flowproof 0.23.2__tar.gz → 0.23.3__tar.gz

{flowproof-0.23.2 → flowproof-0.23.3}/Cargo.lock

@@ -930,7 +930,7 @@ dependencies = [
 
 [[package]]
 name = "flowproof-adapters"
-version = "0.23.2"
+version = "0.23.3"
 dependencies = [
  "anyhow",
  "flowproof-driver",
@@ -949,7 +949,7 @@ dependencies = [
 
 [[package]]
 name = "flowproof-agent"
-version = "0.23.2"
+version = "0.23.3"
 dependencies = [
  "chrono",
  "flowproof-driver",
@@ -967,7 +967,7 @@ dependencies = [
 
 [[package]]
 name = "flowproof-cli"
-version = "0.23.2"
+version = "0.23.3"
 dependencies = [
  "ab_glyph",
  "chrono",
@@ -992,7 +992,7 @@ dependencies = [
 
 [[package]]
 name = "flowproof-driver"
-version = "0.23.2"
+version = "0.23.3"
 dependencies = [
  "calamine",
  "flowproof-trace",
@@ -1008,7 +1008,7 @@ dependencies = [
 
 [[package]]
 name = "flowproof-python"
-version = "0.23.2"
+version = "0.23.3"
 dependencies = [
  "flowproof-agent",
  "flowproof-cli",
@@ -1020,7 +1020,7 @@ dependencies = [
 
 [[package]]
 name = "flowproof-replay"
-version = "0.23.2"
+version = "0.23.3"
 dependencies = [
  "chrono",
  "flowproof-agent",
@@ -1034,7 +1034,7 @@ dependencies = [
 
 [[package]]
 name = "flowproof-trace"
-version = "0.23.2"
+version = "0.23.3"
 dependencies = [
  "jsonschema",
  "regex",

{flowproof-0.23.2 → flowproof-0.23.3}/Cargo.toml

@@ -22,7 +22,7 @@ members = ["crates/flowproof-driver", "crates/flowproof-trace", "crates/flowproo
 exclude = ["spike/windows-containment"]
 
 [workspace.package]
-version = "0.23.2"
+version = "0.23.3"
 edition = "2021"
 license = "Apache-2.0"
 repository = "https://github.com/automators-com/flowproof"

{flowproof-0.23.2 → flowproof-0.23.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: flowproof
-Version: 0.23.2
+Version: 0.23.3
 Classifier: Development Status :: 3 - Alpha
 Classifier: Intended Audience :: Developers
 Classifier: Programming Language :: Python :: 3

{flowproof-0.23.2 → flowproof-0.23.3}/crates/flowproof-adapters/src/web.rs

@@ -4691,8 +4691,14 @@ impl AppDriver for WebAppDriver {
                 KeyMod::Meta => ModifierKey::Meta,
             })
             .collect();
+        // CDP's key table uses the literal space as the key value; models
+        // and human-authored specs commonly use its code/name instead.
+        let browser_key = match key {
+            "Space" | "Spacebar" => " ",
+            _ => key,
+        };
         self.tab()?
-            .press_key_with_modifiers(key, (!mods.is_empty()).then_some(mods.as_slice()))
+            .press_key_with_modifiers(browser_key, (!mods.is_empty()).then_some(mods.as_slice()))
             .map_err(|e| web_err(&format!("pressing key '{key}'"), e))?;
         Ok(())
     }

{flowproof-0.23.2 → flowproof-0.23.3}/crates/flowproof-adapters/tests/static_area_e2e.rs

@@ -32,6 +32,7 @@ fn serve(html: &'static str) -> String {
 }
 
 const FIXTURE: &str = r#"<!doctype html><html><body>
+    <input id="space-toggle" type="checkbox">
     <div id="app">
       <button id="open" onclick="
         var d = document.createElement('div');
@@ -69,6 +70,26 @@ fn search_reaches_content_outside_the_app_root() {
         .invoke(&UiaSelector::css("#open"))
         .expect("opening the dialog succeeds");
 
+    // Space aliases must produce real keyboard activation, not text insertion.
+    driver
+        .invoke(&UiaSelector::css("#space-toggle"))
+        .expect("focus checkbox");
+    assert!(driver
+        .element_exists(&UiaSelector::css("#space-toggle:checked"))
+        .expect("checked"));
+    for key in ["Space", "Spacebar", " "] {
+        driver
+            .press_key(key, &[])
+            .expect("space key activates checkbox");
+        assert!(driver
+            .element_exists(&UiaSelector::css("#space-toggle:not(:checked)"))
+            .expect("unchecked"));
+        driver.press_key(key, &[]).expect("space toggles back");
+        assert!(driver
+            .element_exists(&UiaSelector::css("#space-toggle:checked"))
+            .expect("checked again"));
+    }
+
     // Native-id (css) reach: the sibling div is appended to <body>, a
     // structural ancestor of neither #app nor anything inside it.
     assert!(

