browser_oxide 0.1.3

Stealth headless browser engine in Rust: real HTML/CSS/DOM/JS, V8 via deno_core, own BoringSSL TLS/JA4 fingerprint, no Chromium, no CDP — for anti-bot web scraping, archival, and AI agents
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210

//! Verify the sigma-lognormal humanizer produces the signals
//! (mouse / scroll / click / key) that anti-bot sensors gate on,
//! with the right shape characteristics.
//!
//! This test runs `Page::navigate_humanized` (which is just
//! `Page::navigate` — humanize is default-on) on a synthetic page
//! that records every dispatched event into `globalThis.__events`.
//! After ~3.2 s of event-loop time we read the array back and check:
//!
//! 1. **Mouse path** — at least 30 mousemove events, distance >800px
//! 2. **Scroll** — 4 wheel + 4 scroll events with monotonically
//!    decreasing deltaY (decel)
//! 3. **Click** — full mousedown / mouseup / click sequence
//! 4. **Key** — Tab keydown + keyup
//! 5. **Sigma-lognormal shape** — mousemove inter-arrival times peak
//!    in the early third (mode), with a long tail. We test this by
//!    checking that the modal interval is < median, which is what an
//!    asymmetric-right distribution requires.

use browser_oxide::stealth::presets::chrome_148_macos;
use browser_oxide::Page;

const HTML: &str = r#"<!doctype html><html><head><title>humanize</title></head><body>
<script>
  globalThis.__events = [];
  globalThis.__lastMove = null;
  const record = (kind, ev) => {
    const e = { kind, t: Date.now() };
    if (ev) {
      if (typeof ev.clientX === 'number') { e.x = ev.clientX; e.y = ev.clientY; }
      if (typeof ev.deltaY === 'number') e.deltaY = ev.deltaY;
      if (typeof ev.key === 'string') e.key = ev.key;
      e.isTrusted = ev.isTrusted;
    }
    globalThis.__events.push(e);
  };
  document.addEventListener('mousemove', e => record('mousemove', e));
  document.addEventListener('mousedown', e => record('mousedown', e));
  document.addEventListener('mouseup',   e => record('mouseup',   e));
  document.addEventListener('click',     e => record('click',     e));
  document.addEventListener('wheel',     e => record('wheel',     e), { passive: true });
  document.addEventListener('scroll',    e => record('scroll',    e), { passive: true });
  document.addEventListener('keydown',   e => record('keydown',   e));
  document.addEventListener('keyup',     e => record('keyup',     e));
  document.addEventListener('visibilitychange', e => record('visibilitychange', e));
  window.addEventListener('focus',       e => record('focus',     e));
  window.addEventListener('scroll',      e => record('scroll-win', e));
</script>
</body></html>"#;

#[tokio::test]
#[ignore = "not yet implemented: wheel-event emission in the synthetic input path"]
async fn humanize_emits_full_signal_set() {
    // navigate_humanized installs humanize.js as an init script; from_html
    // does NOT, so we'd see nothing on a default Page::from_html. Use
    // navigate_humanized via the existing entrypoint.
    //
    // Trick: navigate_humanized takes a URL. We don't want a network test
    // here, so we use a data: URL that decodes to the probe HTML.
    let mut page = Page::from_html(HTML, Some(chrome_148_macos()))
        .await
        .unwrap();
    // Manually inject humanize.js and drive enough time for it to fire.
    let humanize = include_str!("../src/js/humanize.js");
    page.evaluate(humanize).unwrap();

    // Drive the event loop until quiescent (or timeout). humanize.js
    // schedules setTimeouts up to ~3.2 s out, so a 4 s budget covers
    // everything. `page.evaluate("0")` alone doesn't fire pending
    // timers — we need run_until_idle.
    let _ = page
        .event_loop()
        .run_until_idle(std::time::Duration::from_secs(4))
        .await;

    let n_total = page.evaluate("globalThis.__events.length").unwrap();
    let n_mousemove = page
        .evaluate("globalThis.__events.filter(e => e.kind === 'mousemove').length")
        .unwrap();
    let n_wheel = page
        .evaluate("globalThis.__events.filter(e => e.kind === 'wheel').length")
        .unwrap();
    let n_click = page
        .evaluate("globalThis.__events.filter(e => e.kind === 'click').length")
        .unwrap();
    let n_keydown = page
        .evaluate("globalThis.__events.filter(e => e.kind === 'keydown').length")
        .unwrap();
    let total_distance = page
        .evaluate(
            "(() => {\
                const m = globalThis.__events.filter(e => e.kind === 'mousemove' && typeof e.x === 'number');\
                let d = 0;\
                for (let i = 1; i < m.length; i++) { d += Math.hypot(m[i].x-m[i-1].x, m[i].y-m[i-1].y); }\
                return Math.round(d);\
            })()",
        )
        .unwrap();
    let monotonic_decel = page
        .evaluate(
            "(() => {\
                const w = globalThis.__events.filter(e => e.kind === 'wheel');\
                if (w.length < 2) return 'too-few';\
                let prev = Infinity;\
                for (const e of w) { if (e.deltaY > prev) return 'not-monotonic'; prev = e.deltaY; }\
                return 'ok';\
            })()",
        )
        .unwrap();
    let trusted_count = page
        .evaluate("globalThis.__events.filter(e => e.isTrusted === true).length")
        .unwrap();

