is_none()); |
| 1800 | } |
| 1801 | |
| 1802 | #[test] |
| 1803 | fn direct_urls_remain_compatible_open_refs() { |
| 1804 | assert!(looks_like_url("https://example.com")); |
| 1805 | assert!(looks_like_url("http://example.com")); |
| 1806 | assert!(!looks_like_url("turn0search0")); |
| 1807 | } |
| 1808 | |
| 1809 | #[test] |
| 1810 | fn network_policy_denies_direct_open_url() { |
| 1811 | use crate::network_policy::{Decision, NetworkPolicy, NetworkPolicyDecider}; |
| 1812 | |
| 1813 | let policy = NetworkPolicy { |
| 1814 | default: Decision::Deny.into(), |
| 1815 | allow: vec!["api.deepseek.com".to_string()], |
| 1816 | deny: vec![], |
| 1817 | audit: false, |
| 1818 | }; |
| 1819 | let decider = NetworkPolicyDecider::new(policy, None); |
| 1820 | let ctx = ToolContext::new(PathBuf::from(".")).with_network_policy(decider); |
| 1821 | |
| 1822 | let err = check_network_policy("https://example.com/private", &ctx) |
| 1823 | .expect_err("blocked host should fail"); |
| 1824 | assert!(format!("{err}").contains("blocked by network policy")); |
| 1825 | } |
| 1826 | } |
| 1827 |