stepproof 0.2.22 → 0.4.0

package/dist/core/types.d.ts

@@ -1,30 +1,52 @@
-export type Provider = 'openai' | 'anthropic';
-export type AssertionType = 'contains' | 'not_contains' | 'regex' | 'json_schema' | 'llm_judge';
+export type Provider = 'openai' | 'anthropic' | 'gemini' | 'ollama';
+export type AssertionType = 'contains' | 'not_contains' | 'regex' | 'json_schema' | 'llm_judge' | 'similarity' | 'sentiment' | 'toxicity' | 'starts_with' | 'ends_with' | 'length' | 'word_count' | 'cost_under' | 'latency_under';
+export interface ConversationTurn {
+    role: 'user' | 'assistant' | 'system';
+    content?: string;
+}
 export interface Assertion {
     type: AssertionType;
-    /** For contains, not_contains, regex */
-    value?: string;
+    /** For contains, not_contains, regex, starts_with, ends_with, cost_under, latency_under */
+    value?: string | number;
     /** For json_schema: path to JSON schema file (relative to scenario file) */
     schema?: string;
     /** For llm_judge: the evaluation prompt */
     prompt?: string;
     /** For llm_judge: the expected response prefix (default: "yes") */
     pass_on?: string;
-    /** For llm_judge: override provider (default: anthropic) */
+    /** For llm_judge, similarity, sentiment, toxicity: override provider (default: anthropic) */
     provider?: Provider;
-    /** For llm_judge: override model (default: claude-haiku or gpt-4o-mini) */
+    /** For llm_judge, similarity, sentiment, toxicity: override model */
     model?: string;
+    /** For similarity: minimum similarity score 0.0-1.0 (default: 0.7) */
+    threshold?: number;
+    /** For toxicity: maximum toxicity score 0.0-1.0 (default: 0.5) */
+    max_score?: number;
+    /** For length, word_count: minimum value */
+    min?: number;
+    /** For length, word_count: maximum value */
+    max?: number;
 }
 export interface Step {
     id: string;
     provider: Provider;
     model: string;
-    /** Prompt template. Use {{variable}} for variables, {{step_id.output}} for prior step outputs. */
-    prompt: string;
+    /** Prompt template. Use {{variable}} for variables, {{step_id.output}} for prior step outputs. Required unless conversation is set. */
+    prompt?: string;
     /** Optional system prompt */
     system?: string;
-    /** Minimum pass rate threshold (0.0–1.0). Default: 0.8 */
+    /** Multi-turn conversation. The last assistant turn without content is where the LLM responds. */
+    conversation?: ConversationTurn[];
+    /** Step IDs this step depends on. Steps without dependencies run in parallel. */
+    depends_on?: string[];
+    /** Condition string — if false, step is skipped. Supports contains, not_contains, matches operators. */
+    if?: string;
+    /** Minimum pass rate threshold (0.0-1.0). Default: 0.8 */
     min_pass_rate?: number;
+    /** Retry up to N times on assertion failure (0 = no retries). Default: 0 */
+    retry?: number;
+    /** Delay in ms between retries. Default: 1000 */
+    retry_delay?: number;
     assertions: Assertion[];
 }
 export interface Scenario {
@@ -45,9 +67,17 @@ export interface StepResult {
     iteration: number;
     output: string;
     passed: boolean;
+    skipped?: boolean;
+    /** Number of retries used before final result (0 = passed on first try) */
+    retriesUsed?: number;
+    /** Dataset row index (0-based) when running in dataset mode */
+    datasetRow?: number;
     assertionResults: AssertionResult[];
     error?: string;
     durationMs: number;
+    inputTokens?: number;
+    outputTokens?: number;
+    costUsd?: number;
 }
 export interface StepSummary {
     stepId: string;
@@ -57,6 +87,23 @@ export interface StepSummary {
     passRate: number;
     minPassRate: number;
     belowThreshold: boolean;
+    avgDurationMs: number;
+    totalCostUsd: number;
+    avgCostUsd: number;
+    /** Number of step results that required at least one retry */
+    retriedCount?: number;
+}
+/** Per-row result summary when running in dataset mode */
+export interface DatasetRowSummary {
+    rowIndex: number;
+    /** First few columns from the row for display */
+    rowPreview: Record<string, string>;
+    allStepsPassed: boolean;
+    stepResults: Array<{
+        stepId: string;
+        passed: boolean;
+        passRate: number;
+    }>;
 }
 export interface ScenarioReport {
     scenarioName: string;
@@ -67,5 +114,12 @@ export interface ScenarioReport {
     steps: StepSummary[];
     allPassed: boolean;
     results: StepResult[];
+    /** Dataset mode info — present when run with --dataset */
+    dataset?: {
+        path: string;
+        totalRows: number;
+        rowsPassed: number;
+        rowSummaries: DatasetRowSummary[];
+    };
 }
 //# sourceMappingURL=types.d.ts.map

