@claude-flow/cli 3.0.0-alpha.66 → 3.0.0-alpha.68

package/dist/src/commands/hooks.js

@@ -596,7 +596,7 @@ const explainCommand = {
 // Pretrain subcommand
 const pretrainCommand = {
     name: 'pretrain',
-    description: 'Bootstrap intelligence from repository (4-step pipeline)',
+    description: 'Bootstrap intelligence from repository (4-step pipeline + embeddings)',
     options: [
         {
             name: 'path',
@@ -618,17 +618,41 @@ const pretrainCommand = {
             description: 'Skip cached analysis',
             type: 'boolean',
             default: false
+        },
+        {
+            name: 'with-embeddings',
+            description: 'Index documents for semantic search during pretraining',
+            type: 'boolean',
+            default: true
+        },
+        {
+            name: 'embedding-model',
+            description: 'ONNX embedding model',
+            type: 'string',
+            default: 'all-MiniLM-L6-v2',
+            choices: ['all-MiniLM-L6-v2', 'all-mpnet-base-v2']
+        },
+        {
+            name: 'file-types',
+            description: 'File extensions to index (comma-separated)',
+            type: 'string',
+            default: 'ts,js,py,md,json'
         }
     ],
     examples: [
-        { command: 'claude-flow hooks pretrain', description: 'Pretrain from current repository' },
-        { command: 'claude-flow hooks pretrain -p ../my-project --depth deep', description: 'Deep analysis of specific project' }
+        { command: 'claude-flow hooks pretrain', description: 'Pretrain with embeddings indexing' },
+        { command: 'claude-flow hooks pretrain -p ../my-project --depth deep', description: 'Deep analysis of specific project' },
+        { command: 'claude-flow hooks pretrain --no-with-embeddings', description: 'Skip embedding indexing' },
+        { command: 'claude-flow hooks pretrain --file-types ts,tsx,js', description: 'Index only TypeScript/JS files' }
     ],
     action: async (ctx) => {
-        const path = ctx.flags.path || '.';
+        const repoPath = ctx.flags.path || '.';
         const depth = ctx.flags.depth || 'medium';
+        const withEmbeddings = ctx.flags['with-embeddings'] !== false && ctx.flags.withEmbeddings !== false;
+        const embeddingModel = (ctx.flags['embedding-model'] || ctx.flags.embeddingModel || 'all-MiniLM-L6-v2');
+        const fileTypes = (ctx.flags['file-types'] || ctx.flags.fileTypes || 'ts,js,py,md,json');
         output.writeln();
-        output.writeln(output.bold('Pretraining Intelligence (4-Step Pipeline)'));
+        output.writeln(output.bold('Pretraining Intelligence (4-Step Pipeline + Embeddings)'));
         output.writeln();
         const steps = [
             { name: 'RETRIEVE', desc: 'Top-k memory injection with MMR diversity' },
@@ -636,6 +660,10 @@ const pretrainCommand = {
             { name: 'DISTILL', desc: 'Extract strategy memories from trajectories' },
             { name: 'CONSOLIDATE', desc: 'Dedup, detect contradictions, prune old patterns' }
         ];
+        // Add embedding steps if enabled
+        if (withEmbeddings) {
+            steps.push({ name: 'EMBED', desc: `Index documents with ${embeddingModel} (ONNX)` }, { name: 'HYPERBOLIC', desc: 'Project to Poincaré ball for hierarchy preservation' });
+        }
         const spinner = output.createSpinner({ text: 'Starting pretraining...', spinner: 'dots' });
         try {
             spinner.start();
@@ -646,9 +674,12 @@ const pretrainCommand = {
             }
             // Call MCP tool for pretraining
             const result = await callMCPTool('hooks/pretrain', {
-                path,
+                path: repoPath,
                 depth,
                 skipCache: ctx.flags.skipCache || false,
+                withEmbeddings,
+                embeddingModel,
+                fileTypes: fileTypes.split(',').map((t) => t.trim()),
             });
             spinner.succeed('Pretraining completed');
             if (ctx.flags.format === 'json') {
@@ -656,22 +687,31 @@ const pretrainCommand = {
                 return { success: true, data: result };
             }
             output.writeln();
+            // Base stats
+            const tableData = [
+                { metric: 'Files Analyzed', value: result.stats.filesAnalyzed },
+                { metric: 'Patterns Extracted', value: result.stats.patternsExtracted },
+                { metric: 'Strategies Learned', value: result.stats.strategiesLearned },
+                { metric: 'Trajectories Evaluated', value: result.stats.trajectoriesEvaluated },
+                { metric: 'Contradictions Resolved', value: result.stats.contradictionsResolved },
+            ];
+            // Add embedding stats if available
+            if (withEmbeddings && result.stats.documentsIndexed !== undefined) {
+                tableData.push({ metric: 'Documents Indexed', value: result.stats.documentsIndexed }, { metric: 'Embeddings Generated', value: result.stats.embeddingsGenerated || 0 }, { metric: 'Hyperbolic Projections', value: result.stats.hyperbolicProjections || 0 });
+            }
+            tableData.push({ metric: 'Duration', value: `${(result.duration / 1000).toFixed(1)}s` });
             output.printTable({
                 columns: [
                     { key: 'metric', header: 'Metric', width: 30 },
                     { key: 'value', header: 'Value', width: 15, align: 'right' }
                 ],
-                data: [
-                    { metric: 'Files Analyzed', value: result.stats.filesAnalyzed },
-                    { metric: 'Patterns Extracted', value: result.stats.patternsExtracted },
-                    { metric: 'Strategies Learned', value: result.stats.strategiesLearned },
-                    { metric: 'Trajectories Evaluated', value: result.stats.trajectoriesEvaluated },
-                    { metric: 'Contradictions Resolved', value: result.stats.contradictionsResolved },
-                    { metric: 'Duration', value: `${(result.duration / 1000).toFixed(1)}s` }
-                ]
+                data: tableData
             });
             output.writeln();
             output.printSuccess('Repository intelligence bootstrapped successfully');
+            if (withEmbeddings) {
+                output.writeln(output.dim('  Semantic search enabled: Use "embeddings search -q <query>" to search'));
+            }
             output.writeln(output.dim('  Next step: Run "claude-flow hooks build-agents" to generate optimized configs'));
             return { success: true, data: result };
         }