YC (S26) | Open Computer History | Continuously record your company computer work, map your workflows, help you find work worth automating, and power your agents' context
Rust agentsagiaiai-memoryaudio-recordingcomputer-visionhermeshermes-agentllmlocal-ailocal-firstmachine-learningmcpmultimodalopenclawprivacyrewindscreen-recordingspeech-to-textycombinator
fix(meeting): host-boundary URL matching to kill substring false positives (#4442)
Vitalik176f14aceec9
diff --git a/crates/screenpipe-engine/src/meeting_detector.rs b/crates/screenpipe-engine/src/meeting_detector.rsindex 7380cd1b7..fafb0a4c8 100644--- a/crates/screenpipe-engine/src/meeting_detector.rs+++ b/crates/screenpipe-engine/src/meeting_detector.rs@@ -2132,7 +2132,7 @@ fn browser_window_matches_meeting(return ids.browser_url_patterns.iter()- .any(|p| contains_case_insensitive(doc, p));+ .any(|p| browser_url_pattern_matches(doc, p));}if let Some(t) = title {let t_lower = t.to_lowercase();@@ -2160,6 +2160,64 @@ fn contains_case_insensitive(haystack: &str, needle: &str) -> bool {haystack.to_lowercase().contains(&needle.to_lowercase())}+/// Match a `browser_url_patterns` entry against a URL or window-title string at+/// host-name boundaries.+///+/// A bare substring match leaks: `daily.co` is a substring of `thedaily.com`,+/// so an unrelated news site became a phantom meeting candidate (the latent+/// vector behind #4246, follow-up here). This matches a dotted host pattern+/// only as a whole hostname or a subdomain of it, and a path-qualified pattern+/// (`cal.com/video`, `zoom.us/j`) only when both the host segment and the path+/// component are bounded:+/// * `daily.co` → matches `daily.co`, `app.daily.co/room`; NOT+/// `thedaily.com`, `daily.com`, `daily.co.uk`.+/// * `cal.com/video` → matches `app.cal.com/video/uid`; NOT `cal.com/videos`,+/// `cal.com/pricing`.+///+/// Non-domain patterns (containing a space, or with no `.`) keep the previous+/// case-insensitive substring behavior — they are free-text markers+/// (e.g. "zoom meeting"), not hostnames.+fn browser_url_pattern_matches(haystack: &str, pattern: &str) -> bool {+ if pattern.contains(' ') || !pattern.contains('.') {+ return contains_case_insensitive(haystack, pattern);+ }+ contains_at_domain_boundary(haystack, pattern)+}++/// True when the dotted `pattern` appears in `haystack` bounded so its leading+/// host label is whole (or a subdomain boundary, i.e. preceded by `.`) and its+/// trailing character does not extend the matched host/path component. ASCII+/// case-insensitive; `pattern` is expected lowercase-or-mixed (compared+/// case-folded).+fn contains_at_domain_boundary(haystack: &str, pattern: &str) -> bool {+ let hay = haystack.to_ascii_lowercase();+ let pat = pattern.to_ascii_lowercase();+ let (hb, pb) = (hay.as_bytes(), pat.as_bytes());+ if pb.is_empty() || pb.len() > hb.len() {+ return false;+ }+ // A char that, on the LEFT of the match, would make the host label longer+ // (so the pattern is a tail of a bigger label, e.g. `daily` in `thedaily`).+ // `.` is intentionally NOT in this set: it marks a subdomain boundary.+ let extends_left = |c: u8| c.is_ascii_alphanumeric() || c == b'-';+ // On the RIGHT, anything that continues the host or path component: alnum,+ // `-`, or `.` (which would make `daily.co` part of `daily.com`/`daily.co.uk`).+ let extends_right = |c: u8| c.is_ascii_alphanumeric() || c == b'-' || c == b'.';+ let mut i = 0;+ while i + pb.len() <= hb.len() {+ if &hb[i..i + pb.len()] == pb {+ let left_ok = i == 0 || !extends_left(hb[i - 1]);+ let after = i + pb.len();+ let right_ok = after == hb.len() || !extends_right(hb[after]);+ if left_ok && right_ok {+ return true;+ }+ }+ i += 1;+ }+ false+}+fn ends_with_ascii_case_insensitive(haystack: &str, suffix: &str) -> bool {let haystack = haystack.as_bytes();let suffix = suffix.as_bytes();@@ -2339,7 +2397,7 @@ fn has_browser_meeting_url(pid: i32, url_patterns: &[&str]) -> bool {let doc_for_match = url_without_query_or_fragment(&doc);if url_patterns.iter()- .any(|p| contains_case_insensitive(doc_for_match, p))+ .any(|p| browser_url_pattern_matches(doc_for_match, p)){return true;}@@ -2352,7 +2410,7 @@ fn has_browser_meeting_url(pid: i32, url_patterns: &[&str]) -> bool {if url_patterns.iter().filter(|p| p.contains('.'))