helixmind 0.5.4

package/dist/cli/bench/harness.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"harness.js","sourceRoot":"","sources":["../../../src/cli/bench/harness.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAC3C,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAC5C,OAAO,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AAC9C,OAAO,EAAE,gBAAgB,EAAE,gBAAgB,EAAE,YAAY,EAAE,mBAAmB,EAAE,MAAM,aAAa,CAAC;AACpG,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACpC,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AACtD,OAAO,EAAE,UAAU,IAAI,gBAAgB,EAAE,MAAM,uBAAuB,CAAC;AAEvE;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,YAAY,CAChC,MAAmB,EACnB,WAAmE,EACnE,SAAsE;IAEtE,MAAM,QAAQ,GAAG,IAAI,CAAC,OAAO,EAAE,EAAE,YAAY,EAAE,OAAO,CAAC,CAAC;IAExD,MAAM,KAAK,GAAG,MAAM,WAAW,CAAC,MAAM,CAAC,OAAO,EAAE,QAAQ,EAAE;QACxD,MAAM,EAAE,MAAM,CAAC,UAAU;QACzB,KAAK,EAAE,MAAM,CAAC,SAAS;QACvB,OAAO,EAAE,MAAM,CAAC,OAAO;KACxB,CAAC,CAAC;IAEH,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,OAAO,cAAc,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IACpC,CAAC;IAED,yDAAyD;IACzD,IAAI,YAAY,GAAG,IAAI,GAAG,EAAU,CAAC;IACrC,IAAI,MAAM,CAAC,WAAW,EAAE,CAAC;QACvB,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,EAAE,EAAE,YAAY,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,WAAW,CAAC,CAAC;QACrF,YAAY,GAAG,mBAAmB,CAAC,SAAS,CAAC,CAAC;IAChD,CAAC;IACD,kEAAkE;IAClE,MAAM,gBAAgB,GAAG,mBAAmB,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;IAC/D,KAAK,MAAM,EAAE,IAAI,gBAAgB;QAAE,YAAY,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;IAExD,MAAM,cAAc,GAAG,YAAY,CAAC,IAAI,GAAG,CAAC;QAC1C,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC;QACrD,CAAC,CAAC,KAAK,CAAC;IAEV,MAAM,YAAY,GAAG,KAAK,CAAC,MAAM,GAAG,cAAc,CAAC,MAAM,CAAC;IAC1D,IAAI,YAAY,GAAG,CAAC,IAAI,WAAW,EAAE,CAAC;QACpC,6BAA6B;QAC7B,OAAO,CAAC,MAAM,EAAE,KAAK,EAAE,CAAC,eAAe,YAAY,6BAA6B,cAAc,CAAC,MAAM,cAAc,CAAC,CAAC;IACvH,CAAC;IAED,+CAA+C;IAC/C,IAAI,YAAsC,CAAC;IAC3C,IAAI,MAAM,CAAC,UAAU,IAAI,MAAM,CAAC,UAAU,KAAK,UAAU,EAAE,CAAC;QAC1D,MAAM,aAAa,GAAG,IAAI,CAAC,MAAM,CAAC,SAAS,EAAE,iBAAiB,CAAC,CAAC;QAChE,SAAS,CAAC,aAAa,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAC9C,MAAM,YAAY,GAAG,gBAAgB,CAAC,aAAa,CAAC,CAAC;QACrD,YAAY,GAAG,IAAI,YAAY,CAAC,YAAY,CAAC,CAAC;QAC9C,MAAM,YAAY,CAAC,UAAU,EAAE,CAAC;IAClC,CAAC;IAED,MAAM,OAAO,GAAiB,EAAE,CAAC;IAEjC,IAAI,CAAC;QACH,MAAM,YAAY,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,+BAA+B;QAClE,MAAM,aAAa,GAAG,YAAY,CAAC,CAAC,gBAAgB;QAEpD,IAAI,MAAM,CAAC,WAAW,IAAI,CAAC,EAAE,CAAC;YAC5B,uBAAuB;YACvB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,cAAc,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBAC/C,MAAM,UAAU,GAAG,aAAa,GAAG,CAAC,CAAC;gBACrC,WAAW,EAAE,CAAC,cAAc,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,YAAY,CAAC,CAAC;gBAE3D,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,cAAc,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,SAAS,EAAE,YAAY,CAAC,CAAC;gBACvF,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBAErB,gDAAgD;gBAChD,IAAI,YAAY,EAAE,CAAC;oBACjB,IAAI,CAAC;wBAAC,YAAY,CAAC,MAAM,EAAE,CAAC;oBAAC,CAAC;oBAAC,MAAM,CAAC,CAAC,kBAAkB,CAAC,CAAC;gBAC7D,CAAC;gBAED,mCAAmC;gBACnC,gBAAgB,CAAC,MAAM,CAAC,SAAS,EAAE;oBACjC,WAAW,EAAE,MAAM,CAAC,WAAW;oBAC/B,kBAAkB,EAAE,GAAG,MAAM,CAAC,QAAQ,IAAI,MAAM,CAAC,KAAK,EAAE;oBACxD,WAAW,EAAE,MAAM,CAAC,WAAW;iBAChC,CAAC,CAAC;gBACH,gBAAgB,CAAC,MAAM,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;gBAE3C,SAAS,EAAE,CAAC,MAAM,EAAE,UAAU,EAAE,YAAY,CAAC,CAAC;YAChD,CAAC;QACH,CAAC;aAAM,CAAC;YACN,4CAA4C;YAC5C,MAAM,IAAI,CAAC,cAAc,EAAE,MAAM,CAAC,WAAW,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,EAAE,EAAE;gBAC/D,MAAM,UAAU,GAAG,aAAa,GAAG,CAAC,CAAC;gBACrC,WAAW,EAAE,CAAC,IAAI,EAAE,UAAU,EAAE,YAAY,CAAC,CAAC;gBAE9C,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,IAAI,EAAE,MAAM,EAAE,SAAS,EAAE,YAAY,CAAC,CAAC;gBAC1E,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBAErB,gBAAgB,CAAC,MAAM,CAAC,SAAS,EAAE;oBACjC,WAAW,EAAE,MAAM,CAAC,WAAW;oBAC/B,kBAAkB,EAAE,GAAG,MAAM,CAAC,QAAQ,IAAI,MAAM,CAAC,KAAK,EAAE;oBACxD,WAAW,EAAE,MAAM,CAAC,WAAW;iBAChC,CAAC,CAAC;gBACH,gBAAgB,CAAC,MAAM,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;gBAE3C,SAAS,EAAE,CAAC,MAAM,EAAE,UAAU,EAAE,YAAY,CAAC,CAAC;YAChD,CAAC,CAAC,CAAC;QACL,CAAC;QAED,oDAAoD;QACpD,IAAI,YAAY,EAAE,CAAC;YACjB,MAAM,QAAQ,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,UAAU,CAAC,CAAC,MAAM,CAAC;YACrE,MAAM,KAAK,GAAG,OAAO,CAAC,MAAM,CAAC;YAC7B,IAAI,CAAC;gBACH,MAAM,YAAY,CAAC,KAAK,CACtB,4BAA4B,QAAQ,IAAI,KAAK,cAAc,CAAC,CAAC,QAAQ,GAAG,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM;oBACnH,YAAY,MAAM,CAAC,OAAO,YAAY,MAAM,CAAC,KAAK,IAAI;oBACtD,oBAAoB,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,UAAU,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,EAClH,SAAS,EACT,EAAE,IAAI,EAAE,CAAC,WAAW,EAAE,SAAS,EAAE,MAAM,CAAC,OAAO,CAAC,EAAE,CACnD,CAAC;YACJ,CAAC;YAAC,MAAM,CAAC,CAAC,kBAAkB,CAAC,CAAC;QAChC,CAAC;IACH,CAAC;YAAS,CAAC;QACT,wBAAwB;QACxB,IAAI,YAAY,EAAE,CAAC;YACjB,IAAI,CAAC;gBAAC,YAAY,CAAC,KAAK,EAAE,CAAC;YAAC,CAAC;YAAC,MAAM,CAAC,CAAC,YAAY,CAAC,CAAC;QACtD,CAAC;IACH,CAAC;IAED,MAAM,OAAO,GAAG,cAAc,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAChD,YAAY,CAAC,MAAM,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC;IACxC,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,8CAA8C;AAC9C,KAAK,UAAU,IAAI,CACjB,KAAU,EACV,WAAmB,EACnB,EAA6C;IAE7C,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,MAAM,OAAO,GAAoB,EAAE,CAAC;IAEpC,KAAK,UAAU,MAAM;QACnB,OAAO,SAAS,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC;YAChC,MAAM,CAAC,GAAG,SAAS,EAAE,CAAC;YACtB,MAAM,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;QACxB,CAAC;IACH,CAAC;IAED,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,WAAW,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC7D,OAAO,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC;IACzB,CAAC;IAED,MAAM,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;AAC7B,CAAC"}