package/dist/core/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/core/types.ts"],"names":[],"mappings":"AAAA,MAAM,MAAM,QAAQ,GAAG,QAAQ,GAAG,WAAW,CAAC;AAE9C,MAAM,MAAM,aAAa,GAAG,UAAU,GAAG,cAAc,GAAG,OAAO,GAAG,aAAa,GAAG,WAAW,CAAC;AAEhG,MAAM,WAAW,SAAS;IACxB,IAAI,EAAE,aAAa,CAAC;IACpB,wCAAwC;IACxC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,4EAA4E;IAC5E,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,2CAA2C;IAC3C,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,mEAAmE;IACnE,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,4DAA4D;IAC5D,QAAQ,CAAC,EAAE,QAAQ,CAAC;IACpB,2EAA2E;IAC3E,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,IAAI;IACnB,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,QAAQ,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,kGAAkG;IAClG,MAAM,EAAE,MAAM,CAAC;IACf,6BAA6B;IAC7B,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,0DAA0D;IAC1D,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,UAAU,EAAE,SAAS,EAAE,CAAC;CACzB;AAED,MAAM,WAAW,QAAQ;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,+CAA+C;IAC/C,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,iDAAiD;IACjD,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACnC,KAAK,EAAE,IAAI,EAAE,CAAC;CACf;AAED,MAAM,WAAW,eAAe;IAC9B,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,OAAO,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,UAAU;IACzB,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,OAAO,CAAC;IAChB,gBAAgB,EAAE,eAAe,EAAE,CAAC;IACpC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,WAAW;IAC1B,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,cAAc,EAAE,OAAO,CAAC;CACzB;AAED,MAAM,WAAW,cAAc;IAC7B,YAAY,EAAE,MAAM,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,MAAM,CAAC;IACnB,KAAK,EAAE,WAAW,EAAE,CAAC;IACrB,SAAS,EAAE,OAAO,CAAC;IACnB,OAAO,EAAE,UAAU,EAAE,CAAC;CACvB"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/core/types.ts"],"names":[],"mappings":"AAAA,MAAM,MAAM,QAAQ,GAAG,QAAQ,GAAG,WAAW,GAAG,QAAQ,GAAG,QAAQ,CAAC;AAEpE,MAAM,MAAM,aAAa,GACrB,UAAU,GACV,cAAc,GACd,OAAO,GACP,aAAa,GACb,WAAW,GACX,YAAY,GACZ,WAAW,GACX,UAAU,GACV,aAAa,GACb,WAAW,GACX,QAAQ,GACR,YAAY,GACZ,YAAY,GACZ,eAAe,CAAC;AAEpB,MAAM,WAAW,gBAAgB;IAC/B,IAAI,EAAE,MAAM,GAAG,WAAW,GAAG,QAAQ,CAAC;IACtC,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,SAAS;IACxB,IAAI,EAAE,aAAa,CAAC;IACpB,2FAA2F;IAC3F,KAAK,CAAC,EAAE,MAAM,GAAG,MAAM,CAAC;IACxB,4EAA4E;IAC5E,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,2CAA2C;IAC3C,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,mEAAmE;IACnE,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,6FAA6F;IAC7F,QAAQ,CAAC,EAAE,QAAQ,CAAC;IACpB,qEAAqE;IACrE,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,sEAAsE;IACtE,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,kEAAkE;IAClE,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,4CAA4C;IAC5C,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,4CAA4C;IAC5C,GAAG,CAAC,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,IAAI;IACnB,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,QAAQ,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,uIAAuI;IACvI,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,6BAA6B;IAC7B,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,kGAAkG;IAClG,YAAY,CAAC,EAAE,gBAAgB,EAAE,CAAC;IAClC,iFAAiF;IACjF,UAAU,CAAC,EAAE,MAAM,EAAE,CAAC;IACtB,wGAAwG;IACxG,EAAE,CAAC,EAAE,MAAM,CAAC;IACZ,0DAA0D;IAC1D,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,4EAA4E;IAC5E,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,iDAAiD;IACjD,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,UAAU,EAAE,SAAS,EAAE,CAAC;CACzB;AAED,MAAM,WAAW,QAAQ;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,+CAA+C;IAC/C,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,iDAAiD;IACjD,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACnC,KAAK,EAAE,IAAI,EAAE,CAAC;CACf;AAED,MAAM,WAAW,eAAe;IAC9B,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,OAAO,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,UAAU;IACzB,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,OAAO,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,2EAA2E;IAC3E,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,+DAA+D;IAC/D,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,gBAAgB,EAAE,eAAe,EAAE,CAAC;IACpC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,WAAW;IAC1B,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,cAAc,EAAE,OAAO,CAAC;IACxB,aAAa,EAAE,MAAM,CAAC;IACtB,YAAY,EAAE,MAAM,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;IACnB,8DAA8D;IAC9D,YAAY,CAAC,EAAE,MAAM,CAAC;CACvB;AAED,0DAA0D;AAC1D,MAAM,WAAW,iBAAiB;IAChC,QAAQ,EAAE,MAAM,CAAC;IACjB,iDAAiD;IACjD,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACnC,cAAc,EAAE,OAAO,CAAC;IACxB,WAAW,EAAE,KAAK,CAAC;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,OAAO,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;CAC3E;AAED,MAAM,WAAW,cAAc;IAC7B,YAAY,EAAE,MAAM,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,MAAM,CAAC;IACnB,KAAK,EAAE,WAAW,EAAE,CAAC;IACrB,SAAS,EAAE,OAAO,CAAC;IACnB,OAAO,EAAE,UAAU,EAAE,CAAC;IACtB,0DAA0D;IAC1D,OAAO,CAAC,EAAE;QACR,IAAI,EAAE,MAAM,CAAC;QACb,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,EAAE,MAAM,CAAC;QACnB,YAAY,EAAE,iBAAiB,EAAE,CAAC;KACnC,CAAC;CACH"}

package/dist/dataset.d.ts

@@ -0,0 +1,6 @@
+/**
+ * Lightweight CSV parser. Handles quoted fields (including escaped quotes "").
+ * O(n) where n = file size.
+ */
+export declare function loadDataset(csvPath: string): Array<Record<string, string>>;
+//# sourceMappingURL=dataset.d.ts.map

package/dist/dataset.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"dataset.d.ts","sourceRoot":"","sources":["../src/dataset.ts"],"names":[],"mappings":"AAEA;;;GAGG;AACH,wBAAgB,WAAW,CAAC,OAAO,EAAE,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAkC1E"}

