Skip to main content

llm_browser_testkit/
scenario.rs

1//! Scenario types for human-readable browser test case definitions.
2//!
3//! Scenarios are written in [TOML](https://toml.io) and describe groups of
4//! browser interaction tests with reusable assertion definitions and
5//! configurable test-level overrides.
6//!
7//! # Structure
8//!
9//! ```toml
10//! [config]                         # Global defaults
11//! start_url = "/dashboard"
12//!
13//! [[definitions]]                  # Reusable assertion definitions
14//! name = "no_errors"
15//! preset = "no_error_on_page"
16//!
17//! [[test]]                         # Test group
18//! name = "Dashboard Smoke"
19//! start_url = "/dashboard"         # Override global start_url (optional)
20//!
21//! [[test.steps]]                   # Ordered steps — the `kind` field
22//! kind = "navigate"                # determines which other fields apply
23//! url = "/dashboard"
24//!
25//! [[test.steps]]
26//! kind = "click"
27//! target = "the Login button"      # Natural language — LLM resolves to selector
28//!
29//! [[test.steps]]
30//! kind = "assert"
31//! definition = "no_errors"
32//! ```
33//!
34//! ## Step Kinds
35//!
36//! | `kind`        | Required fields  | Optional fields                          |
37//! |---------------|-----------------|------------------------------------------|
38//! | `navigate`    | `url`           | `wait_after_ms`                          |
39//! | `click`       | `target`        | `selector`, `wait_after_ms`              |
40//! | `type`        | `target`, `text`| `selector`, `wait_after_ms`              |
41//! | `wait`        | `target`        | `selector`, `timeout_ms`                 |
42//! | `assert`      | *one of below*  | —                                        |
43//! | `screenshot`  | —               | `path`                                   |
44//! | `agent`       | `agent`, `task` | —                                        |
45//! | `mcp`         | `server`, `tool`| `args`                                   |
46//!
47//! Assert steps require one of: `definition` (references a named
48//! `[[definitions]]` entry), `preset` (built-in preset name), or `prompt`
49//! (custom LLM evaluation prompt).
50
51use std::collections::HashMap;
52
53use serde::Deserialize;
54use serde_json::Value;
55
56/// Top-level scenario file, deserialized from TOML.
57#[derive(Debug, Deserialize)]
58pub struct Scenario {
59    /// Global configuration (overridable per test).
60    #[serde(default)]
61    pub config: ScenarioConfig,
62
63    /// Reusable assertion definitions referenced by name in `assert` steps.
64    #[serde(default)]
65    pub definitions: Vec<AssertDefinition>,
66
67    /// Ordered test groups to execute.
68    #[serde(default)]
69    pub test: Vec<TestGroup>,
70}
71
72/// Global scenario configuration with per-test overridable fields.
73#[derive(Debug, Deserialize, Default, Clone)]
74pub struct ScenarioConfig {
75    /// Base URL for relative navigation.
76    #[serde(default)]
77    pub base_url: Option<String>,
78    /// LLM server base URL (deprecated; prefer `[config.endpoints]`).
79    #[serde(default)]
80    pub llm_url: Option<String>,
81    /// LLM model name (deprecated; prefer `[config.endpoints]`).
82    #[serde(default)]
83    pub llm_model: Option<String>,
84    /// LLM API key (Bearer token).
85    #[serde(default)]
86    pub llm_api_key: Option<String>,
87    /// Custom HTTP headers as JSON key-value pairs.
88    #[serde(default, deserialize_with = "deserialize_headers")]
89    pub llm_headers: HashMap<String, String>,
90    /// Run browser in headless mode.
91    #[serde(default)]
92    pub browser_headless: Option<bool>,
93    /// HTTP / browser action timeout in seconds.
94    #[serde(default)]
95    pub timeout_secs: Option<u64>,
96    /// Browser viewport width.
97    #[serde(default)]
98    pub viewport_width: Option<u32>,
99    /// Browser viewport height.
100    #[serde(default)]
101    pub viewport_height: Option<u32>,
102    /// Default URL every test auto-navigates to before running its steps.
103    #[serde(default)]
104    pub start_url: Option<String>,
105    /// Whether to auto-navigate to `start_url` before test steps.
106    ///
107    /// Disable when a test starts with click-based navigation.
108    #[serde(default = "default_auto_navigate")]
109    pub auto_navigate: bool,
110    /// LLM temperature (0.0–1.0). Lower = more deterministic.