package/dist/cli/bench/metrics.d.ts

@@ -0,0 +1,15 @@
+import type { BenchConfig, TaskResult, BenchRunMetrics } from './types.js';
+/** Estimate cost for a given model and token counts */
+export declare function estimateCost(model: string, tokens: {
+    input: number;
+    output: number;
+}): number;
+/** Format USD with appropriate precision */
+export declare function formatCurrency(usd: number): string;
+/** Format token count for display */
+export declare function formatTokens(count: number): string;
+/** Format duration in human-readable form */
+export declare function formatDuration(ms: number): string;
+/** Compute aggregated metrics from a set of task results */
+export declare function computeMetrics(config: BenchConfig, results: TaskResult[]): BenchRunMetrics;
+//# sourceMappingURL=metrics.d.ts.map

package/dist/cli/bench/metrics.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"metrics.d.ts","sourceRoot":"","sources":["../../../src/cli/bench/metrics.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,WAAW,EAAE,UAAU,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AA4B3E,uDAAuD;AACvD,wBAAgB,YAAY,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,GAAG,MAAM,CAI7F;AAED,4CAA4C;AAC5C,wBAAgB,cAAc,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,CAIlD;AAED,qCAAqC;AACrC,wBAAgB,YAAY,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAIlD;AAED,6CAA6C;AAC7C,wBAAgB,cAAc,CAAC,EAAE,EAAE,MAAM,GAAG,MAAM,CASjD;AAED,4DAA4D;AAC5D,wBAAgB,cAAc,CAAC,MAAM,EAAE,WAAW,EAAE,OAAO,EAAE,UAAU,EAAE,GAAG,eAAe,CAoC1F"}