package/dist/dataset.js

@@ -0,0 +1,108 @@
+import * as fs from 'node:fs';
+/**
+ * Lightweight CSV parser. Handles quoted fields (including escaped quotes "").
+ * O(n) where n = file size.
+ */
+export function loadDataset(csvPath) {
+    let content;
+    try {
+        content = fs.readFileSync(csvPath, 'utf-8');
+    }
+    catch (e) {
+        throw new Error(`Cannot read dataset file: ${csvPath}`);
+    }
+    const lines = parseCSVLines(content);
+    if (lines.length < 2) {
+        throw new Error(`Dataset file must have a header row and at least one data row: ${csvPath}`);
+    }
+    const headers = lines[0];
+    const rows = [];
+    for (let i = 1; i < lines.length; i++) {
+        const fields = lines[i];
+        if (fields.length === 1 && fields[0] === '')
+            continue; // skip blank lines
+        if (fields.length !== headers.length) {
+            throw new Error(`Dataset row ${i} has ${fields.length} fields, expected ${headers.length}`);
+        }
+        const row = {};
+        for (let j = 0; j < headers.length; j++) {
+            row[headers[j]] = fields[j];
+        }
+        rows.push(row);
+    }
+    if (rows.length === 0) {
+        throw new Error(`Dataset file has no data rows: ${csvPath}`);
+    }
+    return rows;
+}
+/**
+ * Parse CSV content into array of field arrays.
+ * Handles: quoted fields, embedded commas, embedded newlines in quotes, escaped quotes ("").
+ */
+function parseCSVLines(content) {
+    const results = [];
+    let currentRow = [];
+    let currentField = '';
+    let inQuotes = false;
+    let i = 0;
+    while (i < content.length) {
+        const ch = content[i];
+        if (inQuotes) {
+            if (ch === '"') {
+                if (i + 1 < content.length && content[i + 1] === '"') {
+                    // Escaped quote
+                    currentField += '"';
+                    i += 2;
+                }
+                else {
+                    // End of quoted field
+                    inQuotes = false;
+                    i++;
+                }
+            }
+            else {
+                currentField += ch;
+                i++;
+            }
+        }
+        else {
+            if (ch === '"') {
+                inQuotes = true;
+                i++;
+            }
+            else if (ch === ',') {
+                currentRow.push(currentField);
+                currentField = '';
+                i++;
+            }
+            else if (ch === '\r') {
+                // Handle \r\n and lone \r
+                currentRow.push(currentField);
+                currentField = '';
+                results.push(currentRow);
+                currentRow = [];
+                i++;
+                if (i < content.length && content[i] === '\n')
+                    i++;
+            }
+            else if (ch === '\n') {
+                currentRow.push(currentField);
+                currentField = '';
+                results.push(currentRow);
+                currentRow = [];
+                i++;
+            }
+            else {
+                currentField += ch;
+                i++;
+            }
+        }
+    }
+    // Flush last field/row
+    if (currentField || currentRow.length > 0) {
+        currentRow.push(currentField);
+        results.push(currentRow);
+    }
+    return results;
+}
+//# sourceMappingURL=dataset.js.map

package/dist/dataset.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"dataset.js","sourceRoot":"","sources":["../src/dataset.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,MAAM,SAAS,CAAC;AAE9B;;;GAGG;AACH,MAAM,UAAU,WAAW,CAAC,OAAe;IACzC,IAAI,OAAe,CAAC;IACpB,IAAI,CAAC;QACH,OAAO,GAAG,EAAE,CAAC,YAAY,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;IAC9C,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACX,MAAM,IAAI,KAAK,CAAC,6BAA6B,OAAO,EAAE,CAAC,CAAC;IAC1D,CAAC;IAED,MAAM,KAAK,GAAG,aAAa,CAAC,OAAO,CAAC,CAAC;IACrC,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACrB,MAAM,IAAI,KAAK,CAAC,kEAAkE,OAAO,EAAE,CAAC,CAAC;IAC/F,CAAC;IAED,MAAM,OAAO,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;IACzB,MAAM,IAAI,GAAkC,EAAE,CAAC;IAE/C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACtC,MAAM,MAAM,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;QACxB,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,IAAI,MAAM,CAAC,CAAC,CAAC,KAAK,EAAE;YAAE,SAAS,CAAC,mBAAmB;QAC1E,IAAI,MAAM,CAAC,MAAM,KAAK,OAAO,CAAC,MAAM,EAAE,CAAC;YACrC,MAAM,IAAI,KAAK,CAAC,eAAe,CAAC,QAAQ,MAAM,CAAC,MAAM,qBAAqB,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;QAC9F,CAAC;QACD,MAAM,GAAG,GAA2B,EAAE,CAAC;QACvC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACxC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;QAC9B,CAAC;QACD,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IACjB,CAAC;IAED,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACtB,MAAM,IAAI,KAAK,CAAC,kCAAkC,OAAO,EAAE,CAAC,CAAC;IAC/D,CAAC;IAED,OAAO,IAAI,CAAC;AACd,CAAC;AAED;;;GAGG;AACH,SAAS,aAAa,CAAC,OAAe;IACpC,MAAM,OAAO,GAAe,EAAE,CAAC;IAC/B,IAAI,UAAU,GAAa,EAAE,CAAC;IAC9B,IAAI,YAAY,GAAG,EAAE,CAAC;IACtB,IAAI,QAAQ,GAAG,KAAK,CAAC;IACrB,IAAI,CAAC,GAAG,CAAC,CAAC;IAEV,OAAO,CAAC,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC;QAC1B,MAAM,EAAE,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;QAEtB,IAAI,QAAQ,EAAE,CAAC;YACb,IAAI,EAAE,KAAK,GAAG,EAAE,CAAC;gBACf,IAAI,CAAC,GAAG,CAAC,GAAG,OAAO,CAAC,MAAM,IAAI,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,GAAG,EAAE,CAAC;oBACrD,gBAAgB;oBAChB,YAAY,IAAI,GAAG,CAAC;oBACpB,CAAC,IAAI,CAAC,CAAC;gBACT,CAAC;qBAAM,CAAC;oBACN,sBAAsB;oBACtB,QAAQ,GAAG,KAAK,CAAC;oBACjB,CAAC,EAAE,CAAC;gBACN,CAAC;YACH,CAAC;iBAAM,CAAC;gBACN,YAAY,IAAI,EAAE,CAAC;gBACnB,CAAC,EAAE,CAAC;YACN,CAAC;QACH,CAAC;aAAM,CAAC;YACN,IAAI,EAAE,KAAK,GAAG,EAAE,CAAC;gBACf,QAAQ,GAAG,IAAI,CAAC;gBAChB,CAAC,EAAE,CAAC;YACN,CAAC;iBAAM,IAAI,EAAE,KAAK,GAAG,EAAE,CAAC;gBACtB,UAAU,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;gBAC9B,YAAY,GAAG,EAAE,CAAC;gBAClB,CAAC,EAAE,CAAC;YACN,CAAC;iBAAM,IAAI,EAAE,KAAK,IAAI,EAAE,CAAC;gBACvB,0BAA0B;gBAC1B,UAAU,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;gBAC9B,YAAY,GAAG,EAAE,CAAC;gBAClB,OAAO,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;gBACzB,UAAU,GAAG,EAAE,CAAC;gBAChB,CAAC,EAAE,CAAC;gBACJ,IAAI,CAAC,GAAG,OAAO,CAAC,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,KAAK,IAAI;oBAAE,CAAC,EAAE,CAAC;YACrD,CAAC;iBAAM,IAAI,EAAE,KAAK,IAAI,EAAE,CAAC;gBACvB,UAAU,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;gBAC9B,YAAY,GAAG,EAAE,CAAC;gBAClB,OAAO,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;gBACzB,UAAU,GAAG,EAAE,CAAC;gBAChB,CAAC,EAAE,CAAC;YACN,CAAC;iBAAM,CAAC;gBACN,YAAY,IAAI,EAAE,CAAC;gBACnB,CAAC,EAAE,CAAC;YACN,CAAC;QACH,CAAC;IACH,CAAC;IAED,uBAAuB;IACvB,IAAI,YAAY,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC1C,UAAU,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QAC9B,OAAO,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;IAC3B,CAAC;IAED,OAAO,OAAO,CAAC;AACjB,CAAC"}