- .any(|p| contains_case_insensitive(&title, p))+ .any(|p| browser_url_pattern_matches(&title, p)){return true;}@@ -2536,11 +2594,15 @@ pub fn find_running_meeting_apps(continue;}+ // url_patterns are matched against the window TITLE on Windows (no+ // per-tab URL available). Boundary matching still applies — it keeps+ // a real meeting host in a title matching while closing the+ // `daily.co` ⊂ `thedaily.com` substring leak.let url_match = profile.app_identifiers.browser_url_patterns.iter()- .any(|p| contains_case_insensitive(title, p));+ .any(|p| browser_url_pattern_matches(title, p));// See `browser_title_matches_pattern` for the matching rules.let title_match = !profile.app_identifiers.browser_title_patterns.is_empty() && {let title_lower = title.to_lowercase();@@ -4229,13 +4291,105 @@ mod tests {.expect("generic fallback profile present")}- /// Mirrors the lowercase substring match used by `has_browser_meeting_url`- /// and `db_find_browser_meetings`.+ /// Exercises the production matcher used by `browser_window_matches_meeting`+ /// / `has_browser_meeting_url` / `db_find_browser_meetings` so these tests+ /// validate the real host-boundary logic, not a stand-in.fn url_matches_any_pattern(url: &str, patterns: &[&str]) -> bool {- let url_lower = url.to_lowercase();- patterns- .iter()- .any(|p| url_lower.contains(&p.to_lowercase()))+ patterns.iter().any(|p| browser_url_pattern_matches(url, p))+ }++ #[test]+ fn test_url_boundary_matcher_host_patterns() {+ // A bare host pattern matches the host and its subdomains, but never a+ // longer label that merely ends with the same letters.+ for hit in [+ "daily.co",+ "https://daily.co",+ "https://daily.co/room/abc",+ "https://app.daily.co/room",+ "https://my.team.daily.co/x",+ ] {+ assert!(+ browser_url_pattern_matches(hit, "daily.co"),+ "{hit:?} should match host pattern daily.co"+ );+ }+ for miss in [+ "https://www.thedaily.com",+ "https://thedaily.com/news",+ "https://dailywire.com",+ "https://daily.com", // different TLD+ "https://daily.co.uk", // different (longer) domain+ "https://notdaily.co.uk",+ ] {+ assert!(+ !browser_url_pattern_matches(miss, "daily.co"),+ "{miss:?} must NOT match host pattern daily.co (substring leak)"+ );+ }+ }++ #[test]+ fn test_url_boundary_matcher_path_qualified() {+ for hit in [+ "https://app.cal.com/video/uid",+ "https://cal.com/video/8f3e",+ "https://cal.com/video", // exact, no trailing segment+ ] {+ assert!(+ browser_url_pattern_matches(hit, "cal.com/video"),+ "{hit:?} should match cal.com/video"+ );+ }+ for miss in [+ "https://cal.com/videos", // /video is not a prefix component+ "https://cal.com/pricing",+ "https://app.cal.com/event-types",+ "https://thecal.com/video", // host label extended on the left+ ] {+ assert!(+ !browser_url_pattern_matches(miss, "cal.com/video"),+ "{miss:?} must NOT match cal.com/video"+ );+ }+ }++ #[test]+ fn test_url_boundary_matcher_keeps_freetext_markers() {+ // Non-host patterns (space / no dot) keep substring behavior so existing+ // free-text markers aren't silently disabled.+ assert!(browser_url_pattern_matches(+ "some Zoom Meeting in progress",+ "zoom meeting"+ ));+ }++ #[test]+ fn test_generic_profile_rejects_daily_co_lookalike() {+ // The concrete vector this PR closes: daily.co (a bare host pattern) must+ // not match thedaily.com / dailywire.com, while real daily.co calls do.+ let profile = generic_profile();+ let patterns = profile.app_identifiers.browser_url_patterns;+ assert!(+ patterns.contains(&"daily.co"),+ "fixture: daily.co host pattern present"+ );+ for miss in [+ "https://www.thedaily.com",+ "https://thedaily.com/podcast",+ "https://dailywire.com/news",+ ] {+ assert!(+ !url_matches_any_pattern(miss, patterns),+ "{miss:?} should NOT match a meeting profile (daily.co substring leak)"+ );+ }+ for hit in ["https://daily.co/standup", "https://app.daily.co/room/x"] {+ assert!(+ url_matches_any_pattern(hit, patterns),+ "real Daily call {hit:?} should still match"+ );+ }}#[test]