package/dist/cli/bench/metrics.js

@@ -0,0 +1,98 @@
+/** Model pricing per 1M tokens in USD */
+const MODEL_PRICING = {
+    // Anthropic
+    'claude-sonnet-4-6': { input: 3.0, output: 15.0 },
+    'claude-opus-4-6': { input: 15.0, output: 75.0 },
+    'claude-haiku-4-5-20251001': { input: 1.0, output: 5.0 },
+    // OpenAI
+    'gpt-4o': { input: 2.5, output: 10.0 },
+    'gpt-4o-mini': { input: 0.15, output: 0.6 },
+    'gpt-4-turbo': { input: 10.0, output: 30.0 },
+    'o1': { input: 15.0, output: 60.0 },
+    'o1-mini': { input: 3.0, output: 12.0 },
+    'o3-mini': { input: 1.1, output: 4.4 },
+    // DeepSeek
+    'deepseek-chat': { input: 0.14, output: 0.28 },
+    'deepseek-reasoner': { input: 0.55, output: 2.19 },
+    // Z.AI
+    'glm-5': { input: 0.5, output: 1.0 },
+    'glm-4.7': { input: 0.3, output: 0.6 },
+    'glm-4.6': { input: 0.2, output: 0.4 },
+    'glm-4.5': { input: 0.15, output: 0.3 },
+    // OpenRouter (pass-through pricing varies)
+    'anthropic/claude-sonnet-4': { input: 3.0, output: 15.0 },
+    'openai/gpt-4o': { input: 2.5, output: 10.0 },
+};
+/** Estimate cost for a given model and token counts */
+export function estimateCost(model, tokens) {
+    const pricing = MODEL_PRICING[model];
+    if (!pricing)
+        return 0;
+    return (tokens.input / 1_000_000) * pricing.input + (tokens.output / 1_000_000) * pricing.output;
+}
+/** Format USD with appropriate precision */
+export function formatCurrency(usd) {
+    if (usd < 0.01)
+        return `$${usd.toFixed(4)}`;
+    if (usd < 1)
+        return `$${usd.toFixed(3)}`;
+    return `$${usd.toFixed(2)}`;
+}
+/** Format token count for display */
+export function formatTokens(count) {
+    if (count >= 1_000_000)
+        return `${(count / 1_000_000).toFixed(1)}M`;
+    if (count >= 1_000)
+        return `${(count / 1_000).toFixed(1)}k`;
+    return String(count);
+}
+/** Format duration in human-readable form */
+export function formatDuration(ms) {
+    if (ms < 1000)
+        return `${ms}ms`;
+    const s = Math.round(ms / 1000);
+    if (s < 60)
+        return `${s}s`;
+    const m = Math.floor(s / 60);
+    const rem = s % 60;
+    if (m < 60)
+        return `${m}m ${rem}s`;
+    const h = Math.floor(m / 60);
+    return `${h}h ${m % 60}m`;
+}
+/** Compute aggregated metrics from a set of task results */
+export function computeMetrics(config, results) {
+    const resolved = results.filter(r => r.status === 'resolved').length;
+    const failed = results.filter(r => r.status === 'failed').length;
+    const errors = results.filter(r => r.status === 'error').length;
+    const timeouts = results.filter(r => r.status === 'timeout').length;
+    const totalTokensIn = results.reduce((sum, r) => sum + r.tokens.input, 0);
+    const totalTokensOut = results.reduce((sum, r) => sum + r.tokens.output, 0);
+    const totalToolCalls = results.reduce((sum, r) => sum + r.toolCalls, 0);
+    const totalDuration = results.reduce((sum, r) => sum + r.durationMs, 0);
+    const n = results.length || 1;
+    return {
+        runId: config.runId,
+        timestamp: new Date().toISOString(),
+        dataset: config.dataset,
+        provider: config.provider,
+        model: config.model,
+        totalTasks: results.length,
+        resolved,
+        failed,
+        errors,
+        timeouts,
+        resolutionRate: results.length > 0 ? (resolved / results.length) * 100 : 0,
+        avgTokensPerTask: {
+            input: Math.round(totalTokensIn / n),
+            output: Math.round(totalTokensOut / n),
+        },
+        avgToolCallsPerTask: Math.round((totalToolCalls / n) * 10) / 10,
+        avgDurationMs: Math.round(totalDuration / n),
+        totalCostEstimate: estimateCost(config.model, { input: totalTokensIn, output: totalTokensOut }),
+        totalDurationMs: totalDuration,
+        taskResults: results,
+        spiralMode: config.withSpiral ? (config.spiralMode ?? 'fresh') : undefined,
+    };
+}
+//# sourceMappingURL=metrics.js.map