package/dist/reporters/github-comment.d.ts

@@ -0,0 +1,8 @@
+import type { ScenarioReport } from '../core/types.js';
+import type { BaselineComparison } from '../baseline.js';
+/**
+ * Format a ScenarioReport as a GitHub PR comment in markdown.
+ * Includes a marker comment so the action can update existing comments.
+ */
+export declare function formatPRComment(report: ScenarioReport, baselineComparison?: BaselineComparison): string;
+//# sourceMappingURL=github-comment.d.ts.map

package/dist/reporters/github-comment.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"github-comment.d.ts","sourceRoot":"","sources":["../../src/reporters/github-comment.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAe,MAAM,kBAAkB,CAAC;AACpE,OAAO,KAAK,EAAE,kBAAkB,EAAkB,MAAM,gBAAgB,CAAC;AA6BzE;;;GAGG;AACH,wBAAgB,eAAe,CAC7B,MAAM,EAAE,cAAc,EACtB,kBAAkB,CAAC,EAAE,kBAAkB,GACtC,MAAM,CA0FR"}

package/dist/reporters/github-comment.js

@@ -0,0 +1,114 @@
+function formatDuration(ms) {
+    if (ms < 1000)
+        return `${ms}ms`;
+    if (ms < 60_000)
+        return `${(ms / 1000).toFixed(1)}s`;
+    const mins = Math.floor(ms / 60_000);
+    const secs = ((ms % 60_000) / 1000).toFixed(0);
+    return `${mins}m ${secs}s`;
+}
+function formatCost(usd) {
+    if (usd === 0)
+        return '$0.00';
+    if (usd < 0.01)
+        return `$${usd.toFixed(4)}`;
+    return `$${usd.toFixed(2)}`;
+}
+function baselineCell(step, baselineMap) {
+    const cmp = baselineMap.get(step.stepId);
+    if (!cmp)
+        return '—';
+    const deltaPct = (cmp.delta * 100).toFixed(1);
+    if (cmp.delta > 0)
+        return `↑ +${deltaPct}%`;
+    if (cmp.delta < 0)
+        return `↓ ${deltaPct}%`;
+    return '→ same';
+}
+function statusEmoji(belowThreshold) {
+    return belowThreshold ? '❌ Fail' : '✅ Pass';
+}
+/**
+ * Format a ScenarioReport as a GitHub PR comment in markdown.
+ * Includes a marker comment so the action can update existing comments.
+ */
+export function formatPRComment(report, baselineComparison) {
+    const baselineMap = new Map();
+    if (baselineComparison) {
+        for (const s of baselineComparison.steps) {
+            baselineMap.set(s.stepId, s);
+        }
+    }
+    const hasBaseline = baselineMap.size > 0;
+    const totalCost = report.steps.reduce((sum, s) => sum + s.totalCostUsd, 0);
+    const lines = [];
+    // Marker for upsert
+    lines.push('<!-- stepproof-results -->');
+    lines.push(`## Stepproof Results: ${report.scenarioName}`);
+    lines.push('');
+    // Table header
+    if (hasBaseline) {
+        lines.push('| Step | Pass Rate | Threshold | Status | Baseline |');
+        lines.push('|------|-----------|-----------|--------|----------|');
+    }
+    else {
+        lines.push('| Step | Pass Rate | Threshold | Status |');
+        lines.push('|------|-----------|-----------|--------|');
+    }
+    // Table rows
+    for (const step of report.steps) {
+        const pct = (step.passRate * 100).toFixed(1);
+        const threshold = (step.minPassRate * 100).toFixed(0);
+        const passRateCell = `${pct}% (${step.passes}/${step.totalRuns})`;
+        const status = statusEmoji(step.belowThreshold);
+        if (hasBaseline) {
+            const bl = baselineCell(step, baselineMap);
+            lines.push(`| ${step.stepId} | ${passRateCell} | ${threshold}% | ${status} | ${bl} |`);
+        }
+        else {
+            lines.push(`| ${step.stepId} | ${passRateCell} | ${threshold}% | ${status} |`);
+        }
+    }
+    lines.push('');
+    // Summary line
+    const overallStatus = report.allPassed ? '✅ All steps passed' : '❌ Some steps failed';
+    const costStr = totalCost > 0 ? ` | 💰 ${formatCost(totalCost)}` : '';
+    lines.push(`**Overall: ${overallStatus}** | ⏱ ${formatDuration(report.durationMs)}${costStr}`);
+    // Regression warning
+    if (baselineComparison?.hasRegression) {
+        const regSteps = baselineComparison.steps.filter(s => s.regression);
+        lines.push('');
+        lines.push(`> ⚠️ **Regression detected** in ${regSteps.map(s => `\`${s.stepId}\``).join(', ')}`);
+    }
+    // Iteration details in collapsible section
+    if (report.results.length > 0) {
+        lines.push('');
+        lines.push('<details><summary>View iteration details</summary>');
+        lines.push('');
+        // Group results by step
+        const byStep = new Map();
+        for (const r of report.results) {
+            const arr = byStep.get(r.stepId) ?? [];
+            arr.push(r);
+            byStep.set(r.stepId, arr);
+        }
+        for (const [stepId, results] of byStep) {
+            lines.push(`### ${stepId}`);
+            lines.push('');
+            lines.push('| Iteration | Passed | Duration | Assertions |');
+            lines.push('|-----------|--------|----------|------------|');
+            for (const r of results) {
+                const passIcon = r.passed ? '✅' : '❌';
+                const dur = `${(r.durationMs / 1000).toFixed(1)}s`;
+                const assertions = r.assertionResults
+                    .map(a => `${a.passed ? '✅' : '❌'} ${a.type}`)
+                    .join(', ');
+                lines.push(`| ${r.iteration} | ${passIcon} | ${dur} | ${assertions} |`);
+            }
+            lines.push('');
+        }
+        lines.push('</details>');
+    }
+    return lines.join('\n');
+}
+//# sourceMappingURL=github-comment.js.map