111    #[serde(default = "default_temperature")]
112    pub temperature: f64,
113    /// Enable thinking/reasoning tokens. `None` means the provider default
114    /// is used (no `thinking` key is sent). Set to `true`/`false` to
115    /// explicitly enable or disable.
116    #[serde(default)]
117    pub thinking: Option<bool>,
118    /// Provider-specific model parameters merged into the chat completion
119    /// request body (e.g. `effort = "high"` for Anthropic).
120    #[serde(default, deserialize_with = "deserialize_model_params")]
121    pub model_params: HashMap<String, Value>,
122    /// Named endpoints (LLM, MCP, A2A agents) with pricing.
123    #[serde(default)]
124    pub endpoints: HashMap<String, EndpointConfig>,
125    /// Global and per-test budgets for cost/token/call limits.
126    #[serde(default)]
127    pub budgets: BudgetsConfig,
128    /// MCP server exposure configuration.
129    #[serde(default)]
130    pub mcp_server: Option<McpServerConfig>,
131    /// A2A agent server exposure configuration.
132    #[serde(default)]
133    pub a2a_server: Option<A2aServerConfig>,
134    /// Whether to continue running the remaining steps of a test after a
135    /// step fails. Default `false` = fail fast: the first failed step ends
136    /// the test and the rest are reported as skipped. Set to `true` to run
137    /// every step (more diagnostics, more LLM cost on broken apps).
138    #[serde(default)]
139    pub continue_on_failure: bool,
140    /// Longest edge (px) of screenshots attached to `screenshot = true`
141    /// assert steps, before they are JPEG-encoded and sent to the vision
142    /// endpoint. Downscaling keeps vision token cost/quality sane.
143    /// Default: 1400.
144    #[serde(default)]
145    pub screenshot_max_dimension: Option<u32>,
146    /// Directory for failure artifacts (screenshots, page snapshots).
147    /// Defaults to `artifacts`.
148    #[serde(default)]
149    pub artifacts_dir: Option<String>,
150    /// Optional viewport matrix: when set, every test in the scenario is
151    /// expanded into one variant per viewport (e.g. mobile/tablet/desktop).
152    /// Each variant overrides the test's viewport and gets a ` — <name>`
153    /// suffix on the test name. Per-test budgets apply per variant.
154    #[serde(default)]
155    pub viewport_matrix: Option<ViewportMatrix>,
156    /// Class-name prefixes the `layout_no_issues` scan skips: elements
157    /// whose class matches any prefix are ignored by the fixed-element,
158    /// text-clipped, and overlap checks. Defaults cover the Angular CDK
159    /// screen-reader helpers (`cdk-visually-hidden`,
160    /// `cdk-describedby-message-container`, `cdk-overlay-container`),
161    /// which are intentionally 1x1 / off-screen.
162    #[serde(default = "default_layout_ignore_classes")]
163    pub layout_ignore_classes: Vec<String>,
164}
165
166/// A list of named viewports a scenario is expanded across.
167#[derive(Debug, Deserialize, Clone, Default)]
168pub struct ViewportMatrix {
169    /// The viewport variants (`{name, width, height}`).
170    #[serde(default)]
171    pub viewports: Vec<ViewportDef>,
172}
173
174/// One named viewport size in a matrix.
175#[derive(Debug, Deserialize, Clone)]
176pub struct ViewportDef {
177    /// Human-readable variant name (appended to test names, e.g.
178    /// `— mobile`).
179    pub name: String,
180    /// Browser viewport width in pixels.
181    pub width: u32,
182    /// Browser viewport height in pixels.
183    pub height: u32,
184}
185
186/// A named endpoint definition with pricing.
187#[derive(Debug, Deserialize, Clone, Default)]
188pub struct EndpointConfig {
189    /// Endpoint type: `llm`, `mcp`, or `a2a`.
190    #[serde(rename = "type")]
191    pub endpoint_type: EndpointType,
192    /// Base URL for the endpoint.
193    #[serde(default)]
194    pub url: Option<String>,
195    /// Model name (LLM endpoints only).
196    #[serde(default)]
197    pub model: Option<String>,
198    /// API key / bearer token.
199    #[serde(default)]
200    pub api_key: Option<String>,
201    /// Custom HTTP headers as JSON key-value pairs.
202    #[serde(default, deserialize_with = "deserialize_headers")]
203    pub headers: HashMap<String, String>,
204    /// Pricing configuration.
205    #[serde(default)]
206    pub pricing: Option<PricingConfig>,
207    /// Task types this endpoint serves by default
208    /// (e.g. `["targeting", "assertion"]`).