package/dist/cli/bench/metrics.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"metrics.js","sourceRoot":"","sources":["../../../src/cli/bench/metrics.ts"],"names":[],"mappings":"AAEA,yCAAyC;AACzC,MAAM,aAAa,GAAsD;IACvE,YAAY;IACZ,mBAAmB,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE;IACjD,iBAAiB,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE;IAChD,2BAA2B,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,GAAG,EAAE;IACxD,SAAS;IACT,QAAQ,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE;IACtC,aAAa,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,GAAG,EAAE;IAC3C,aAAa,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE;IAC5C,IAAI,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE;IACnC,SAAS,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE;IACvC,SAAS,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,GAAG,EAAE;IACtC,WAAW;IACX,eAAe,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE;IAC9C,mBAAmB,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE;IAClD,OAAO;IACP,OAAO,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,GAAG,EAAE;IACpC,SAAS,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,GAAG,EAAE;IACtC,SAAS,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,GAAG,EAAE;IACtC,SAAS,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,GAAG,EAAE;IACvC,2CAA2C;IAC3C,2BAA2B,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE;IACzD,eAAe,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE;CAC9C,CAAC;AAEF,uDAAuD;AACvD,MAAM,UAAU,YAAY,CAAC,KAAa,EAAE,MAAyC;IACnF,MAAM,OAAO,GAAG,aAAa,CAAC,KAAK,CAAC,CAAC;IACrC,IAAI,CAAC,OAAO;QAAE,OAAO,CAAC,CAAC;IACvB,OAAO,CAAC,MAAM,CAAC,KAAK,GAAG,SAAS,CAAC,GAAG,OAAO,CAAC,KAAK,GAAG,CAAC,MAAM,CAAC,MAAM,GAAG,SAAS,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC;AACnG,CAAC;AAED,4CAA4C;AAC5C,MAAM,UAAU,cAAc,CAAC,GAAW;IACxC,IAAI,GAAG,GAAG,IAAI;QAAE,OAAO,IAAI,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC;IAC5C,IAAI,GAAG,GAAG,CAAC;QAAE,OAAO,IAAI,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC;IACzC,OAAO,IAAI,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC;AAC9B,CAAC;AAED,qCAAqC;AACrC,MAAM,UAAU,YAAY,CAAC,KAAa;IACxC,IAAI,KAAK,IAAI,SAAS;QAAE,OAAO,GAAG,CAAC,KAAK,GAAG,SAAS,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;IACpE,IAAI,KAAK,IAAI,KAAK;QAAE,OAAO,GAAG,CAAC,KAAK,GAAG,KAAK,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;IAC5D,OAAO,MAAM,CAAC,KAAK,CAAC,CAAC;AACvB,CAAC;AAED,6CAA6C;AAC7C,MAAM,UAAU,cAAc,CAAC,EAAU;IACvC,IAAI,EAAE,GAAG,IAAI;QAAE,OAAO,GAAG,EAAE,IAAI,CAAC;IAChC,MAAM,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,EAAE,GAAG,IAAI,CAAC,CAAC;IAChC,IAAI,CAAC,GAAG,EAAE;QAAE,OAAO,GAAG,CAAC,GAAG,CAAC;IAC3B,MAAM,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC;IAC7B,MAAM,GAAG,GAAG,CAAC,GAAG,EAAE,CAAC;IACnB,IAAI,CAAC,GAAG,EAAE;QAAE,OAAO,GAAG,CAAC,KAAK,GAAG,GAAG,CAAC;IACnC,MAAM,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC;IAC7B,OAAO,GAAG,CAAC,KAAK,CAAC,GAAG,EAAE,GAAG,CAAC;AAC5B,CAAC;AAED,4DAA4D;AAC5D,MAAM,UAAU,cAAc,CAAC,MAAmB,EAAE,OAAqB;IACvE,MAAM,QAAQ,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,UAAU,CAAC,CAAC,MAAM,CAAC;IACrE,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,MAAM,CAAC;IACjE,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,OAAO,CAAC,CAAC,MAAM,CAAC;IAChE,MAAM,QAAQ,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,MAAM,CAAC;IAEpE,MAAM,aAAa,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;IAC1E,MAAM,cAAc,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IAC5E,MAAM,cAAc,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC;IACxE,MAAM,aAAa,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC;IAExE,MAAM,CAAC,GAAG,OAAO,CAAC,MAAM,IAAI,CAAC,CAAC;IAE9B,OAAO;QACL,KAAK,EAAE,MAAM,CAAC,KAAK;QACnB,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QACnC,OAAO,EAAE,MAAM,CAAC,OAAO;QACvB,QAAQ,EAAE,MAAM,CAAC,QAAQ;QACzB,KAAK,EAAE,MAAM,CAAC,KAAK;QACnB,UAAU,EAAE,OAAO,CAAC,MAAM;QAC1B,QAAQ;QACR,MAAM;QACN,MAAM;QACN,QAAQ;QACR,cAAc,EAAE,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,GAAG,OAAO,CAAC,MAAM,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;QAC1E,gBAAgB,EAAE;YAChB,KAAK,EAAE,IAAI,CAAC,KAAK,CAAC,aAAa,GAAG,CAAC,CAAC;YACpC,MAAM,EAAE,IAAI,CAAC,KAAK,CAAC,cAAc,GAAG,CAAC,CAAC;SACvC;QACD,mBAAmB,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,cAAc,GAAG,CAAC,CAAC,GAAG,EAAE,CAAC,GAAG,EAAE;QAC/D,aAAa,EAAE,IAAI,CAAC,KAAK,CAAC,aAAa,GAAG,CAAC,CAAC;QAC5C,iBAAiB,EAAE,YAAY,CAAC,MAAM,CAAC,KAAK,EAAE,EAAE,KAAK,EAAE,aAAa,EAAE,MAAM,EAAE,cAAc,EAAE,CAAC;QAC/F,eAAe,EAAE,aAAa;QAC9B,WAAW,EAAE,OAAO;QACpB,UAAU,EAAE,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,UAAU,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS;KAC3E,CAAC;AACJ,CAAC"}

package/dist/cli/bench/output.d.ts

@@ -0,0 +1,42 @@
+import type { SWEPrediction, TaskResult, BenchRunMetrics, RunSummary } from './types.js';
+/**
+ * Append a single prediction to the JSONL file (SWE-bench compatible).
+ */
+export declare function appendPrediction(outputDir: string, prediction: SWEPrediction): void;
+/**
+ * Append a task result to the detailed results JSONL.
+ */
+export declare function appendTaskResult(outputDir: string, result: TaskResult): void;
+/**
+ * Write the final run summary.
+ */
+export declare function writeSummary(outputDir: string, metrics: BenchRunMetrics): void;
+/**
+ * List all past benchmark runs from the bench/runs/ directory.
+ */
+export declare function listRuns(benchDir: string): RunSummary[];
+/**
+ * Get IDs of tasks that were already completed (resolved/failed) in a previous run.
+ * Reads results.jsonl line by line and returns Set of instance_ids.
+ */
+export declare function getCompletedTaskIds(outputDir: string): Set<string>;
+/**
+ * Load previous run's task results for merging into resumed run.
+ */
+export declare function loadPreviousResults(outputDir: string): Array<{
+    instance_id: string;
+    status: string;
+    tokens: {
+        input: number;
+        output: number;
+    };
+    toolCalls: number;
+    durationMs: number;
+    errors: string[];
+    stepCount: number;
+}>;
+/**
+ * Load a specific run's summary metrics.
+ */
+export declare function loadRunSummary(benchDir: string, runId: string): BenchRunMetrics | null;
+//# sourceMappingURL=output.d.ts.map

package/dist/cli/bench/output.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"output.d.ts","sourceRoot":"","sources":["../../../src/cli/bench/output.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,aAAa,EAAE,UAAU,EAAE,eAAe,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AAEzF;;GAEG;AACH,wBAAgB,gBAAgB,CAAC,SAAS,EAAE,MAAM,EAAE,UAAU,EAAE,aAAa,GAAG,IAAI,CAOnF;AAED;;GAEG;AACH,wBAAgB,gBAAgB,CAAC,SAAS,EAAE,MAAM,EAAE,MAAM,EAAE,UAAU,GAAG,IAAI,CAiB5E;AAED;;GAEG;AACH,wBAAgB,YAAY,CAAC,SAAS,EAAE,MAAM,EAAE,OAAO,EAAE,eAAe,GAAG,IAAI,CAS9E;AAED;;GAEG;AACH,wBAAgB,QAAQ,CAAC,QAAQ,EAAE,MAAM,GAAG,UAAU,EAAE,CAwCvD;AAED;;;GAGG;AACH,wBAAgB,mBAAmB,CAAC,SAAS,EAAE,MAAM,GAAG,GAAG,CAAC,MAAM,CAAC,CAoBlE;AAED;;GAEG;AACH,wBAAgB,mBAAmB,CAAC,SAAS,EAAE,MAAM,GAAG,KAAK,CAAC;IAAE,WAAW,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC;IAAC,SAAS,EAAE,MAAM,CAAC;IAAC,UAAU,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,EAAE,CAAC;IAAC,SAAS,EAAE,MAAM,CAAA;CAAE,CAAC,CAY5N;AAED;;GAEG;AACH,wBAAgB,cAAc,CAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,eAAe,GAAG,IAAI,CAUtF"}