package/dist/reporters/github-comment.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"github-comment.js","sourceRoot":"","sources":["../../src/reporters/github-comment.ts"],"names":[],"mappings":"AAGA,SAAS,cAAc,CAAC,EAAU;IAChC,IAAI,EAAE,GAAG,IAAI;QAAE,OAAO,GAAG,EAAE,IAAI,CAAC;IAChC,IAAI,EAAE,GAAG,MAAM;QAAE,OAAO,GAAG,CAAC,EAAE,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;IACrD,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,EAAE,GAAG,MAAM,CAAC,CAAC;IACrC,MAAM,IAAI,GAAG,CAAC,CAAC,EAAE,GAAG,MAAM,CAAC,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;IAC/C,OAAO,GAAG,IAAI,KAAK,IAAI,GAAG,CAAC;AAC7B,CAAC;AAED,SAAS,UAAU,CAAC,GAAW;IAC7B,IAAI,GAAG,KAAK,CAAC;QAAE,OAAO,OAAO,CAAC;IAC9B,IAAI,GAAG,GAAG,IAAI;QAAE,OAAO,IAAI,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC;IAC5C,OAAO,IAAI,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC;AAC9B,CAAC;AAED,SAAS,YAAY,CAAC,IAAiB,EAAE,WAAwC;IAC/E,MAAM,GAAG,GAAG,WAAW,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IACzC,IAAI,CAAC,GAAG;QAAE,OAAO,GAAG,CAAC;IACrB,MAAM,QAAQ,GAAG,CAAC,GAAG,CAAC,KAAK,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;IAC9C,IAAI,GAAG,CAAC,KAAK,GAAG,CAAC;QAAE,OAAO,MAAM,QAAQ,GAAG,CAAC;IAC5C,IAAI,GAAG,CAAC,KAAK,GAAG,CAAC;QAAE,OAAO,KAAK,QAAQ,GAAG,CAAC;IAC3C,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,SAAS,WAAW,CAAC,cAAuB;IAC1C,OAAO,cAAc,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC;AAC9C,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAC7B,MAAsB,EACtB,kBAAuC;IAEvC,MAAM,WAAW,GAAG,IAAI,GAAG,EAA0B,CAAC;IACtD,IAAI,kBAAkB,EAAE,CAAC;QACvB,KAAK,MAAM,CAAC,IAAI,kBAAkB,CAAC,KAAK,EAAE,CAAC;YACzC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;QAC/B,CAAC;IACH,CAAC;IAED,MAAM,WAAW,GAAG,WAAW,CAAC,IAAI,GAAG,CAAC,CAAC;IACzC,MAAM,SAAS,GAAG,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC;IAE3E,MAAM,KAAK,GAAa,EAAE,CAAC;IAE3B,oBAAoB;IACpB,KAAK,CAAC,IAAI,CAAC,4BAA4B,CAAC,CAAC;IACzC,KAAK,CAAC,IAAI,CAAC,yBAAyB,MAAM,CAAC,YAAY,EAAE,CAAC,CAAC;IAC3D,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAEf,eAAe;IACf,IAAI,WAAW,EAAE,CAAC;QAChB,KAAK,CAAC,IAAI,CAAC,sDAAsD,CAAC,CAAC;QACnE,KAAK,CAAC,IAAI,CAAC,sDAAsD,CAAC,CAAC;IACrE,CAAC;SAAM,CAAC;QACN,KAAK,CAAC,IAAI,CAAC,2CAA2C,CAAC,CAAC;QACxD,KAAK,CAAC,IAAI,CAAC,2CAA2C,CAAC,CAAC;IAC1D,CAAC;IAED,aAAa;IACb,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;QAChC,MAAM,GAAG,GAAG,CAAC,IAAI,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC7C,MAAM,SAAS,GAAG,CAAC,IAAI,CAAC,WAAW,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QACtD,MAAM,YAAY,GAAG,GAAG,GAAG,MAAM,IAAI,CAAC,MAAM,IAAI,IAAI,CAAC,SAAS,GAAG,CAAC;QAClE,MAAM,MAAM,GAAG,WAAW,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;QAEhD,IAAI,WAAW,EAAE,CAAC;YAChB,MAAM,EAAE,GAAG,YAAY,CAAC,IAAI,EAAE,WAAW,CAAC,CAAC;YAC3C,KAAK,CAAC,IAAI,CAAC,KAAK,IAAI,CAAC,MAAM,MAAM,YAAY,MAAM,SAAS,OAAO,MAAM,MAAM,EAAE,IAAI,CAAC,CAAC;QACzF,CAAC;aAAM,CAAC;YACN,KAAK,CAAC,IAAI,CAAC,KAAK,IAAI,CAAC,MAAM,MAAM,YAAY,MAAM,SAAS,OAAO,MAAM,IAAI,CAAC,CAAC;QACjF,CAAC;IACH,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAEf,eAAe;IACf,MAAM,aAAa,GAAG,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,qBAAqB,CAAC;IACtF,MAAM,OAAO,GAAG,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,UAAU,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IACtE,KAAK,CAAC,IAAI,CAAC,cAAc,aAAa,UAAU,cAAc,CAAC,MAAM,CAAC,UAAU,CAAC,GAAG,OAAO,EAAE,CAAC,CAAC;IAE/F,qBAAqB;IACrB,IAAI,kBAAkB,EAAE,aAAa,EAAE,CAAC;QACtC,MAAM,QAAQ,GAAG,kBAAkB,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC;QACpE,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,mCAAmC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACnG,CAAC;IAED,2CAA2C;IAC3C,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC9B,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,oDAAoD,CAAC,CAAC;QACjE,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAEf,wBAAwB;QACxB,MAAM,MAAM,GAAG,IAAI,GAAG,EAAiC,CAAC;QACxD,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;YAC/B,MAAM,GAAG,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;YACvC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YACZ,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;QAC5B,CAAC;QAED,KAAK,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,IAAI,MAAM,EAAE,CAAC;YACvC,KAAK,CAAC,IAAI,CAAC,OAAO,MAAM,EAAE,CAAC,CAAC;YAC5B,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YACf,KAAK,CAAC,IAAI,CAAC,gDAAgD,CAAC,CAAC;YAC7D,KAAK,CAAC,IAAI,CAAC,gDAAgD,CAAC,CAAC;YAC7D,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;gBACxB,MAAM,QAAQ,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;gBACtC,MAAM,GAAG,GAAG,GAAG,CAAC,CAAC,CAAC,UAAU,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;gBACnD,MAAM,UAAU,GAAG,CAAC,CAAC,gBAAgB;qBAClC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC,IAAI,EAAE,CAAC;qBAC7C,IAAI,CAAC,IAAI,CAAC,CAAC;gBACd,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,SAAS,MAAM,QAAQ,MAAM,GAAG,MAAM,UAAU,IAAI,CAAC,CAAC;YAC1E,CAAC;YACD,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACjB,CAAC;QAED,KAAK,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;IAC3B,CAAC;IAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC"}

