opencode-swarm-plugin 0.21.0 → 0.23.0

package/dist/eval-capture.d.ts

@@ -0,0 +1,206 @@
+/**
+ * Eval Data Capture - Captures real swarm execution data for evals
+ *
+ * Records decomposition inputs, outputs, and outcomes to JSONL files
+ * that can be used as ground truth for Evalite evals.
+ *
+ * Data flow:
+ * 1. swarm_decompose captures: task, context, generated decomposition
+ * 2. swarm_complete captures: outcome signals per subtask
+ * 3. swarm_record_outcome captures: learning signals
+ * 4. Human feedback (optional): accept/reject/modify
+ *
+ * @module eval-capture
+ */
+import { z } from "zod";
+/**
+ * Subtask outcome - what actually happened
+ */
+export declare const SubtaskOutcomeSchema: z.ZodObject<{
+    bead_id: z.ZodString;
+    title: z.ZodString;
+    planned_files: z.ZodArray<z.ZodString>;
+    actual_files: z.ZodArray<z.ZodString>;
+    duration_ms: z.ZodNumber;
+    error_count: z.ZodNumber;
+    retry_count: z.ZodNumber;
+    success: z.ZodBoolean;
+    failure_mode: z.ZodOptional<z.ZodString>;
+}, z.core.$strip>;
+export type SubtaskOutcome = z.infer<typeof SubtaskOutcomeSchema>;
+/**
+ * Complete eval record - input, output, and outcome
+ */
+export declare const EvalRecordSchema: z.ZodObject<{
+    id: z.ZodString;
+    timestamp: z.ZodString;
+    project_path: z.ZodString;
+    task: z.ZodString;
+    context: z.ZodOptional<z.ZodString>;
+    strategy: z.ZodEnum<{
+        "file-based": "file-based";
+        "feature-based": "feature-based";
+        "risk-based": "risk-based";
+        auto: "auto";
+    }>;
+    max_subtasks: z.ZodNumber;
+    epic_title: z.ZodString;
+    epic_description: z.ZodOptional<z.ZodString>;
+    subtasks: z.ZodArray<z.ZodObject<{
+        title: z.ZodString;
+        description: z.ZodOptional<z.ZodString>;
+        files: z.ZodArray<z.ZodString>;
+        dependencies: z.ZodOptional<z.ZodArray<z.ZodNumber>>;
+        estimated_complexity: z.ZodOptional<z.ZodNumber>;
+    }, z.core.$strip>>;
+    outcomes: z.ZodOptional<z.ZodArray<z.ZodObject<{
+        bead_id: z.ZodString;
+        title: z.ZodString;
+        planned_files: z.ZodArray<z.ZodString>;
+        actual_files: z.ZodArray<z.ZodString>;
+        duration_ms: z.ZodNumber;
+        error_count: z.ZodNumber;
+        retry_count: z.ZodNumber;
+        success: z.ZodBoolean;
+        failure_mode: z.ZodOptional<z.ZodString>;
+    }, z.core.$strip>>>;
+    overall_success: z.ZodOptional<z.ZodBoolean>;
+    total_duration_ms: z.ZodOptional<z.ZodNumber>;
+    total_errors: z.ZodOptional<z.ZodNumber>;
+    human_accepted: z.ZodOptional<z.ZodBoolean>;
+    human_modified: z.ZodOptional<z.ZodBoolean>;
+    human_notes: z.ZodOptional<z.ZodString>;
+    file_overlap_count: z.ZodOptional<z.ZodNumber>;
+    scope_accuracy: z.ZodOptional<z.ZodNumber>;
+    time_balance_ratio: z.ZodOptional<z.ZodNumber>;
+}, z.core.$strip>;
+export type EvalRecord = z.infer<typeof EvalRecordSchema>;
+/**
+ * Partial record for in-progress capture
+ */
+export type PartialEvalRecord = Partial<EvalRecord> & {
+    id: string;
+    timestamp: string;
+    task: string;
+};
+/**
+ * Default path for eval data
+ */
+export declare const DEFAULT_EVAL_DATA_PATH = ".opencode/eval-data.jsonl";
+/**
+ * Get the eval data file path for a project
+ */
+export declare function getEvalDataPath(projectPath: string): string;
+/**
+ * Ensure the eval data directory exists
+ */
+export declare function ensureEvalDataDir(projectPath: string): void;
+/**
+ * Append an eval record to the JSONL file
+ */
+export declare function appendEvalRecord(projectPath: string, record: EvalRecord | PartialEvalRecord): void;
+/**
+ * Read all eval records from a project
+ */
+export declare function readEvalRecords(projectPath: string): EvalRecord[];
+/**
+ * Read partial records (for updating in-progress records)
+ */
+export declare function readPartialRecords(projectPath: string): PartialEvalRecord[];
+/**
+ * Update an existing record by ID
+ */
+export declare function updateEvalRecord(projectPath: string, id: string, updates: Partial<EvalRecord>): boolean;
+/**
+ * Start capturing a decomposition
+ *
+ * Called when swarm_decompose generates a decomposition.