package/dist/cli/bench/output.js

@@ -0,0 +1,140 @@
+import { appendFileSync, writeFileSync, readFileSync, readdirSync, existsSync, mkdirSync } from 'node:fs';
+import { join } from 'node:path';
+/**
+ * Append a single prediction to the JSONL file (SWE-bench compatible).
+ */
+export function appendPrediction(outputDir, prediction) {
+    mkdirSync(outputDir, { recursive: true });
+    appendFileSync(join(outputDir, 'predictions.jsonl'), JSON.stringify(prediction) + '\n', 'utf-8');
+}
+/**
+ * Append a task result to the detailed results JSONL.
+ */
+export function appendTaskResult(outputDir, result) {
+    mkdirSync(outputDir, { recursive: true });
+    // Write a compact version without the full patch/agentText
+    const compact = {
+        instance_id: result.instance_id,
+        status: result.status,
+        tokens: result.tokens,
+        toolCalls: result.toolCalls,
+        durationMs: result.durationMs,
+        errors: result.errors,
+        stepCount: result.steps.length,
+    };
+    appendFileSync(join(outputDir, 'results.jsonl'), JSON.stringify(compact) + '\n', 'utf-8');
+}
+/**
+ * Write the final run summary.
+ */
+export function writeSummary(outputDir, metrics) {
+    mkdirSync(outputDir, { recursive: true });
+    // Don't include taskResults in summary (too large)
+    const { taskResults: _, ...summary } = metrics;
+    writeFileSync(join(outputDir, 'summary.json'), JSON.stringify(summary, null, 2), 'utf-8');
+}
+/**
+ * List all past benchmark runs from the bench/runs/ directory.
+ */
+export function listRuns(benchDir) {
+    const runsDir = join(benchDir, 'runs');
+    if (!existsSync(runsDir))
+        return [];
+    const runs = [];
+    try {
+        const dirs = readdirSync(runsDir, { withFileTypes: true })
+            .filter(d => d.isDirectory())
+            .map(d => d.name)
+            .sort()
+            .reverse(); // Newest first
+        for (const dir of dirs) {
+            const summaryPath = join(runsDir, dir, 'summary.json');
+            if (!existsSync(summaryPath))
+                continue;
+            try {
+                const data = JSON.parse(readFileSync(summaryPath, 'utf-8'));
+                runs.push({
+                    runId: data.runId ?? dir,
+                    timestamp: data.timestamp ?? '',
+                    dataset: data.dataset ?? '',
+                    provider: data.provider ?? '',
+                    model: data.model ?? '',
+                    resolved: data.resolved ?? 0,
+                    totalTasks: data.totalTasks ?? 0,
+                    resolutionRate: data.resolutionRate ?? 0,
+                    totalCostEstimate: data.totalCostEstimate ?? 0,
+                    spiralMode: data.spiralMode,
+                });
+            }
+            catch {
+                // Corrupted summary, skip
+            }
+        }
+    }
+    catch {
+        // Runs directory doesn't exist or can't be read
+    }
+    return runs;
+}
+/**
+ * Get IDs of tasks that were already completed (resolved/failed) in a previous run.
+ * Reads results.jsonl line by line and returns Set of instance_ids.
+ */
+export function getCompletedTaskIds(outputDir) {
+    const resultsPath = join(outputDir, 'results.jsonl');
+    const completed = new Set();
+    if (!existsSync(resultsPath))
+        return completed;
+    try {
+        const lines = readFileSync(resultsPath, 'utf-8').split('\n').filter(l => l.trim());
+        for (const line of lines) {
+            try {
+                const data = JSON.parse(line);
+                // Only skip tasks that actually completed (resolved or failed)
+                // Re-run errored/timed-out tasks (they might succeed with budget)
+                if (data.instance_id && (data.status === 'resolved' || data.status === 'failed')) {
+                    completed.add(data.instance_id);
+                }
+            }
+            catch { /* skip malformed lines */ }
+        }
+    }
+    catch { /* file read error */ }
+    return completed;
+}
+/**
+ * Load previous run's task results for merging into resumed run.
+ */
+export function loadPreviousResults(outputDir) {
+    const resultsPath = join(outputDir, 'results.jsonl');
+    if (!existsSync(resultsPath))
+        return [];
+    const results = [];
+    try {
+        const lines = readFileSync(resultsPath, 'utf-8').split('\n').filter(l => l.trim());
+        for (const line of lines) {
+            try {
+                results.push(JSON.parse(line));
+            }
+            catch { /* skip */ }
+        }
+    }
+    catch { /* ignore */ }
+    return results;
+}
+/**
+ * Load a specific run's summary metrics.
+ */
+export function loadRunSummary(benchDir, runId) {
+    const summaryPath = join(benchDir, 'runs', runId, 'summary.json');
+    if (!existsSync(summaryPath))
+        return null;
+    try {
+        const data = JSON.parse(readFileSync(summaryPath, 'utf-8'));
+        return { ...data, taskResults: [] };
+    }
+    catch {
+        return null;
+    }
+}
+//# sourceMappingURL=output.js.map

