rdesktop-dev 0.1.2

Agent-first development server for rdesktop - browser mode for AI agent workflows
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
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367

//! Agent API endpoints.
//!
//! These endpoints allow AI agents to interact with the running application
//! through structured HTTP requests, without needing native desktop control.
//!
//! ## How it works
//!
//! 1. The browser app includes the rdesktop bridge script
//! 2. The bridge script periodically sends DOM snapshots to the server
//! 3. Agents query the server for DOM/state information
//! 4. Agents send actions to the server, which forwards them to the browser
//!
//! ## Design Philosophy
//!
//! Instead of requiring agents to take screenshots and use vision models to
//! understand the UI, the Agent API provides direct DOM access and structured
//! state information. This is:
//!
//! - **Faster**: No screenshot encoding/decoding overhead
//! - **More reliable**: Exact element selectors, not pixel coordinates
//! - **More informative**: Full DOM tree, computed styles, accessibility info
//! - **Easier to test**: Standard HTTP endpoints, can be scripted

use axum::extract::{Query, State};
use axum::http::StatusCode;
use axum::response::IntoResponse;
use axum::Json;
use serde::{Deserialize, Serialize};
use std::collections::HashMap;

use crate::server::DevServerState;

/// Query parameters for element selection.
#[derive(Debug, Deserialize)]
pub struct ElementQuery {
    /// CSS selector to query elements
    pub selector: Option<String>,

    /// Text content to search for
    pub text: Option<String>,

    /// Role attribute to filter by
    pub role: Option<String>,
}

/// An action that an agent can execute on the UI.
#[derive(Debug, Deserialize, Serialize)]
pub struct AgentAction {
    /// The type of action
    pub action: ActionType,

    /// CSS selector of the target element
    pub selector: String,

    /// Value for type/fill actions
    pub value: Option<String>,

    /// Coordinates for scroll actions
    pub coordinates: Option<(f64, f64)>,
}

/// Types of actions agents can execute.
#[derive(Debug, Deserialize, Serialize)]
#[serde(rename_all = "snake_case")]
pub enum ActionType {
    Click,
    DoubleClick,
    RightClick,
    Type,
    Fill,
    Clear,
    Scroll,
    Hover,
    Focus,
    Select,
}

/// Response from a DOM query.
#[derive(Debug, Serialize)]
pub struct DomSnapshot {
    /// The full HTML content
    pub html: String,

    /// The page URL
    pub url: String,

    /// The page title
    pub title: String,

    /// Timestamp of the snapshot
    pub timestamp: String,
}

/// Response from an element query.
#[derive(Debug, Serialize)]
pub struct ElementInfo {
    /// CSS selector that uniquely identifies this element
    pub selector: String,

    /// Tag name
    pub tag: String,

    /// Text content
    pub text: String,

    /// Element attributes
    pub attributes: HashMap<String, String>,

    /// Whether the element is visible
    pub visible: bool,

    /// Whether the element is enabled (for interactive elements)
    pub enabled: bool,

    /// Accessibility role
    pub role: Option<String>,

    /// Accessibility label
    pub label: Option<String>,
}

/// Result of an action execution.
#[derive(Debug, Serialize)]
pub struct ActionResult {
    /// Whether the action succeeded
    pub success: bool,

    /// Error message if the action failed
    pub error: Option<String>,

    /// Any side effects (e.g., navigation that occurred)
    pub side_effects: Vec<String>,
}

/// GET /__rdesktop__/agent/dom
///
/// Returns a full DOM snapshot of the current page.
/// The snapshot is collected from the browser via the bridge script.
pub async fn get_dom(
    State(state): State<DevServerState>,
) -> impl IntoResponse {
    let snapshot = state.last_dom_snapshot.read().await;

    let html = snapshot.clone().unwrap_or_else(|| {
        r#"<!DOCTYPE html>
<html>
<head><title>rdesktop</title></head>
<body>
  <p>No DOM snapshot available yet. Make sure the app is loaded in the browser.</p>
  <p>The bridge script will send DOM updates automatically.</p>
</body>
</html>"#
            .to_string()
    });

    let dom = DomSnapshot {
        html,
        url: "http://localhost".to_string(),
        title: "rdesktop App".to_string(),
        timestamp: timestamp(),
    };

    Json(dom).into_response()
}

/// GET /__rdesktop__/agent/elements?selector=...
///
/// Query elements matching a CSS selector or text content.
pub async fn query_elements(
    State(state): State<DevServerState>,
    Query(query): Query<ElementQuery>,
) -> impl IntoResponse {
    let snapshot = state.last_dom_snapshot.read().await;

    // Parse the DOM and find matching elements
    let elements: Vec<ElementInfo> = if let Some(ref html) = *snapshot {
        find_elements(html, &query)
    } else {
        vec![]
    };

    Json(serde_json::json!({
        "query": {
            "selector": query.selector,
            "text": query.text,
            "role": query.role,
        },
        "count": elements.len(),
        "elements": elements,
    }))
    .into_response()
}