209    #[serde(default)]
210    pub default_for: Vec<String>,
211    /// Command to launch an MCP server subprocess (stdio transport).
212    #[serde(default)]
213    pub command: Option<String>,
214    /// Arguments for the MCP server command.
215    #[serde(default)]
216    pub args: Vec<String>,
217    /// Whether this LLM endpoint accepts image parts (vision) in addition
218    /// to text. `assert` steps with `screenshot = true` require a vision
219    /// endpoint.
220    #[serde(default)]
221    pub vision: bool,
222}
223
224/// Type discriminator for endpoint configuration.
225#[derive(Debug, Deserialize, Clone, PartialEq, Eq, Default)]
226#[serde(rename_all = "lowercase")]
227pub enum EndpointType {
228    /// OpenAI-compatible LLM API.
229    #[default]
230    Llm,
231    /// Model Context Protocol server.
232    Mcp,
233    /// Agent-to-Agent protocol agent.
234    A2a,
235}
236
237/// Pricing configuration for an endpoint.
238#[derive(Debug, Deserialize, Clone, Default)]
239pub struct PricingConfig {
240    /// Cost per 1M input tokens (USD).
241    #[serde(default)]
242    pub input_per_1m_tokens: f64,
243    /// Cost per 1M output tokens (USD).
244    #[serde(default)]
245    pub output_per_1m_tokens: f64,
246    /// Flat cost per call (USD), used for MCP/agent endpoints.
247    #[serde(default)]
248    pub per_call: f64,
249}
250
251/// Budget limits for test execution.
252#[derive(Debug, Deserialize, Clone, Default)]
253pub struct BudgetsConfig {
254    /// Global budget across all tests in the scenario.
255    #[serde(default)]
256    pub global: Option<BudgetDef>,
257    /// Default per-test budget. Individual tests can override.
258    #[serde(default)]
259    pub per_test_default: Option<BudgetDef>,
260}
261
262/// A budget definition with limits and enforcement mode.
263#[derive(Debug, Deserialize, Clone)]
264pub struct BudgetDef {
265    /// Maximum cost in USD.
266    #[serde(default)]
267    pub max_cost: Option<f64>,
268    /// Maximum total tokens (input + output).
269    #[serde(default)]
270    pub max_tokens: Option<u64>,
271    /// Maximum number of calls (LLM, MCP, agent combined).
272    #[serde(default)]
273    pub max_calls: Option<u64>,
274    /// Enforcement mode: `hard` (abort) or `soft` (warn and continue).
275    #[serde(default)]
276    pub enforcement: Option<BudgetEnforcement>,
277}
278
279/// Budget enforcement strategy.
280#[derive(Debug, Deserialize, Clone, PartialEq, Eq)]
281#[serde(rename_all = "lowercase")]
282pub enum BudgetEnforcement {
283    /// Abort the test or run when budget is exceeded.
284    Hard,
285    /// Log a warning but continue execution.
286    Soft,
287}
288
289/// MCP server exposure configuration.
290#[derive(Debug, Deserialize, Clone)]
291pub struct McpServerConfig {
292    /// Whether to enable the embedded MCP server.
293    #[serde(default)]
294    pub enabled: bool,
295    /// Port to listen on.
296    #[serde(default = "default_mcp_port")]
297    pub port: u16,
298}
299
300const fn default_mcp_port() -> u16 {
301    3000
302}
303
304/// A2A agent server exposure configuration.
305#[derive(Debug, Deserialize, Clone)]
306pub struct A2aServerConfig {
307    /// Whether to enable the embedded A2A agent server.
308    #[serde(default)]
309    pub enabled: bool,
310    /// Port to listen on.