package/dist/cli/bench/output.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"output.js","sourceRoot":"","sources":["../../../src/cli/bench/output.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,aAAa,EAAE,YAAY,EAAE,WAAW,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AAC1G,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAGjC;;GAEG;AACH,MAAM,UAAU,gBAAgB,CAAC,SAAiB,EAAE,UAAyB;IAC3E,SAAS,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAC1C,cAAc,CACZ,IAAI,CAAC,SAAS,EAAE,mBAAmB,CAAC,EACpC,IAAI,CAAC,SAAS,CAAC,UAAU,CAAC,GAAG,IAAI,EACjC,OAAO,CACR,CAAC;AACJ,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,gBAAgB,CAAC,SAAiB,EAAE,MAAkB;IACpE,SAAS,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAC1C,2DAA2D;IAC3D,MAAM,OAAO,GAAG;QACd,WAAW,EAAE,MAAM,CAAC,WAAW;QAC/B,MAAM,EAAE,MAAM,CAAC,MAAM;QACrB,MAAM,EAAE,MAAM,CAAC,MAAM;QACrB,SAAS,EAAE,MAAM,CAAC,SAAS;QAC3B,UAAU,EAAE,MAAM,CAAC,UAAU;QAC7B,MAAM,EAAE,MAAM,CAAC,MAAM;QACrB,SAAS,EAAE,MAAM,CAAC,KAAK,CAAC,MAAM;KAC/B,CAAC;IACF,cAAc,CACZ,IAAI,CAAC,SAAS,EAAE,eAAe,CAAC,EAChC,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,GAAG,IAAI,EAC9B,OAAO,CACR,CAAC;AACJ,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,YAAY,CAAC,SAAiB,EAAE,OAAwB;IACtE,SAAS,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAC1C,mDAAmD;IACnD,MAAM,EAAE,WAAW,EAAE,CAAC,EAAE,GAAG,OAAO,EAAE,GAAG,OAAO,CAAC;IAC/C,aAAa,CACX,IAAI,CAAC,SAAS,EAAE,cAAc,CAAC,EAC/B,IAAI,CAAC,SAAS,CAAC,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,EAChC,OAAO,CACR,CAAC;AACJ,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,QAAQ,CAAC,QAAgB;IACvC,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;IACvC,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC;QAAE,OAAO,EAAE,CAAC;IAEpC,MAAM,IAAI,GAAiB,EAAE,CAAC;IAE9B,IAAI,CAAC;QACH,MAAM,IAAI,GAAG,WAAW,CAAC,OAAO,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC;aACvD,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC;aAC5B,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;aAChB,IAAI,EAAE;aACN,OAAO,EAAE,CAAC,CAAC,eAAe;QAE7B,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;YACvB,MAAM,WAAW,GAAG,IAAI,CAAC,OAAO,EAAE,GAAG,EAAE,cAAc,CAAC,CAAC;YACvD,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC;gBAAE,SAAS;YAEvC,IAAI,CAAC;gBACH,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC,CAAC;gBAC5D,IAAI,CAAC,IAAI,CAAC;oBACR,KAAK,EAAE,IAAI,CAAC,KAAK,IAAI,GAAG;oBACxB,SAAS,EAAE,IAAI,CAAC,SAAS,IAAI,EAAE;oBAC/B,OAAO,EAAE,IAAI,CAAC,OAAO,IAAI,EAAE;oBAC3B,QAAQ,EAAE,IAAI,CAAC,QAAQ,IAAI,EAAE;oBAC7B,KAAK,EAAE,IAAI,CAAC,KAAK,IAAI,EAAE;oBACvB,QAAQ,EAAE,IAAI,CAAC,QAAQ,IAAI,CAAC;oBAC5B,UAAU,EAAE,IAAI,CAAC,UAAU,IAAI,CAAC;oBAChC,cAAc,EAAE,IAAI,CAAC,cAAc,IAAI,CAAC;oBACxC,iBAAiB,EAAE,IAAI,CAAC,iBAAiB,IAAI,CAAC;oBAC9C,UAAU,EAAE,IAAI,CAAC,UAAU;iBAC5B,CAAC,CAAC;YACL,CAAC;YAAC,MAAM,CAAC;gBACP,0BAA0B;YAC5B,CAAC;QACH,CAAC;IACH,CAAC;IAAC,MAAM,CAAC;QACP,gDAAgD;IAClD,CAAC;IAED,OAAO,IAAI,CAAC;AACd,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,mBAAmB,CAAC,SAAiB;IACnD,MAAM,WAAW,GAAG,IAAI,CAAC,SAAS,EAAE,eAAe,CAAC,CAAC;IACrD,MAAM,SAAS,GAAG,IAAI,GAAG,EAAU,CAAC;IACpC,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC;QAAE,OAAO,SAAS,CAAC;IAE/C,IAAI,CAAC;QACH,MAAM,KAAK,GAAG,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;QACnF,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACzB,IAAI,CAAC;gBACH,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;gBAC9B,+DAA+D;gBAC/D,kEAAkE;gBAClE,IAAI,IAAI,CAAC,WAAW,IAAI,CAAC,IAAI,CAAC,MAAM,KAAK,UAAU,IAAI,IAAI,CAAC,MAAM,KAAK,QAAQ,CAAC,EAAE,CAAC;oBACjF,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;gBAClC,CAAC;YACH,CAAC;YAAC,MAAM,CAAC,CAAC,0BAA0B,CAAC,CAAC;QACxC,CAAC;IACH,CAAC;IAAC,MAAM,CAAC,CAAC,qBAAqB,CAAC,CAAC;IAEjC,OAAO,SAAS,CAAC;AACnB,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB,CAAC,SAAiB;IACnD,MAAM,WAAW,GAAG,IAAI,CAAC,SAAS,EAAE,eAAe,CAAC,CAAC;IACrD,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC;QAAE,OAAO,EAAE,CAAC;IAExC,MAAM,OAAO,GAAe,EAAE,CAAC;IAC/B,IAAI,CAAC;QACH,MAAM,KAAK,GAAG,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;QACnF,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACzB,IAAI,CAAC;gBAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;YAAC,CAAC;YAAC,MAAM,CAAC,CAAC,UAAU,CAAC,CAAC;QAC9D,CAAC;IACH,CAAC;IAAC,MAAM,CAAC,CAAC,YAAY,CAAC,CAAC;IACxB,OAAO,OAAO,CAAC;AACjB,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,cAAc,CAAC,QAAgB,EAAE,KAAa;IAC5D,MAAM,WAAW,GAAG,IAAI,CAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,cAAc,CAAC,CAAC;IAClE,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC;QAAE,OAAO,IAAI,CAAC;IAE1C,IAAI,CAAC;QACH,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC,CAAC;QAC5D,OAAO,EAAE,GAAG,IAAI,EAAE,WAAW,EAAE,EAAE,EAAE,CAAC;IACtC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC"}