{flowproof-0.23.2 → flowproof-0.23.3}/crates/flowproof-agent/src/author.rs

@@ -67,7 +67,7 @@ action that reaches the goal.
 - Respond with ONLY JSON, no prose, no code fences.
 - The JSON action is one of: \"click\", \"click_at\", \"drag\", \"type_text\", \
 \"assert_text\", \"capture_text\", \"capture_count\", \"type_captured\", \
-\"select_option\", \"select_options\", \"scroll\", \"press_key\", \"rule_step\", or \
+\"select_option\", \"select_options\", \"scroll\", \"scroll_into_view\", \"press_key\", \"rule_step\", or \
 \"capture_ambiguity\".
 - UI actions MUST include \"target\": \"<target token of a listed element>\". \
 Clicking and typing require an entry whose \"actionable\" field is true. \
@@ -77,7 +77,11 @@ type_text also needs \"text\"; assert_text needs \"expected\" and optional \
 drag needs \"onto\" with a second listed actionable target. click_at needs \
 \"x_pct\" and \"y_pct\" from 0 through 100. capture_count needs a safe \"name\". \
 select_option needs one \"text\" value; select_options is only for a multi-select \
-and needs a non-empty \"values\" array. scroll needs \"to_px\". \
+and needs a non-empty \"values\" array. scroll needs \"to_px\" and means an exact \
+vertical offset inside a scroll container. To reveal a field or cell, use \
+scroll_into_view with that individual target instead; it needs no offset. \
+Never scroll a collection (tag=collection): it represents multiple rows for counting, \
+not a scroll container. For horizontal grids, reveal the desired cell, not a row collection. \
 press_key needs \"key\" and has no target. \
 For another action, use rule_step with \"step\" containing exact \
 deterministic grammar and copy listed target tokens into quoted targets, for \
@@ -881,12 +885,32 @@ fn ground_one(
                 .ok_or("select_option needs a non-empty 'text'")?;
             Ok(vec![ResolvedAction::TypeText { target, text }])
         }
-        "scroll" => Ok(vec![ResolvedAction::Scroll {
-            target: Some(target),
-            to: crate::rules::ScrollTo::Offset(
-                authored.to_px.ok_or("scroll needs a 'to_px' offset")?,
-            ),
-        }]),
+        "scroll" | "scroll_into_view" => {
+            let entries: Vec<serde_json::Value> = serde_json::from_str(scene).unwrap_or_default();
+            if entries.iter().any(|entry| {
+                (entry["target"].as_str() == Some(token)
+                    || entry["css"]
+                        .as_str()
+                        .is_some_and(|css| token == css || token == format!("css:{css}")))
+                    && entry["tag"] == "collection"
+            }) {
+                return Err("cannot scroll a collection; choose an individual cell with scroll_into_view or a scroll container with scroll".into());
+            }
+            let to = if authored.action == "scroll_into_view" {
+                if authored.to_px.is_some() {
+                    return Err("scroll_into_view does not accept to_px".into());
+                }
+                crate::rules::ScrollTo::IntoView
+            } else {
+                crate::rules::ScrollTo::Offset(
+                    authored.to_px.ok_or("scroll needs a 'to_px' offset")?,
+                )
+            };
+            Ok(vec![ResolvedAction::Scroll {
+                target: Some(target),
+                to,
+            }])
+        }
         "type_captured" => {
             let name = authored
                 .capture
@@ -2329,6 +2353,34 @@ mod tests {
         }
     }
 
+    #[test]
+    fn collection_scroll_is_rejected_and_reauthored_as_reveal_cell() {
+        let mut client = Scripted {
+            replies: vec![
+                r#"{"action":"scroll","target":"css:#rows tr","to_px":0}"#.into(),
+                r#"{"action":"scroll_into_view","target":"css:#task-1"}"#.into(),
+            ],
+            calls: 0,
+        };
+        let result = author_step(
+            &mut client,
+            &AuthorContext {
+                intent: "Bring the first task into view",
+                scene: HUMAN_PRIMITIVE_SCENE,
+                ..ctx()
+            },
+        )
+        .expect("collection rejection should reauthor an individual target");
+        assert_eq!(client.calls, 2);
+        assert_eq!(
+            result,
+            ResolvedAction::Scroll {
+                target: Some(Target::css("#task-1")),
+                to: crate::rules::ScrollTo::IntoView,
+            }
+        );
+    }
+
     #[test]
     fn rule_step_covers_grounded_actions_outside_the_small_json_vocabulary() {
         let mut client = Scripted {