package/dist/reporters/html-reporter.d.ts

@@ -0,0 +1,3 @@
+import type { ScenarioReport } from '../core/types.js';
+export declare function generateHtmlReport(report: ScenarioReport): string;
+//# sourceMappingURL=html-reporter.d.ts.map

package/dist/reporters/html-reporter.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"html-reporter.d.ts","sourceRoot":"","sources":["../../src/reporters/html-reporter.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAA2B,MAAM,kBAAkB,CAAC;AAEhF,wBAAgB,kBAAkB,CAAC,MAAM,EAAE,cAAc,GAAG,MAAM,CAuFjE"}

package/dist/reporters/html-reporter.js

@@ -0,0 +1,152 @@
+export function generateHtmlReport(report) {
+    const reportJson = JSON.stringify(report);
+    const passColor = '#22c55e';
+    const failColor = '#ef4444';
+    const verdictColor = report.allPassed ? passColor : failColor;
+    const verdictText = report.allPassed ? 'PASSED' : 'FAILED';
+    return `<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="UTF-8">
+<meta name="viewport" content="width=device-width, initial-scale=1.0">
+<title>Stepproof — ${escHtml(report.scenarioName)}</title>
+<style>
+  *, *::before, *::after { box-sizing: border-box; margin: 0; padding: 0; }
+  body { font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif; background: #0f172a; color: #e2e8f0; line-height: 1.6; padding: 2rem; max-width: 900px; margin: 0 auto; }
+  h1 { font-size: 1.5rem; font-weight: 600; margin-bottom: 0.25rem; }
+  .meta { color: #94a3b8; font-size: 0.875rem; margin-bottom: 1.5rem; }
+  .meta span { margin-right: 1.5rem; }
+  .verdict { display: inline-block; padding: 0.25rem 0.75rem; border-radius: 4px; font-weight: 700; font-size: 0.875rem; background: ${verdictColor}20; color: ${verdictColor}; border: 1px solid ${verdictColor}40; margin-bottom: 1.5rem; }
+  .card { background: #1e293b; border-radius: 8px; padding: 1.25rem; margin-bottom: 1rem; border: 1px solid #334155; }
+  .step-header { display: flex; align-items: center; justify-content: space-between; cursor: pointer; user-select: none; }
+  .step-header:hover { opacity: 0.85; }
+  .step-name { font-weight: 600; font-size: 1.05rem; }
+  .step-icon { font-size: 1.1rem; margin-right: 0.5rem; }
+  .step-rate { font-size: 0.875rem; color: #94a3b8; }
+  .bar-container { background: #334155; border-radius: 4px; height: 8px; margin: 0.75rem 0; overflow: hidden; }
+  .bar-fill { height: 100%; border-radius: 4px; transition: width 0.3s ease; }
+  .bar-pass { background: ${passColor}; }
+  .bar-fail { background: ${failColor}; }
+  .step-stats { display: flex; gap: 1.5rem; font-size: 0.8rem; color: #94a3b8; }
+  .iterations { display: none; margin-top: 1rem; border-top: 1px solid #334155; padding-top: 0.75rem; }
+  .iterations.open { display: block; }
+  .iter-row { display: flex; align-items: center; gap: 0.75rem; padding: 0.4rem 0; font-size: 0.8rem; border-bottom: 1px solid #1e293b; }
+  .iter-row:last-child { border-bottom: none; }
+  .iter-badge { display: inline-block; width: 18px; height: 18px; border-radius: 50%; text-align: center; line-height: 18px; font-size: 0.65rem; font-weight: 700; color: #fff; flex-shrink: 0; }
+  .iter-pass { background: ${passColor}; }
+  .iter-fail { background: ${failColor}; }
+  .iter-assertions { color: #94a3b8; flex: 1; }
+  .iter-duration { color: #64748b; white-space: nowrap; }
+  .iter-output { margin-top: 0.25rem; font-family: 'SF Mono', 'Fira Code', monospace; font-size: 0.75rem; color: #cbd5e1; background: #0f172a; padding: 0.5rem; border-radius: 4px; max-height: 120px; overflow-y: auto; white-space: pre-wrap; word-break: break-word; display: none; }
+  .iter-output.open { display: block; }
+  .iter-toggle { cursor: pointer; color: #60a5fa; font-size: 0.75rem; }
+  .iter-toggle:hover { text-decoration: underline; }
+  .assertion-line { font-size: 0.75rem; padding-left: 1.5rem; }
+  .assertion-pass { color: ${passColor}; }
+  .assertion-fail { color: ${failColor}; }
+  .summary-bar { display: flex; gap: 1.5rem; margin-top: 1rem; font-size: 0.8rem; color: #94a3b8; }
+  .chevron { transition: transform 0.2s; display: inline-block; }
+  .chevron.open { transform: rotate(90deg); }
+  @media (max-width: 640px) { body { padding: 1rem; } .step-stats { flex-wrap: wrap; gap: 0.75rem; } }