package/dist/cli/bench/prompt.d.ts

@@ -0,0 +1,13 @@
+import type { SWETask } from './types.js';
+import type { SpiralQueryResult } from '../../types.js';
+/**
+ * Build a system prompt optimized for SWE-bench task resolution.
+ * Focused on minimal, targeted bug fixes.
+ */
+export declare function buildBenchSystemPrompt(task: SWETask, projectRoot: string): string;
+/**
+ * Build a system prompt with Spiral Memory context for enhanced SWE-bench solving.
+ * Combines standard bench instructions with spiral context sections.
+ */
+export declare function buildBenchSystemPromptWithSpiral(task: SWETask, projectRoot: string, spiralContext: SpiralQueryResult): string;
+//# sourceMappingURL=prompt.d.ts.map

package/dist/cli/bench/prompt.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"prompt.d.ts","sourceRoot":"","sources":["../../../src/cli/bench/prompt.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,YAAY,CAAC;AAC1C,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,gBAAgB,CAAC;AAExD;;;GAGG;AACH,wBAAgB,sBAAsB,CAAC,IAAI,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,GAAG,MAAM,CA0CjF;AAED;;;GAGG;AACH,wBAAgB,gCAAgC,CAC9C,IAAI,EAAE,OAAO,EACb,WAAW,EAAE,MAAM,EACnB,aAAa,EAAE,iBAAiB,GAC/B,MAAM,CAmER"}

package/dist/cli/bench/prompt.js

@@ -0,0 +1,106 @@
+/**
+ * Build a system prompt optimized for SWE-bench task resolution.
+ * Focused on minimal, targeted bug fixes.
+ */
+export function buildBenchSystemPrompt(task, projectRoot) {
+    const sections = [];
+    sections.push(`You are an expert software engineer. Your task is to fix a GitHub issue in a real codebase.
+
+## Repository
+- Repo: ${task.repo}
+- Working directory: ${projectRoot}
+- The repository has been checked out at the exact commit where the issue exists.
+
+## Issue to Fix
+${task.problem_statement}`);
+    if (task.hints_text && task.hints_text.trim()) {
+        sections.push(`## Hints
+${task.hints_text}`);
+    }
+    sections.push(`## Instructions
+1. Read the issue carefully and understand what is expected.
+2. Explore the codebase to find the relevant files using search_files and find_files.
+3. Read and understand the existing code before making changes.
+4. Make the MINIMAL changes needed to fix the issue. Do not refactor, do not add features.
+5. Do NOT modify any test files. Only fix the source code.
+6. After making changes, verify your fix by reading the modified files.
+7. Do NOT run the test suite — that happens automatically after your fix.
+
+## Rules
+- Change as few files as possible.
+- Change as few lines as possible.
+- Do not add type annotations, docstrings, or comments unless they are part of the fix.
+- Do not rename variables or move code unless required by the fix.
+- Do not use git commit — just edit the files directly.
+- If the issue mentions a specific file, start there.
+- If unsure, search for error messages or class/function names from the issue.`);
+    sections.push(`## Available Tools
+You have: read_file, write_file, edit_file, list_directory, search_files, find_files, run_command, git_status, git_diff, git_log.
+
+Use edit_file for targeted changes. Use search_files to find relevant code. Use run_command for exploring (e.g., grep, python -c).`);
+    return sections.join('\n\n');
+}
+/**
+ * Build a system prompt with Spiral Memory context for enhanced SWE-bench solving.
+ * Combines standard bench instructions with spiral context sections.
+ */
+export function buildBenchSystemPromptWithSpiral(task, projectRoot, spiralContext) {
+    const sections = [];
+    sections.push(`You are an expert software engineer with access to a spiral memory system that provides contextual knowledge from similar codebases and past bug fixes.
+
+## Repository
+- Repo: ${task.repo}
+- Working directory: ${projectRoot}
+- The repository has been checked out at the exact commit where the issue exists.
+
+## Issue to Fix
+${task.problem_statement}`);
+    if (task.hints_text && task.hints_text.trim()) {
+        sections.push(`## Hints
+${task.hints_text}`);
+    }
+    // Inject spiral context if available
+    if (spiralContext.node_count > 0) {
+        const spiralSections = ['## Spiral Memory Context\nRelevant knowledge from the spiral memory system:'];
+        const levelMap = [
+            { label: 'Focus (highly relevant)', nodes: spiralContext.level_1 },
+            { label: 'Active (recent patterns)', nodes: spiralContext.level_2 },
+            { label: 'Reference (background)', nodes: spiralContext.level_3 },
+        ];
+        for (const { label, nodes } of levelMap) {
+            if (nodes.length > 0) {
+                spiralSections.push(`### ${label}:`);
+                for (const node of nodes) {
+                    spiralSections.push(`- [${node.type}] ${node.content}`);
+                }
+            }
+        }
+        sections.push(spiralSections.join('\n'));
+    }
+    sections.push(`## Instructions
+1. Read the issue carefully and understand what is expected.
+2. Review the Spiral Memory Context above for relevant patterns and prior fixes.
+3. Explore the codebase to find the relevant files using search_files and find_files.
+4. Read and understand the existing code before making changes.
+5. Make the MINIMAL changes needed to fix the issue. Do not refactor, do not add features.
+6. Do NOT modify any test files. Only fix the source code.
+7. After making changes, verify your fix by reading the modified files.
+8. Do NOT run the test suite — that happens automatically after your fix.
+9. Use spiral_query to search for similar patterns or fixes if stuck.
+10. Use spiral_store to save important patterns you discover for future tasks.
+
+## Rules
+- Change as few files as possible.
+- Change as few lines as possible.
+- Do not add type annotations, docstrings, or comments unless they are part of the fix.
+- Do not rename variables or move code unless required by the fix.
+- Do not use git commit — just edit the files directly.
+- If the issue mentions a specific file, start there.
+- If unsure, search for error messages or class/function names from the issue.`);
+    sections.push(`## Available Tools
+You have: read_file, write_file, edit_file, list_directory, search_files, find_files, run_command, git_status, git_diff, git_log, spiral_query, spiral_store.
+
+Use edit_file for targeted changes. Use search_files to find relevant code. Use spiral_query to leverage memory from similar fixes. Use spiral_store to save fix patterns for future reference.`);
+    return sections.join('\n\n');
+}
+//# sourceMappingURL=prompt.js.map