/// POST /__rdesktop__/agent/action
///
/// Execute a UI action (click, type, scroll, etc.)
/// The action is stored and picked up by the bridge script.
pub async fn execute_action(
    State(_state): State<DevServerState>,
    Json(action): Json<AgentAction>,
) -> impl IntoResponse {
    tracing::info!(
        action = ?action.action,
        selector = %action.selector,
        "Agent action received"
    );

    // In a full implementation, this would:
    // 1. Store the action in a shared queue
    // 2. The bridge script polls for pending actions
    // 3. The bridge executes the action in the browser
    // 4. The result is returned

    let result = ActionResult {
        success: true,
        error: None,
        side_effects: vec![format!("Action {:?} on '{}' queued", action.action, action.selector)],
    };

    Json(result).into_response()
}

/// GET /__rdesktop__/agent/state
///
/// Get the current application state.
pub async fn get_state(
    State(state): State<DevServerState>,
) -> impl IntoResponse {
    let app_state = state.last_app_state.read().await;

    match app_state.as_ref() {
        Some(state) => Json(state.clone()).into_response(),
        None => Json(serde_json::json!({
            "message": "No application state available yet.",
            "hint": "Use fetch('/__rdesktop__/state', { method: 'POST', body: JSON.stringify(state) }) from your app."
        }))
        .into_response(),
    }
}

/// POST /__rdesktop__/agent/ipc
///
/// Send an IPC message from the agent to the app.
pub async fn send_ipc(
    State(_state): State<DevServerState>,
    Json(message): Json<serde_json::Value>,
) -> impl IntoResponse {
    let cmd = message["cmd"].as_str().unwrap_or("unknown");
    let payload = message["payload"].clone();
    let id = message["id"].as_str().unwrap_or("0");

    tracing::info!(cmd = cmd, "Agent IPC message received");

    // In a full implementation, this would forward to the Rust IPC handler.
    // For now, handle basic commands directly.
    let response = match cmd {
        "greet" => {
            let name = payload["name"].as_str().unwrap_or("World");
            serde_json::json!({
                "id": id,
                "success": true,
                "data": { "message": format!("Hello, {}!", name) }
            })
        }
        "ping" => {
            serde_json::json!({
                "id": id,
                "success": true,
                "data": { "pong": true }
            })
        }
        _ => {
            serde_json::json!({
                "id": id,
                "success": false,
                "data": { "error": format!("Unknown command: {}", cmd) }
            })
        }
    };

    Json(response).into_response()
}

/// GET /__rdesktop__/agent/screenshot
///
/// Capture a screenshot. In browser mode, this delegates to the browser.
pub async fn take_screenshot(
    State(_state): State<DevServerState>,
) -> impl IntoResponse {
    (
        StatusCode::NOT_IMPLEMENTED,
        Json(serde_json::json!({
            "message": "Screenshot not implemented in browser mode.",
            "hint": "Use Playwright's page.screenshot() directly."
        })),
    )
        .into_response()
}

/// Simple timestamp helper.
fn timestamp() -> String {
    let now = std::time::SystemTime::now()
        .duration_since(std::time::UNIX_EPOCH)
        .unwrap_or_default();
    format!("{}", now.as_secs())
}

/// Find elements in HTML matching the query.
/// This is a simple text-based search, not a full DOM parser.
fn find_elements(html: &str, query: &ElementQuery) -> Vec<ElementInfo> {
    let mut elements = vec![];

    if let Some(ref selector) = query.selector {
        // Simple tag selector matching (e.g., "button", "input", "h1")
        let tag = selector.trim_start_matches('<').trim_end_matches('>');
        let open_tag = format!("<{}", tag);

        let mut start = 0;
        while let Some(pos) = html[start..].find(&open_tag) {
            let abs_pos = start + pos;
            let end = html[abs_pos..].find('>').unwrap_or(0);
            let _tag_content = &html[abs_pos..abs_pos + end + 1];

            // Extract text content between tags
            let close_tag = format!("</{}>", tag);
            let text_start = abs_pos + end + 1;
            let text = if let Some(text_end) = html[text_start..].find(&close_tag) {
                html[text_start..text_start + text_end].trim().to_string()
            } else {
                String::new()
            };

            elements.push(ElementInfo {
                selector: format!("{}:nth-of-type({})", tag, elements.len() + 1),
                tag: tag.to_string(),
                text,
                attributes: HashMap::new(),
                visible: true,
                enabled: true,
                role: None,
                label: None,
            });

            start = abs_pos + end + 1;
        }
    }

    if let Some(ref text_query) = query.text {
        // Search for text content
        let lower_html = html.to_lowercase();
        let lower_query = text_query.to_lowercase();
        if lower_html.contains(&lower_query) {
            elements.push(ElementInfo {
                selector: format!("*:contains(\"{}\")", text_query),
                tag: "*".to_string(),
                text: text_query.clone(),
                attributes: HashMap::new(),
                visible: true,
                enabled: true,
                role: None,
                label: None,
            });
        }
    }

    elements
}