311    #[serde(default = "default_a2a_port")]
312    pub port: u16,
313}
314
315const fn default_a2a_port() -> u16 {
316    3100
317}
318
319fn deserialize_headers<'de, D>(deserializer: D) -> Result<HashMap<String, String>, D::Error>
320where
321    D: serde::Deserializer<'de>,
322{
323    let raw: Option<serde_json::Value> = Option::deserialize(deserializer)?;
324    let Some(json) = raw else {
325        return Ok(HashMap::new());
326    };
327    let serde_json::Value::Object(obj) = json else {
328        return Ok(HashMap::new());
329    };
330    Ok(obj
331        .into_iter()
332        .filter_map(|(k, v)| v.as_str().map(|s| (k, s.to_owned())))
333        .collect())
334}
335
336const fn default_auto_navigate() -> bool {
337    true
338}
339
340fn default_layout_ignore_classes() -> Vec<String> {
341    vec![
342        "cdk-visually-hidden".to_owned(),
343        "cdk-describedby-message-container".to_owned(),
344        "cdk-overlay-container".to_owned(),
345    ]
346}
347
348const fn default_temperature() -> f64 {
349    0.0
350}
351
352fn deserialize_model_params<'de, D>(deserializer: D) -> Result<HashMap<String, Value>, D::Error>
353where
354    D: serde::Deserializer<'de>,
355{
356    #[derive(Deserialize)]
357    #[serde(untagged)]
358    enum Raw {
359        Map(HashMap<String, Value>),
360        Table(HashMap<String, Value>),
361    }
362    let raw: Option<Raw> = Option::deserialize(deserializer)?;
363    Ok(match raw {
364        Some(Raw::Map(m) | Raw::Table(m)) => m,
365        None => HashMap::new(),
366    })
367}
368
369/// Reusable assertion definition referenced by name from `assert` steps.
370///
371/// Definitions can either reference a built-in preset via `preset`, supply a
372/// custom LLM `prompt`, or define a **custom preset** by providing both
373/// `system` and `user_template`. Custom presets support the same template
374/// variables as built-in presets: `{url}`, `{title}`, `{content}`,
375/// `{expected_text}`, `{description}`.
376#[derive(Debug, Deserialize, Clone)]
377pub struct AssertDefinition {
378    /// Unique name used to reference this definition from steps.
379    pub name: String,
380    /// Predefined assertion preset name
381    /// (e.g. `no_error_on_page`, `text_visible`).
382    #[serde(default)]
383    pub preset: Option<String>,
384    /// Custom LLM prompt for assertion evaluation.
385    #[serde(default)]
386    pub prompt: Option<String>,
387    /// System prompt for a custom preset.
388    #[serde(default)]
389    pub system: Option<String>,
390    /// User template (with `{placeholders}`) for a custom preset.
391    #[serde(default)]
392    pub user_template: Option<String>,
393    /// Text that the `text_visible` preset checks for, or the
394    /// `{expected_text}` placeholder value for custom presets.
395    #[serde(default)]
396    pub assert_text: Option<String>,
397    /// Agent endpoint to call for this assertion.
398    #[serde(default)]
399    pub agent: Option<String>,
400    /// Agent task template for this assertion.
401    #[serde(default)]
402    pub task_template: Option<String>,
403}
404
405/// A group of steps that form a single test scenario.
406#[derive(Debug, Deserialize, Clone)]
407pub struct TestGroup {
408    /// Human-readable test name.
409    pub name: String,
410    /// Override the global `start_url` for this test.
411    #[serde(default)]
412    pub start_url: Option<String>,
413    /// Override the global `auto_navigate` for this test.
414    #[serde(default)]
415    pub auto_navigate: Option<bool>,
416    /// Override the global `base_url` for this test.
417    #[serde(default)]
418    pub base_url: Option<String>,
419    /// Override the global `timeout_secs` for this test.
420    #[serde(default)]
421    pub timeout_secs: Option<u64>,
422    /// Override the global `browser_headless` for this test.
423    #[serde(default)]
424    pub browser_headless: Option<bool>,
425    /// Override the global viewport width for this test (applied via CDP
426    /// `Emulation.setDeviceMetricsOverride` before the test runs).
427    #[serde(default)]
428    pub viewport_width: Option<u32>,
429    /// Override the global viewport height for this test.
430    #[serde(default)]
431    pub viewport_height: Option<u32>,
432    /// Per-test budget override.
433    #[serde(default)]
434    pub budget: Option<BudgetDef>,
435    /// Endpoint to use for all steps in this test (can be overridden
436    /// per-step).
437    #[serde(default)]
438    pub endpoint: Option<String>,
439    /// Ordered steps to execute.
440    #[serde(default)]
441    pub steps: Vec<TestStep>,
442}
443
444/// A single step in a test. The `kind` field determines which variant is
445/// deserialized and which field constraints apply.
446#[derive(Debug, Deserialize, Clone)]
447#[serde(tag = "kind")]
448pub enum TestStep {
449    /// Navigate the browser to a URL.
450    #[serde(rename = "navigate")]
451    Navigate {
452        /// URL to navigate to (absolute, or relative to the test's
453        /// base URL).
454        url: String,
455        /// Milliseconds to wait after navigation completes.
456        #[serde(default)]
457        wait_after_ms: Option<u64>,
458    },
459
460    /// Click an element described in natural language.