{flowproof-0.23.2 → flowproof-0.23.3}/crates/flowproof-agent/src/llm.rs

@@ -225,20 +225,24 @@ impl HttpModelClient {
         if let Some(key) = &self.config.api_key {
             request = request.header("authorization", format!("Bearer {key}"));
         }
-        // `temperature` stays here: OpenAI's API and a local vLLM both accept
-        // it, and determinism is worth having where it is free. The
-        // deprecation that forced it out of the Anthropic request is specific
-        // to that API's current models.
+        let mut body = json!({
+            "model": self.model(),
+            "messages": [
+                {"role": "system", "content": system},
+                {"role": "user", "content": user},
+            ],
+        });
+        if self.config.kind == BackendKind::OpenAi {
+            // Includes reasoning tokens. GPT-5 rejects legacy max_tokens and
+            // temperature=0; leave sampling at the model's supported default.
+            body["max_completion_tokens"] = json!(16_384);
+        } else {
+            // Preserve the contract of local OpenAI-compatible servers.
+            body["max_tokens"] = json!(OPENAI_COMPATIBLE_MAX_TOKENS);
+            body["temperature"] = json!(0);
+        }
         let mut response = request
-            .send_json(json!({
-                "model": self.model(),
-                "max_tokens": OPENAI_COMPATIBLE_MAX_TOKENS,
-                "temperature": 0,
-                "messages": [
-                    {"role": "system", "content": system},
-                    {"role": "user", "content": user},
-                ],
-            }))
+            .send_json(body)
             .map_err(|e| Self::http_err("openai-compatible", e))?;
         let response: serde_json::Value = Self::json_or_error(
             "openai-compatible",

{flowproof-0.23.2 → flowproof-0.23.3}/crates/flowproof-agent/tests/model_client_http.rs

@@ -132,3 +132,33 @@ fn an_echoed_key_is_redacted_from_the_error() {
         "and the redaction is visible: {text}"
     );
 }
+
+#[test]
+fn openai_reasoning_request_and_compatible_server_use_their_own_parameters() {
+    for kind in [BackendKind::OpenAi, BackendKind::OpenAiCompatible] {
+        let (base, rx) = serve_once(200, r#"{"choices":[{"message":{"content":"ok"}}]}"#);
+        let mut client = HttpModelClient::new(BackendConfig {
+            kind: kind.clone(),
+            base_url: Some(base),
+            model: Some("gpt-5".into()),
+            api_key: Some("test-key".into()),
+        });
+        assert_eq!(
+            client
+                .complete("system", "user")
+                .expect("completion succeeds"),
+            "ok"
+        );
+        let sent: serde_json::Value = serde_json::from_str(&rx.recv().expect("request captured"))
+            .expect("valid request JSON");
+        if kind == BackendKind::OpenAi {
+            assert_eq!(sent["max_completion_tokens"], 16384);
+            assert!(sent.get("max_tokens").is_none());
+            assert!(sent.get("temperature").is_none());
+        } else {
+            assert_eq!(sent["max_tokens"], 1024);
+            assert_eq!(sent["temperature"], 0);
+            assert!(sent.get("max_completion_tokens").is_none());
+        }
+    }
+}

{flowproof-0.23.2 → flowproof-0.23.3}/flowproof/__init__.py

@@ -24,7 +24,7 @@ from flowproof.flow import (
     run,
 )
 
-__version__ = "0.23.2"
+__version__ = "0.23.3"
 
 __all__ = [
     "ClarificationNeeded",

{flowproof-0.23.2 → flowproof-0.23.3}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "maturin"
 
 [project]
 name = "flowproof"
-version = "0.23.2"
+version = "0.23.3"
 description = "Test that a bot only ever does the job it was given — and never deletes, sends or approves the thing it must not. Record it doing the job once, then replay on every commit with zero LLM calls. For the agents and RPA already running real work in production: web, desktop, SAP, Citrix."
 readme = "README.md"
 license = "Apache-2.0"