    println!("\n=== Humanize signal probe ===");
    println!("  total events:     {n_total}");
    println!("  mousemove:        {n_mousemove}");
    println!("  wheel:            {n_wheel}  (decel: {monotonic_decel})");
    println!("  click:            {n_click}");
    println!("  keydown:          {n_keydown}");
    println!("  path distance:    {total_distance} px");
    println!("  isTrusted=true:   {trusted_count}");

    let mm = n_mousemove.parse::<i64>().unwrap();
    assert!(
        mm >= 30,
        "humanize must emit ≥30 mousemove events, got {mm}"
    );
    let w = n_wheel.parse::<i64>().unwrap();
    assert!(w >= 4, "humanize must emit ≥4 wheel events, got {w}");
    let c = n_click.parse::<i64>().unwrap();
    assert!(c >= 1, "humanize must emit ≥1 click");
    let k = n_keydown.parse::<i64>().unwrap();
    assert!(k >= 1, "humanize must emit ≥1 keydown");
    let dist = total_distance.parse::<i64>().unwrap();
    assert!(
        dist >= 800,
        "mouse path must cover ≥800 px (sensors check magnitude), got {dist}"
    );
    assert_eq!(
        monotonic_decel.trim_matches('"'),
        "ok",
        "scroll wheel deltaY must monotonically decrease (deceleration model)"
    );
    let trusted = trusted_count.parse::<i64>().unwrap();
    assert!(
        trusted >= 30,
        "events must have isTrusted=true so handlers gating on it fire — got {trusted}"
    );
}

/// Sigma-lognormal shape check — mousemove inter-arrival times must
/// peak in the early portion (asymmetric right-skewed distribution).
/// This is what real human cursor motion looks like; the previous
/// uniform-time Bezier produced a flat distribution that's
/// distinguishable from human input.
#[tokio::test]
async fn humanize_mouse_intervals_are_right_skewed() {
    let mut page = Page::from_html(HTML, Some(chrome_148_macos()))
        .await
        .unwrap();
    let humanize = include_str!("../src/js/humanize.js");
    page.evaluate(humanize).unwrap();

    // humanize schedules its mouse stroke on *background* (unref'd) timers so
    // it doesn't pin `run_until_idle` open on benign pages (humanize.js:73 →
    // timer_bootstrap.js `__bgSetTimeout`). A single `run_until_idle` therefore
    // returns at idle before any scheduled move fires, leaving only the 2
    // synchronous seeding moves. The bg-timer callbacks fire on *successive*
    // pumps as wall-clock time elapses (timer_bootstrap.js:33), so drive the
    // loop in short real-time slices instead of one idle wait.
    for _ in 0..90 {
        let _ = page
            .event_loop()
            .run_until_idle(std::time::Duration::from_millis(20))
            .await;
        tokio::time::sleep(std::time::Duration::from_millis(40)).await;
    }

    // Compute inter-arrival times for the first stroke (≈18 events).
    let stats = page.evaluate(
        "(() => {\
            const m = globalThis.__events.filter(e => e.kind === 'mousemove' && typeof e.t === 'number');\
            if (m.length < 10) return JSON.stringify({error: 'too-few', n: m.length});\
            const intervals = [];\
            for (let i = 1; i < m.length; i++) intervals.push(m[i].t - m[i-1].t);\
            intervals.sort((a, b) => a - b);\
            const mid = intervals.length >> 1;\
            const median = intervals.length % 2 ? intervals[mid] : (intervals[mid-1]+intervals[mid])/2;\
            const mean = intervals.reduce((a, b) => a + b, 0) / intervals.length;\
            const minVal = intervals[0];\
            const maxVal = intervals[intervals.length - 1];\
            return JSON.stringify({n: intervals.length, mean: +mean.toFixed(2), median: +median.toFixed(2), min: +minVal.toFixed(2), max: +maxVal.toFixed(2)});\
        })()"
    ).unwrap();
    println!("\nmousemove inter-arrival stats: {}", stats);
    let stats_clean = stats.trim_matches('"').replace("\\\"", "\"");
    let v: serde_json::Value = serde_json::from_str(&stats_clean).expect("stats json");
    let mean = v["mean"].as_f64().unwrap();
    let median = v["median"].as_f64().unwrap();
    let max = v["max"].as_f64().unwrap();
    let min = v["min"].as_f64().unwrap();
    assert!(
        mean > median,
        "right-skewed distribution must have mean > median (sigma-lognormal property), got mean={mean} median={median}"
    );
    assert!(
        max > min * 3.0,
        "interval range must span ≥3× (asymmetric distribution; uniform sampling would have max ≈ min), got min={min} max={max}"
    );
}