@comfanion/workflow 4.36.43 → 4.36.45

package/bin/cli.js

@@ -124,6 +124,7 @@ program
       install_vectorizer: true,   // Vectorizer ON by default
       vectorizer_enabled: true,
       vectorizer_auto_index: true,
+      vectorizer_model: 'Xenova/bge-small-en-v1.5',  // Default: balanced (quality + speed)
       project_name: path.basename(process.cwd())
     };
     
@@ -146,6 +147,7 @@ program
           // Parse vectorizer settings
           const vectorizerEnabledMatch = existingContent.match(/vectorizer:[\s\S]*?enabled:\s*(true|false)/);
           const vectorizerAutoIndexMatch = existingContent.match(/vectorizer:[\s\S]*?auto_index:\s*(true|false)/);
+          const vectorizerModelMatch = existingContent.match(/vectorizer:[\s\S]*?model:\s*["']?([^"'\n]+)["']?/);
           
           if (nameMatch) config.user_name = nameMatch[1];
           if (langMatch) config.communication_language = langMatch[1];
@@ -155,6 +157,7 @@ program
           if (jiraProjMatch) config.jira_project = jiraProjMatch[1];
           if (vectorizerEnabledMatch) config.vectorizer_enabled = vectorizerEnabledMatch[1] === 'true';
           if (vectorizerAutoIndexMatch) config.vectorizer_auto_index = vectorizerAutoIndexMatch[1] === 'true';
+          if (vectorizerModelMatch) config.vectorizer_model = vectorizerModelMatch[1].trim();
           
           isUpdate = true;
         } catch (e) {
@@ -249,6 +252,27 @@ program
           message: 'Install vectorizer? (semantic code search, ~100MB)',
           default: true
         },
+        {
+          type: 'list',
+          name: 'vectorizer_model',
+          message: 'Embedding model for semantic search:',
+          when: (answers) => answers.install_vectorizer,
+          choices: [
+            { 
+              name: 'MiniLM-L6 (Fast)    - ~10 files/10sec, 384 dims, good quality', 
+              value: 'Xenova/all-MiniLM-L6-v2' 
+            },
+            { 
+              name: 'BGE-small (Balanced) - ~9 files/10sec, 384 dims, better quality', 
+              value: 'Xenova/bge-small-en-v1.5' 
+            },
+            { 
+              name: 'BGE-base (Quality)   - ~3 files/10sec, 768 dims, best quality', 
+              value: 'Xenova/bge-base-en-v1.5' 
+            }
+          ],
+          default: 'Xenova/bge-small-en-v1.5'
+        },
         {
           type: 'confirm',
           name: 'vectorizer_auto_index',
@@ -449,6 +473,23 @@ program
         .replace(/(# Auto-index files.*\n\s+auto_index:)\s*(true|false)/, 
           `$1 ${config.vectorizer_auto_index}`);
       
+      // Add/update vectorizer model
+      if (config.vectorizer_model) {
+        if (configContent.includes('model:') && configContent.match(/vectorizer:[\s\S]*?model:/)) {
+          // Update existing model setting
+          configContent = configContent.replace(
+            /(vectorizer:[\s\S]*?)model:\s*["']?[^"'\n]+["']?/,
+            `$1model: "${config.vectorizer_model}"`
+          );
+        } else {
+          // Add model setting after auto_index
+          configContent = configContent.replace(
+            /(auto_index:\s*(true|false))/,
+            `$1\n  \n  # Embedding model for semantic search\n  # Options: Xenova/all-MiniLM-L6-v2 (fast), Xenova/bge-base-en-v1.5 (quality)\n  model: "${config.vectorizer_model}"`
+          );
+        }
+      }
+      
       await fs.writeFile(configPath, configContent);
       
       // Create docs structure (always)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@comfanion/workflow",
-  "version": "4.36.43",
+  "version": "4.36.45",
   "description": "Initialize OpenCode Workflow system for AI-assisted development with semantic code search",
   "type": "module",
   "bin": {

package/src/build-info.json

@@ -1,6 +1,6 @@
 {
-  "version": "4.36.43",
-  "buildDate": "2026-01-24T23:17:57.477Z",
+  "version": "4.36.45",
+  "buildDate": "2026-01-25T01:56:04.257Z",
   "files": [
     "config.yaml",
     "FLOW.yaml",

