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}
151
152/// A named endpoint definition with pricing.
153#[derive(Debug, Deserialize, Clone, Default)]
154pub struct EndpointConfig {
155    /// Endpoint type: `llm`, `mcp`, or `a2a`.
156    #[serde(rename = "type")]
157    pub endpoint_type: EndpointType,
158    /// Base URL for the endpoint.
159    #[serde(default)]
160    pub url: Option<String>,
161    /// Model name (LLM endpoints only).
162    #[serde(default)]
163    pub model: Option<String>,
164    /// API key / bearer token.
165    #[serde(default)]
166    pub api_key: Option<String>,
167    /// Custom HTTP headers as JSON key-value pairs.
168    #[serde(default, deserialize_with = "deserialize_headers")]
169    pub headers: HashMap<String, String>,
170    /// Pricing configuration.
171    #[serde(default)]
172    pub pricing: Option<PricingConfig>,
173    /// Task types this endpoint serves by default
174    /// (e.g. `["targeting", "assertion"]`).
175    #[serde(default)]
176    pub default_for: Vec<String>,
177    /// Command to launch an MCP server subprocess (stdio transport).
178    #[serde(default)]
179    pub command: Option<String>,
180    /// Arguments for the MCP server command.
181    #[serde(default)]
182    pub args: Vec<String>,
183    /// Whether this LLM endpoint accepts image parts (vision) in addition
184    /// to text. `assert` steps with `screenshot = true` require a vision
185    /// endpoint.
186    #[serde(default)]
187    pub vision: bool,
188}
189
190/// Type discriminator for endpoint configuration.
191#[derive(Debug, Deserialize, Clone, PartialEq, Eq, Default)]
192#[serde(rename_all = "lowercase")]
193pub enum EndpointType {
194    /// OpenAI-compatible LLM API.
195    #[default]
196    Llm,
197    /// Model Context Protocol server.
198    Mcp,
199    /// Agent-to-Agent protocol agent.
200    A2a,
201}
202
203/// Pricing configuration for an endpoint.
204#[derive(Debug, Deserialize, Clone, Default)]
205pub struct PricingConfig {
206    /// Cost per 1M input tokens (USD).
207    #[serde(default)]
208    pub input_per_1m_tokens: f64,
209    /// Cost per 1M output tokens (USD).
210    #[serde(default)]
211    pub output_per_1m_tokens: f64,
212    /// Flat cost per call (USD), used for MCP/agent endpoints.
213    #[serde(default)]
214    pub per_call: f64,
215}
216
217/// Budget limits for test execution.
218#[derive(Debug, Deserialize, Clone, Default)]
219pub struct BudgetsConfig {
220    /// Global budget across all tests in the scenario.
221    #[serde(default)]
222    pub global: Option<BudgetDef>,
223    /// Default per-test budget. Individual tests can override.
224    #[serde(default)]
225    pub per_test_default: Option<BudgetDef>,
226}
227
228/// A budget definition with limits and enforcement mode.
229#[derive(Debug, Deserialize, Clone)]
230pub struct BudgetDef {
231    /// Maximum cost in USD.
232    #[serde(default)]
233    pub max_cost: Option<f64>,
234    /// Maximum total tokens (input + output).
235    #[serde(default)]
236    pub max_tokens: Option<u64>,
237    /// Maximum number of calls (LLM, MCP, agent combined).
238    #[serde(default)]
239    pub max_calls: Option<u64>,
240    /// Enforcement mode: `hard` (abort) or `soft` (warn and continue).
241    #[serde(default)]
242    pub enforcement: Option<BudgetEnforcement>,
243}
244
245/// Budget enforcement strategy.
246#[derive(Debug, Deserialize, Clone, PartialEq, Eq)]
247#[serde(rename_all = "lowercase")]
248pub enum BudgetEnforcement {
249    /// Abort the test or run when budget is exceeded.
250    Hard,
251    /// Log a warning but continue execution.
252    Soft,
253}
254
255/// MCP server exposure configuration.
256#[derive(Debug, Deserialize, Clone)]
257pub struct McpServerConfig {
258    /// Whether to enable the embedded MCP server.
259    #[serde(default)]
260    pub enabled: bool,
261    /// Port to listen on.
262    #[serde(default = "default_mcp_port")]
263    pub port: u16,
264}
265
266const fn default_mcp_port() -> u16 {
267    3000
268}
269
270/// A2A agent server exposure configuration.