+</style>
+</head>
+<body>
+
+<h1>${escHtml(report.scenarioName)}</h1>
+<div class="meta">
+  <span>${new Date(report.startedAt).toLocaleString()}</span>
+  <span>${formatDuration(report.durationMs)}</span>
+  <span>${report.iterations} iteration${report.iterations === 1 ? '' : 's'}</span>
+</div>
+<div class="verdict">${verdictText}</div>
+
+${report.steps.map((step, i) => renderStep(step, report.results.filter(r => r.stepId === step.stepId), i)).join('\n')}
+
+<div class="summary-bar">
+  <span>${report.steps.filter(s => !s.belowThreshold).length}/${report.steps.length} steps passed</span>
+  <span>Total: ${formatDuration(report.durationMs)}</span>
+</div>
+
+<script>
+const REPORT = ${reportJson};
+
+function toggleIterations(idx) {
+  const el = document.getElementById('iters-' + idx);
+  const chev = document.getElementById('chev-' + idx);
+  if (el) { el.classList.toggle('open'); }
+  if (chev) { chev.classList.toggle('open'); }
+}
+
+function toggleOutput(id) {
+  const el = document.getElementById(id);
+  if (el) { el.classList.toggle('open'); }
+}
+</script>
+</body>
+</html>`;
+}
+function renderStep(step, results, index) {
+    const pct = (step.passRate * 100).toFixed(1);
+    const threshPct = (step.minPassRate * 100).toFixed(0);
+    const icon = step.belowThreshold ? '&#10005;' : '&#10003;';
+    const iconColor = step.belowThreshold ? '#ef4444' : '#22c55e';
+    const barColor = step.belowThreshold ? 'bar-fail' : 'bar-pass';
+    const avgDur = step.avgDurationMs ? formatDuration(step.avgDurationMs) : '';
+    const costStr = step.totalCostUsd > 0 ? `$${step.totalCostUsd.toFixed(4)} total / $${step.avgCostUsd.toFixed(4)} avg` : '';
+    const iterRows = results.map((r, i) => {
+        const badge = r.passed ? 'iter-pass' : 'iter-fail';
+        const sym = r.passed ? '&#10003;' : '&#10005;';
+        const iterCost = r.costUsd != null && r.costUsd > 0 ? ` &middot; $${r.costUsd.toFixed(4)}` : '';
+        const assertions = r.assertionResults.map(a => {
+            const cls = a.passed ? 'assertion-pass' : 'assertion-fail';
+            const asym = a.passed ? '&#10003;' : '&#10005;';
+            return `<div class="assertion-line ${cls}">${asym} ${escHtml(a.type)}${a.message ? ': ' + escHtml(a.message) : ''}</div>`;
+        }).join('');
+        const outputId = `output-${index}-${i}`;
+        return `<div>
+      <div class="iter-row">
+        <span class="iter-badge ${badge}">${sym}</span>
+        <span>Iteration ${r.iteration}</span>
+        <span class="iter-duration">${formatDuration(r.durationMs)}${iterCost}</span>
+        <span class="iter-toggle" onclick="toggleOutput('${outputId}')">output</span>
+      </div>
+      ${assertions}
+      <div class="iter-output" id="${outputId}">${escHtml(r.output || '(no output)')}</div>
+    </div>`;
+    }).join('');
+    return `<div class="card">
+  <div class="step-header" onclick="toggleIterations(${index})">
+    <div>
+      <span class="step-icon" style="color:${iconColor}">${icon}</span>
+      <span class="step-name">${escHtml(step.stepId)}</span>
+    </div>
+    <div style="display:flex;align-items:center;gap:0.75rem;">
+      <span class="step-rate">${pct}% (threshold: ${threshPct}%)</span>
+      <span class="chevron" id="chev-${index}">&#9654;</span>
+    </div>
+  </div>
+  <div class="bar-container"><div class="bar-fill ${barColor}" style="width:${pct}%"></div></div>
+  <div class="step-stats">
+    <span>${step.passes}/${step.totalRuns} passed</span>
+    <span>${step.failures} failed</span>
+    ${avgDur ? `<span>avg ${avgDur}</span>` : ''}
+    ${costStr ? `<span>${costStr}</span>` : ''}
+  </div>
+  <div class="iterations" id="iters-${index}">
+    ${iterRows}
+  </div>
+</div>`;
+}
+function formatDuration(ms) {
+    if (ms < 1000)
+        return `${ms}ms`;
+    if (ms < 60_000)
+        return `${(ms / 1000).toFixed(1)}s`;
+    const mins = Math.floor(ms / 60_000);
+    const secs = ((ms % 60_000) / 1000).toFixed(0);
+    return `${mins}m ${secs}s`;
+}
+function escHtml(s) {
+    return s.replace(/&/g, '&amp;').replace(/</g, '&lt;').replace(/>/g, '&gt;').replace(/"/g, '&quot;');
+}
+//# sourceMappingURL=html-reporter.js.map