461    #[serde(rename = "click")]
462    Click {
463        /// Natural language description of the element. The LLM resolves
464        /// this to a CSS selector at runtime.
465        target: String,
466        /// Explicit CSS selector override (bypasses LLM resolution).
467        #[serde(default)]
468        selector: Option<String>,
469        /// Milliseconds to wait after the click.
470        #[serde(default)]
471        wait_after_ms: Option<u64>,
472        /// Endpoint to use for LLM element targeting.
473        #[serde(default)]
474        endpoint: Option<String>,
475        /// Idempotent: when the target element is absent the step is
476        /// reported skipped instead of failed (the action was already
477        /// done / not applicable).
478        #[serde(default)]
479        idempotent: bool,
480    },
481
482    /// Type text into an input element.
483    #[serde(rename = "type")]
484    Type {
485        /// Natural language description of the target input element.
486        target: String,
487        /// Text to type into the element.
488        text: String,
489        /// Explicit CSS selector override (bypasses LLM resolution).
490        #[serde(default)]
491        selector: Option<String>,
492        /// Milliseconds to wait after typing.
493        #[serde(default)]
494        wait_after_ms: Option<u64>,
495        /// Endpoint to use for LLM element targeting.
496        #[serde(default)]
497        endpoint: Option<String>,
498        /// Idempotent: when the target element is absent the step is
499        /// reported skipped instead of failed (the action was already
500        /// done / not applicable).
501        #[serde(default)]
502        idempotent: bool,
503    },
504
505    /// Wait for an element to appear on the page.
506    #[serde(rename = "wait")]
507    Wait {
508        /// Natural language description of the element to wait for.
509        target: String,
510        /// Explicit CSS selector override (bypasses LLM resolution).
511        #[serde(default)]
512        selector: Option<String>,
513        /// Wait until the page's visible text contains this substring
514        /// (alternative to `selector`; either or both may be set — both are
515        /// required to hold when both are set).
516        #[serde(default)]
517        text: Option<String>,
518        /// Maximum milliseconds to wait (default: 10000).
519        #[serde(default)]
520        timeout_ms: Option<u64>,
521        /// Endpoint to use for LLM element targeting.
522        #[serde(default)]
523        endpoint: Option<String>,
524        /// Idempotent: when the condition never becomes true within the
525        /// timeout the step is reported skipped instead of failed (the
526        /// condition was not applicable, e.g. already-authenticated
527        /// pages in a viewport matrix).
528        #[serde(default)]
529        idempotent: bool,
530    },
531
532    /// Evaluate an assertion against the current page content.
533    #[serde(rename = "assert")]
534    Assert {
535        /// Reference to a named `[[definitions]]` entry.
536        #[serde(default)]
537        definition: Option<String>,
538        /// Inline predefined assertion preset (e.g. `no_error_on_page`).
539        #[serde(default)]
540        preset: Option<String>,
541        /// Inline custom LLM prompt for assertion evaluation.
542        #[serde(default)]
543        prompt: Option<String>,
544        /// Text that the `text_visible` preset checks for.
545        #[serde(default)]
546        assert_text: Option<String>,
547        /// Endpoint to use for this assertion's LLM call.
548        #[serde(default)]
549        endpoint: Option<String>,
550        /// Attach a screenshot of the current viewport to the assertion so
551        /// the LLM can evaluate visuals (overlaps, clipping, layout).
552        /// Requires the resolved endpoint to declare `vision = true`.
553        #[serde(default)]
554        screenshot: bool,
555    },
556
557    /// Take a screenshot of the current page.
558    #[serde(rename = "screenshot")]
559    Screenshot {
560        /// File path to save the screenshot (default: `screenshot.png`).
561        #[serde(default)]
562        path: Option<String>,
563    },
564
565    /// Call an A2A agent with a task.
566    #[serde(rename = "agent")]
567    Agent {
568        /// Name of the agent endpoint to call.
569        agent: String,
570        /// Task description / prompt for the agent.
571        task: String,
572        /// Optional definition name with a task template.
573        #[serde(default)]
574        definition: Option<String>,
575    },
576
577    /// Call an MCP server tool.
578    #[serde(rename = "mcp")]
579    Mcp {
580        /// Name of the MCP server endpoint.
581        server: String,
582        /// Tool name to invoke on the server.
583        tool: String,
584        /// Tool arguments as JSON.
585        #[serde(default)]
586        args: Option<serde_json::Value>,
587    },
588}