271#[derive(Debug, Deserialize, Clone)]
272pub struct A2aServerConfig {
273    /// Whether to enable the embedded A2A agent server.
274    #[serde(default)]
275    pub enabled: bool,
276    /// Port to listen on.
277    #[serde(default = "default_a2a_port")]
278    pub port: u16,
279}
280
281const fn default_a2a_port() -> u16 {
282    3100
283}
284
285fn deserialize_headers<'de, D>(deserializer: D) -> Result<HashMap<String, String>, D::Error>
286where
287    D: serde::Deserializer<'de>,
288{
289    let raw: Option<serde_json::Value> = Option::deserialize(deserializer)?;
290    let Some(json) = raw else {
291        return Ok(HashMap::new());
292    };
293    let serde_json::Value::Object(obj) = json else {
294        return Ok(HashMap::new());
295    };
296    Ok(obj
297        .into_iter()
298        .filter_map(|(k, v)| v.as_str().map(|s| (k, s.to_owned())))
299        .collect())
300}
301
302const fn default_auto_navigate() -> bool {
303    true
304}
305
306const fn default_temperature() -> f64 {
307    0.0
308}
309
310fn deserialize_model_params<'de, D>(deserializer: D) -> Result<HashMap<String, Value>, D::Error>
311where
312    D: serde::Deserializer<'de>,
313{
314    #[derive(Deserialize)]
315    #[serde(untagged)]
316    enum Raw {
317        Map(HashMap<String, Value>),
318        Table(HashMap<String, Value>),
319    }
320    let raw: Option<Raw> = Option::deserialize(deserializer)?;
321    Ok(match raw {
322        Some(Raw::Map(m) | Raw::Table(m)) => m,
323        None => HashMap::new(),
324    })
325}
326
327/// Reusable assertion definition referenced by name from `assert` steps.
328///
329/// Definitions can either reference a built-in preset via `preset`, supply a
330/// custom LLM `prompt`, or define a **custom preset** by providing both
331/// `system` and `user_template`. Custom presets support the same template
332/// variables as built-in presets: `{url}`, `{title}`, `{content}`,
333/// `{expected_text}`, `{description}`.
334#[derive(Debug, Deserialize, Clone)]
335pub struct AssertDefinition {
336    /// Unique name used to reference this definition from steps.
337    pub name: String,
338    /// Predefined assertion preset name
339    /// (e.g. `no_error_on_page`, `text_visible`).
340    #[serde(default)]
341    pub preset: Option<String>,
342    /// Custom LLM prompt for assertion evaluation.
343    #[serde(default)]
344    pub prompt: Option<String>,
345    /// System prompt for a custom preset.
346    #[serde(default)]
347    pub system: Option<String>,
348    /// User template (with `{placeholders}`) for a custom preset.
349    #[serde(default)]
350    pub user_template: Option<String>,
351    /// Text that the `text_visible` preset checks for, or the
352    /// `{expected_text}` placeholder value for custom presets.
353    #[serde(default)]
354    pub assert_text: Option<String>,
355    /// Agent endpoint to call for this assertion.
356    #[serde(default)]
357    pub agent: Option<String>,
358    /// Agent task template for this assertion.
359    #[serde(default)]
360    pub task_template: Option<String>,
361}
362
363/// A group of steps that form a single test scenario.
364#[derive(Debug, Deserialize)]
365pub struct TestGroup {
366    /// Human-readable test name.
367    pub name: String,
368    /// Override the global `start_url` for this test.
369    #[serde(default)]
370    pub start_url: Option<String>,
371    /// Override the global `auto_navigate` for this test.
372    #[serde(default)]
373    pub auto_navigate: Option<bool>,
374    /// Override the global `base_url` for this test.
375    #[serde(default)]
376    pub base_url: Option<String>,
377    /// Override the global `timeout_secs` for this test.
378    #[serde(default)]
379    pub timeout_secs: Option<u64>,
380    /// Override the global `browser_headless` for this test.
381    #[serde(default)]
382    pub browser_headless: Option<bool>,
383    /// Per-test budget override.
384    #[serde(default)]
385    pub budget: Option<BudgetDef>,
386    /// Endpoint to use for all steps in this test (can be overridden
387    /// per-step).
388    #[serde(default)]
389    pub endpoint: Option<String>,
390    /// Ordered steps to execute.