package/dist/reporters/html-reporter.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"html-reporter.js","sourceRoot":"","sources":["../../src/reporters/html-reporter.ts"],"names":[],"mappings":"AAEA,MAAM,UAAU,kBAAkB,CAAC,MAAsB;IACvD,MAAM,UAAU,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;IAC1C,MAAM,SAAS,GAAG,SAAS,CAAC;IAC5B,MAAM,SAAS,GAAG,SAAS,CAAC;IAC5B,MAAM,YAAY,GAAG,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC;IAC9D,MAAM,WAAW,GAAG,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC;IAE3D,OAAO;;;;;qBAKY,OAAO,CAAC,MAAM,CAAC,YAAY,CAAC;;;;;;;uIAOsF,YAAY,cAAc,YAAY,uBAAuB,YAAY;;;;;;;;;4BASpL,SAAS;4BACT,SAAS;;;;;;;6BAOR,SAAS;6BACT,SAAS;;;;;;;;6BAQT,SAAS;6BACT,SAAS;;;;;;;;;MAShC,OAAO,CAAC,MAAM,CAAC,YAAY,CAAC;;UAExB,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,cAAc,EAAE;UAC3C,cAAc,CAAC,MAAM,CAAC,UAAU,CAAC;UACjC,MAAM,CAAC,UAAU,aAAa,MAAM,CAAC,UAAU,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG;;uBAEnD,WAAW;;EAEhC,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC,EAAE,EAAE,CAAC,UAAU,CAAC,IAAI,EAAE,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC;;;UAG3G,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,MAAM,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM;iBAClE,cAAc,CAAC,MAAM,CAAC,UAAU,CAAC;;;;iBAIjC,UAAU;;;;;;;;;;;;;;;QAenB,CAAC;AACT,CAAC;AAED,SAAS,UAAU,CAAC,IAAiB,EAAE,OAAqB,EAAE,KAAa;IACzE,MAAM,GAAG,GAAG,CAAC,IAAI,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;IAC7C,MAAM,SAAS,GAAG,CAAC,IAAI,CAAC,WAAW,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;IACtD,MAAM,IAAI,GAAG,IAAI,CAAC,cAAc,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC;IAC3D,MAAM,SAAS,GAAG,IAAI,CAAC,cAAc,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC;IAC9D,MAAM,QAAQ,GAAG,IAAI,CAAC,cAAc,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC;IAC/D,MAAM,MAAM,GAAG,IAAI,CAAC,aAAa,CAAC,CAAC,CAAC,cAAc,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;IAC5E,MAAM,OAAO,GAAG,IAAI,CAAC,YAAY,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,aAAa,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC;IAE3H,MAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;QACpC,MAAM,KAAK,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,WAAW,CAAC;QACnD,MAAM,GAAG,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC;QAC/C,MAAM,QAAQ,GAAG,CAAC,CAAC,OAAO,IAAI,IAAI,IAAI,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAChG,MAAM,UAAU,GAAG,CAAC,CAAC,gBAAgB,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE;YAC5C,MAAM,GAAG,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,gBAAgB,CAAC,CAAC,CAAC,gBAAgB,CAAC;YAC3D,MAAM,IAAI,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC;YAChD,OAAO,8BAA8B,GAAG,KAAK,IAAI,IAAI,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,GAAG,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,EAAE,QAAQ,CAAC;QAC5H,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACZ,MAAM,QAAQ,GAAG,UAAU,KAAK,IAAI,CAAC,EAAE,CAAC;QACxC,OAAO;;kCAEuB,KAAK,KAAK,GAAG;0BACrB,CAAC,CAAC,SAAS;sCACC,cAAc,CAAC,CAAC,CAAC,UAAU,CAAC,GAAG,QAAQ;2DAClB,QAAQ;;QAE3D,UAAU;qCACmB,QAAQ,KAAK,OAAO,CAAC,CAAC,CAAC,MAAM,IAAI,aAAa,CAAC;WACzE,CAAC;IACV,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAEZ,OAAO;uDAC8C,KAAK;;6CAEf,SAAS,KAAK,IAAI;gCAC/B,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC;;;gCAGpB,GAAG,iBAAiB,SAAS;uCACtB,KAAK;;;oDAGQ,QAAQ,kBAAkB,GAAG;;YAErE,IAAI,CAAC,MAAM,IAAI,IAAI,CAAC,SAAS;YAC7B,IAAI,CAAC,QAAQ;MACnB,MAAM,CAAC,CAAC,CAAC,aAAa,MAAM,SAAS,CAAC,CAAC,CAAC,EAAE;MAC1C,OAAO,CAAC,CAAC,CAAC,SAAS,OAAO,SAAS,CAAC,CAAC,CAAC,EAAE;;sCAER,KAAK;MACrC,QAAQ;;OAEP,CAAC;AACR,CAAC;AAED,SAAS,cAAc,CAAC,EAAU;IAChC,IAAI,EAAE,GAAG,IAAI;QAAE,OAAO,GAAG,EAAE,IAAI,CAAC;IAChC,IAAI,EAAE,GAAG,MAAM;QAAE,OAAO,GAAG,CAAC,EAAE,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;IACrD,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,EAAE,GAAG,MAAM,CAAC,CAAC;IACrC,MAAM,IAAI,GAAG,CAAC,CAAC,EAAE,GAAG,MAAM,CAAC,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;IAC/C,OAAO,GAAG,IAAI,KAAK,IAAI,GAAG,CAAC;AAC7B,CAAC;AAED,SAAS,OAAO,CAAC,CAAS;IACxB,OAAO,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;AACtG,CAAC"}

package/dist/reporters/terminal-reporter.d.ts

@@ -1,4 +1,13 @@
 import type { ScenarioReport } from '../core/types.js';
-export declare function printReport(report: ScenarioReport, reportPath?: string): void;
+import type { BaselineComparison } from '../baseline.js';
+import type { CacheStats } from '../core/scenario-runner.js';
+import type { ComparisonReport } from '../commands/compare.js';
+export interface PrintReportOptions {
+    reportPath?: string;
+    baselineComparison?: BaselineComparison;
+    cacheStats?: CacheStats;
+}
+export declare function printReport(report: ScenarioReport, reportPathOrOpts?: string | PrintReportOptions): void;
 export declare function printProgress(stepId: string, iteration: number, total: number): void;
+export declare function formatComparison(report: ComparisonReport): string;
 //# sourceMappingURL=terminal-reporter.d.ts.map

package/dist/reporters/terminal-reporter.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"terminal-reporter.d.ts","sourceRoot":"","sources":["../../src/reporters/terminal-reporter.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,cAAc,EAAe,MAAM,kBAAkB,CAAC;AAKpE,wBAAgB,WAAW,CAAC,MAAM,EAAE,cAAc,EAAE,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI,CAwB7E;AAsDD,wBAAgB,aAAa,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,IAAI,CAKpF"}
+{"version":3,"file":"terminal-reporter.d.ts","sourceRoot":"","sources":["../../src/reporters/terminal-reporter.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,cAAc,EAAe,MAAM,kBAAkB,CAAC;AACpE,OAAO,KAAK,EAAE,kBAAkB,EAAkB,MAAM,gBAAgB,CAAC;AACzE,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,4BAA4B,CAAC;AAC7D,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAK/D,MAAM,WAAW,kBAAkB;IACjC,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,kBAAkB,CAAC,EAAE,kBAAkB,CAAC;IACxC,UAAU,CAAC,EAAE,UAAU,CAAC;CACzB;AAED,wBAAgB,WAAW,CAAC,MAAM,EAAE,cAAc,EAAE,gBAAgB,CAAC,EAAE,MAAM,GAAG,kBAAkB,GAAG,IAAI,CA+ExG;AAwFD,wBAAgB,aAAa,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,IAAI,CAKpF;AAED,wBAAgB,gBAAgB,CAAC,MAAM,EAAE,gBAAgB,GAAG,MAAM,CA6DjE"}