+ * Creates a partial record that will be completed when outcomes arrive.
+ */
+export declare function captureDecomposition(params: {
+    epicId: string;
+    projectPath: string;
+    task: string;
+    context?: string;
+    strategy: "file-based" | "feature-based" | "risk-based" | "auto";
+    maxSubtasks: number;
+    epicTitle: string;
+    epicDescription?: string;
+    subtasks: Array<{
+        title: string;
+        description?: string;
+        files: string[];
+        dependencies?: number[];
+        estimated_complexity?: number;
+    }>;
+}): PartialEvalRecord;
+/**
+ * Capture a subtask outcome
+ *
+ * Called when swarm_complete finishes a subtask.
+ * Updates the in-progress record with outcome data.
+ */
+export declare function captureSubtaskOutcome(params: {
+    epicId: string;
+    projectPath: string;
+    beadId: string;
+    title: string;
+    plannedFiles: string[];
+    actualFiles: string[];
+    durationMs: number;
+    errorCount: number;
+    retryCount: number;
+    success: boolean;
+    failureMode?: string;
+}): void;
+/**
+ * Finalize an eval record
+ *
+ * Called when all subtasks are complete.
+ * Computes aggregate metrics and marks record as complete.
+ */
+export declare function finalizeEvalRecord(params: {
+    epicId: string;
+    projectPath: string;
+}): EvalRecord | null;
+/**
+ * Capture human feedback on a decomposition
+ */
+export declare function captureHumanFeedback(params: {
+    epicId: string;
+    projectPath: string;
+    accepted: boolean;
+    modified: boolean;
+    notes?: string;
+}): void;
+/**
+ * Export eval records as Evalite-compatible test cases
+ *
+ * Filters to only complete records with outcomes.
+ */
+export declare function exportForEvalite(projectPath: string): Array<{
+    input: {
+        task: string;
+        context?: string;
+    };
+    expected: {
+        minSubtasks: number;
+        maxSubtasks: number;
+        requiredFiles?: string[];
+        overallSuccess?: boolean;
+    };
+    actual: EvalRecord;
+}>;
+/**
+ * Get statistics about captured eval data
+ */
+export declare function getEvalDataStats(projectPath: string): {
+    totalRecords: number;
+    completeRecords: number;
+    successRate: number;
+    avgSubtasks: number;
+    avgDurationMs: number;
+    avgScopeAccuracy: number;
+    avgTimeBalance: number;
+};
+//# sourceMappingURL=eval-capture.d.ts.map