391    #[serde(default)]
392    pub steps: Vec<TestStep>,
393}
394
395/// A single step in a test. The `kind` field determines which variant is
396/// deserialized and which field constraints apply.
397#[derive(Debug, Deserialize)]
398#[serde(tag = "kind")]
399pub enum TestStep {
400    /// Navigate the browser to a URL.
401    #[serde(rename = "navigate")]
402    Navigate {
403        /// URL to navigate to (absolute, or relative to the test's base
404        /// URL).
405        url: String,
406        /// Milliseconds to wait after navigation completes.
407        #[serde(default)]
408        wait_after_ms: Option<u64>,
409    },
410
411    /// Click an element described in natural language.
412    #[serde(rename = "click")]
413    Click {
414        /// Natural language description of the element. The LLM resolves
415        /// this to a CSS selector at runtime.
416        target: String,
417        /// Explicit CSS selector override (bypasses LLM resolution).
418        #[serde(default)]
419        selector: Option<String>,
420        /// Milliseconds to wait after the click.
421        #[serde(default)]
422        wait_after_ms: Option<u64>,
423        /// Endpoint to use for LLM element targeting.
424        #[serde(default)]
425        endpoint: Option<String>,
426    },
427
428    /// Type text into an input element.
429    #[serde(rename = "type")]
430    Type {
431        /// Natural language description of the target input element.
432        target: String,
433        /// Text to type into the element.
434        text: String,
435        /// Explicit CSS selector override (bypasses LLM resolution).
436        #[serde(default)]
437        selector: Option<String>,
438        /// Milliseconds to wait after typing.
439        #[serde(default)]
440        wait_after_ms: Option<u64>,
441        /// Endpoint to use for LLM element targeting.
442        #[serde(default)]
443        endpoint: Option<String>,
444    },
445
446    /// Wait for an element to appear on the page.
447    #[serde(rename = "wait")]
448    Wait {
449        /// Natural language description of the element to wait for.
450        target: String,
451        /// Explicit CSS selector override (bypasses LLM resolution).
452        #[serde(default)]
453        selector: Option<String>,
454        /// Wait until the page's visible text contains this substring
455        /// (alternative to `selector`; either or both may be set — both are
456        /// required to hold when both are set).
457        #[serde(default)]
458        text: Option<String>,
459        /// Maximum milliseconds to wait (default: 10000).
460        #[serde(default)]
461        timeout_ms: Option<u64>,
462        /// Endpoint to use for LLM element targeting.
463        #[serde(default)]
464        endpoint: Option<String>,
465    },
466
467    /// Evaluate an assertion against the current page content.
468    #[serde(rename = "assert")]
469    Assert {
470        /// Reference to a named `[[definitions]]` entry.
471        #[serde(default)]
472        definition: Option<String>,
473        /// Inline predefined assertion preset (e.g. `no_error_on_page`).
474        #[serde(default)]
475        preset: Option<String>,
476        /// Inline custom LLM prompt for assertion evaluation.
477        #[serde(default)]
478        prompt: Option<String>,
479        /// Text that the `text_visible` preset checks for.
480        #[serde(default)]
481        assert_text: Option<String>,
482        /// Endpoint to use for this assertion's LLM call.
483        #[serde(default)]
484        endpoint: Option<String>,
485        /// Attach a screenshot of the current viewport to the assertion so
486        /// the LLM can evaluate visuals (overlaps, clipping, layout).
487        /// Requires the resolved endpoint to declare `vision = true`.
488        #[serde(default)]
489        screenshot: bool,
490    },
491
492    /// Take a screenshot of the current page.
493    #[serde(rename = "screenshot")]
494    Screenshot {
495        /// File path to save the screenshot (default: `screenshot.png`).
496        #[serde(default)]
497        path: Option<String>,
498    },
499
500    /// Call an A2A agent with a task.
501    #[serde(rename = "agent")]
502    Agent {
503        /// Name of the agent endpoint to call.
504        agent: String,
505        /// Task description / prompt for the agent.
506        task: String,
507        /// Optional definition name with a task template.
508        #[serde(default)]
509        definition: Option<String>,
510    },
511
512    /// Call an MCP server tool.
513    #[serde(rename = "mcp")]
514    Mcp {
515        /// Name of the MCP server endpoint.
516        server: String,
517        /// Tool name to invoke on the server.
518        tool: String,
519        /// Tool arguments as JSON.
520        #[serde(default)]
521        args: Option<serde_json::Value>,
522    },
523}