package/dist/cli/bench/prompt.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"prompt.js","sourceRoot":"","sources":["../../../src/cli/bench/prompt.ts"],"names":[],"mappings":"AAGA;;;GAGG;AACH,MAAM,UAAU,sBAAsB,CAAC,IAAa,EAAE,WAAmB;IACvE,MAAM,QAAQ,GAAa,EAAE,CAAC;IAE9B,QAAQ,CAAC,IAAI,CAAC;;;UAGN,IAAI,CAAC,IAAI;uBACI,WAAW;;;;EAIhC,IAAI,CAAC,iBAAiB,EAAE,CAAC,CAAC;IAE1B,IAAI,IAAI,CAAC,UAAU,IAAI,IAAI,CAAC,UAAU,CAAC,IAAI,EAAE,EAAE,CAAC;QAC9C,QAAQ,CAAC,IAAI,CAAC;EAChB,IAAI,CAAC,UAAU,EAAE,CAAC,CAAC;IACnB,CAAC;IAED,QAAQ,CAAC,IAAI,CAAC;;;;;;;;;;;;;;;;+EAgB+D,CAAC,CAAC;IAE/E,QAAQ,CAAC,IAAI,CAAC;;;mIAGmH,CAAC,CAAC;IAEnI,OAAO,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;AAC/B,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,gCAAgC,CAC9C,IAAa,EACb,WAAmB,EACnB,aAAgC;IAEhC,MAAM,QAAQ,GAAa,EAAE,CAAC;IAE9B,QAAQ,CAAC,IAAI,CAAC;;;UAGN,IAAI,CAAC,IAAI;uBACI,WAAW;;;;EAIhC,IAAI,CAAC,iBAAiB,EAAE,CAAC,CAAC;IAE1B,IAAI,IAAI,CAAC,UAAU,IAAI,IAAI,CAAC,UAAU,CAAC,IAAI,EAAE,EAAE,CAAC;QAC9C,QAAQ,CAAC,IAAI,CAAC;EAChB,IAAI,CAAC,UAAU,EAAE,CAAC,CAAC;IACnB,CAAC;IAED,qCAAqC;IACrC,IAAI,aAAa,CAAC,UAAU,GAAG,CAAC,EAAE,CAAC;QACjC,MAAM,cAAc,GAAa,CAAC,6EAA6E,CAAC,CAAC;QAEjH,MAAM,QAAQ,GAAkE;YAC9E,EAAE,KAAK,EAAE,yBAAyB,EAAE,KAAK,EAAE,aAAa,CAAC,OAAO,EAAE;YAClE,EAAE,KAAK,EAAE,0BAA0B,EAAE,KAAK,EAAE,aAAa,CAAC,OAAO,EAAE;YACnE,EAAE,KAAK,EAAE,wBAAwB,EAAE,KAAK,EAAE,aAAa,CAAC,OAAO,EAAE;SAClE,CAAC;QAEF,KAAK,MAAM,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,QAAQ,EAAE,CAAC;YACxC,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBACrB,cAAc,CAAC,IAAI,CAAC,OAAO,KAAK,GAAG,CAAC,CAAC;gBACrC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;oBACzB,cAAc,CAAC,IAAI,CAAC,MAAM,IAAI,CAAC,IAAI,KAAK,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;gBAC1D,CAAC;YACH,CAAC;QACH,CAAC;QAED,QAAQ,CAAC,IAAI,CAAC,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;IAC3C,CAAC;IAED,QAAQ,CAAC,IAAI,CAAC;;;;;;;;;;;;;;;;;;;+EAmB+D,CAAC,CAAC;IAE/E,QAAQ,CAAC,IAAI,CAAC;;;gMAGgL,CAAC,CAAC;IAEhM,OAAO,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;AAC/B,CAAC"}

package/dist/cli/bench/runner.d.ts

@@ -0,0 +1,14 @@
+import { SpiralEngine } from '../../spiral/engine.js';
+import type { SWETask, BenchConfig, TaskResult, TaskProgressEvent } from './types.js';
+/**
+ * Execute a single SWE-bench task headlessly.
+ * 1. Clone repo at base_commit
+ * 2. (Optional) Initialize/use Spiral Memory
+ * 3. Run agent loop with problem_statement
+ * 4. Extract git diff as patch
+ * 5. Cleanup
+ *
+ * @param sharedSpiral - Pre-initialized SpiralEngine for 'learning' mode (shared across tasks)
+ */
+export declare function runSingleTask(task: SWETask, config: BenchConfig, onProgress?: (event: TaskProgressEvent) => void, sharedSpiral?: SpiralEngine): Promise<TaskResult>;
+//# sourceMappingURL=runner.d.ts.map

package/dist/cli/bench/runner.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../../src/cli/bench/runner.ts"],"names":[],"mappings":"AAYA,OAAO,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AAEtD,OAAO,KAAK,EAAE,OAAO,EAAE,WAAW,EAAE,UAAU,EAAE,iBAAiB,EAAE,MAAM,YAAY,CAAC;AAItF;;;;;;;;;GASG;AACH,wBAAsB,aAAa,CACjC,IAAI,EAAE,OAAO,EACb,MAAM,EAAE,WAAW,EACnB,UAAU,CAAC,EAAE,CAAC,KAAK,EAAE,iBAAiB,KAAK,IAAI,EAC/C,YAAY,CAAC,EAAE,YAAY,GAC1B,OAAO,CAAC,UAAU,CAAC,CA+NrB"}