package/dist/eval-capture.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"eval-capture.d.ts","sourceRoot":"","sources":["../src/eval-capture.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AACH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAQxB;;GAEG;AACH,eAAO,MAAM,oBAAoB;;;;;;;;;;iBAmB/B,CAAC;AACH,MAAM,MAAM,cAAc,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,oBAAoB,CAAC,CAAC;AAElE;;GAEG;AACH,eAAO,MAAM,gBAAgB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA2D3B,CAAC;AACH,MAAM,MAAM,UAAU,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,gBAAgB,CAAC,CAAC;AAE1D;;GAEG;AACH,MAAM,MAAM,iBAAiB,GAAG,OAAO,CAAC,UAAU,CAAC,GAAG;IACpD,EAAE,EAAE,MAAM,CAAC;IACX,SAAS,EAAE,MAAM,CAAC;IAClB,IAAI,EAAE,MAAM,CAAC;CACd,CAAC;AAMF;;GAEG;AACH,eAAO,MAAM,sBAAsB,8BAA8B,CAAC;AAElE;;GAEG;AACH,wBAAgB,eAAe,CAAC,WAAW,EAAE,MAAM,GAAG,MAAM,CAE3D;AAED;;GAEG;AACH,wBAAgB,iBAAiB,CAAC,WAAW,EAAE,MAAM,GAAG,IAAI,CAM3D;AAED;;GAEG;AACH,wBAAgB,gBAAgB,CAC9B,WAAW,EAAE,MAAM,EACnB,MAAM,EAAE,UAAU,GAAG,iBAAiB,GACrC,IAAI,CAKN;AAED;;GAEG;AACH,wBAAgB,eAAe,CAAC,WAAW,EAAE,MAAM,GAAG,UAAU,EAAE,CAajE;AAED;;GAEG;AACH,wBAAgB,kBAAkB,CAAC,WAAW,EAAE,MAAM,GAAG,iBAAiB,EAAE,CAU3E;AAED;;GAEG;AACH,wBAAgB,gBAAgB,CAC9B,WAAW,EAAE,MAAM,EACnB,EAAE,EAAE,MAAM,EACV,OAAO,EAAE,OAAO,CAAC,UAAU,CAAC,GAC3B,OAAO,CAgBT;AAWD;;;;;GAKG;AACH,wBAAgB,oBAAoB,CAAC,MAAM,EAAE;IAC3C,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,MAAM,CAAC;IACpB,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,YAAY,GAAG,eAAe,GAAG,YAAY,GAAG,MAAM,CAAC;IACjE,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,MAAM,CAAC;IAClB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,QAAQ,EAAE,KAAK,CAAC;QACd,KAAK,EAAE,MAAM,CAAC;QACd,WAAW,CAAC,EAAE,MAAM,CAAC;QACrB,KAAK,EAAE,MAAM,EAAE,CAAC;QAChB,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;QACxB,oBAAoB,CAAC,EAAE,MAAM,CAAC;KAC/B,CAAC,CAAC;CACJ,GAAG,iBAAiB,CAsBpB;AAED;;;;;GAKG;AACH,wBAAgB,qBAAqB,CAAC,MAAM,EAAE;IAC5C,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,MAAM,CAAC;IACpB,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB,GAAG,IAAI,CAwBP;AAED;;;;;GAKG;AACH,wBAAgB,kBAAkB,CAAC,MAAM,EAAE;IACzC,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,MAAM,CAAC;CACrB,GAAG,UAAU,GAAG,IAAI,CAoDpB;AAED;;GAEG;AACH,wBAAgB,oBAAoB,CAAC,MAAM,EAAE;IAC3C,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,OAAO,CAAC;IAClB,QAAQ,EAAE,OAAO,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB,GAAG,IAAI,CAMP;AAMD;;;;GAIG;AACH,wBAAgB,gBAAgB,CAAC,WAAW,EAAE,MAAM,GAAG,KAAK,CAAC;IAC3D,KAAK,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IAC1C,QAAQ,EAAE;QACR,WAAW,EAAE,MAAM,CAAC;QACpB,WAAW,EAAE,MAAM,CAAC;QACpB,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;QACzB,cAAc,CAAC,EAAE,OAAO,CAAC;KAC1B,CAAC;IACF,MAAM,EAAE,UAAU,CAAC;CACpB,CAAC,CAkBD;AAED;;GAEG;AACH,wBAAgB,gBAAgB,CAAC,WAAW,EAAE,MAAM,GAAG;IACrD,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,EAAE,MAAM,CAAC;IACxB,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,EAAE,MAAM,CAAC;IACtB,gBAAgB,EAAE,MAAM,CAAC;IACzB,cAAc,EAAE,MAAM,CAAC;CACxB,CAuCA"}