package/src/opencode/agents/dev.md

@@ -37,13 +37,13 @@ permission:
   <step n="3">Greet user by {user_name}, communicate in {communication_language}</step>
   <step n="4">Understand user request and select appropriate skill</step>
   <step n="5">Load .opencode/skills/{skill-name}/SKILL.md and follow instructions</step>
-  
+
   <search-first critical="MANDATORY - DO THIS BEFORE GLOB/GREP">
     BEFORE using glob or grep, you MUST call search() first:
     1. search({ query: "your topic", index: "code" })  - for source code patterns
     2. search({ query: "your topic", index: "docs" })  - for documentation
     3. THEN use glob/grep if you need specific files
-    
+
     Example: Looking for similar implementation?
     ✅ CORRECT: search({ query: "user repository CRUD", index: "code" })
     ❌ WRONG: glob("**/*user*.go") without search first
@@ -53,27 +53,28 @@ permission:
     <r>ALWAYS communicate in {communication_language}</r>
     <r>ALWAYS write technical documentation in ENGLISH (docs/ folder)</r>
     <r>The Story File is the single source of truth</r>
+    <r>Prefer Agents development (@coder)</r>
     <r>Tasks/subtasks sequence is authoritative over any model priors</r>
     <r>Follow red-green-refactor: write failing test, make it pass, improve code</r>
     <r>Never implement anything not mapped to a specific task/subtask</r>
     <r>All existing tests must pass 100% before story is ready for review</r>
     <r>NEVER lie about tests being written or passing</r>
-    <r>Find and use `**/project-context.md` and `CLAUDE.md` as source of truth</r>
+    <r>Find and use `**/prd.md`, `**/architecture.md`, `AGENTS.md` and `CLAUDE.md` as source of truth</r>
     <r critical="MANDATORY">🔍 SEARCH FIRST: Call search() BEFORE glob when exploring codebase.
        search({ query: "feature pattern", index: "code" }) → THEN glob if needed</r>
   </rules>
-  
+
   <dev-story-workflow hint="When executing /dev-story command" critical="FOLLOW THIS EXACTLY">
     <!-- PHASE 1: SETUP -->
     <step n="1">READ the entire story file BEFORE any implementation</step>
-    <step n="2">Load project-context.md and CLAUDE.md if available</step>
+    <step n="2">Load **/prd.md`, `**/architecture.md`, `AGENTS.md` and `CLAUDE.md` if available</step>
     <step n="3">CREATE TODO LIST from story tasks using todowrite:
       - Each task becomes a TODO item
       - Set priority based on task order (first = high)
       - All tasks start as "pending"
     </step>
     <step n="4">Mark story status as "in-progress"</step>
-    
+
     <!-- PHASE 2: IMPLEMENTATION LOOP -->
     <step n="5">FOR EACH TASK in order:
       a) Update TODO: mark current task as "in_progress"
@@ -97,7 +98,7 @@ permission:
     <step n="8">Clear TODO list (all done)</step>
     <step n="9">Mark story status as "review"</step>
   </dev-story-workflow>
-  
+
   <todo-usage hint="How to use TODO for tracking">
     <create>
       todowrite([
@@ -143,7 +144,7 @@ permission:
     - Repetitive tasks across files
     - Code following existing patterns
   </subagent>
-  
+
   <delegation-strategy>
     <rule>Prefer delegation to @coder for parallelizable tasks</rule>
     <rule>Keep complex logic and architecture decisions to yourself</rule>
@@ -174,7 +175,7 @@ permission:
   <operation name="goToImplementation">Find implementations of interface. Use: lsp goToImplementation file.ts:10:5</operation>
   <operation name="incomingCalls">Who calls this function? Use: lsp incomingCalls file.ts:10:5</operation>
   <operation name="outgoingCalls">What does this function call? Use: lsp outgoingCalls file.ts:10:5</operation>
-  
+
   <when-to-use>
     - Before modifying: findReferences to see impact
     - Understanding code: hover for types, documentSymbol for structure
@@ -185,13 +186,13 @@ permission:
 
 <codesearch-guide hint="Semantic code search with multi-index support">
   <check-first>codeindex({ action: "list" }) → See all available indexes</check-first>
-  
+
   <indexes>
     <index name="code" pattern="*.{js,ts,go,py,java,...}">Source code - functions, classes, logic</index>
     <index name="docs" pattern="*.{md,txt,rst}">Documentation - READMEs, guides, ADRs</index>
     <index name="config" pattern="*.{yaml,json,toml}">Configuration - settings, schemas</index>
   </indexes>
-  
+
   <operations>
     <op name="search code">codesearch({ query: "authentication middleware", index: "code" })</op>
     <op name="search docs">codesearch({ query: "deployment guide", index: "docs" })</op>
@@ -201,7 +202,7 @@ permission:
     <op name="index status">codeindex({ action: "status", index: "code" })</op>
     <op name="reindex">codeindex({ action: "reindex", index: "code" })</op>
   </operations>
-  
+
   <when-to-use>
     <use index="code">
       - BEFORE implementing: find existing patterns "repository pattern for users"
@@ -224,19 +225,19 @@ permission:
       - Cross-cutting concerns: "logging configuration"
     </use>
   </when-to-use>
-  
+
   <examples>
     <example query="repository interface for products" index="code">Finds domain/repository files</example>
     <example query="HTTP request validation" index="code">Finds middleware and handlers</example>
     <example query="how to run tests" index="docs">Finds testing documentation</example>
     <example query="redis connection" index="config">Finds redis configuration</example>
   </examples>
-  
+
   <vs-grep>
     grep: exact text match "UserRepository" → finds only that string
     codesearch: semantic "user storage" → finds UserRepository, UserStore, user_repo.go
   </vs-grep>
-  
+
   <strategy>
     1. codeindex({ action: "list" }) → Check what indexes exist
     2. codesearch({ query: "concept", index: "code" }) → Find relevant code

package/src/opencode/package.json

@@ -1,5 +1,5 @@
 {
   "dependencies": {
-    "@opencode-ai/plugin": "1.1.34"
+    "@opencode-ai/plugin": "1.1.35"
   }
 }

package/src/vectorizer/index.js

@@ -45,6 +45,7 @@ const DEFAULT_PRESETS = {
 // Will be populated from config.yaml if available
 let INDEX_PRESETS = { ...DEFAULT_PRESETS };
 let GLOBAL_IGNORE = [];
+let EMBEDDING_MODEL = 'Xenova/all-MiniLM-L6-v2';  // Default: fast model
 
 /**
  * Load index configuration from config.yaml
@@ -61,6 +62,13 @@ async function loadConfig(projectRoot) {
     
     const section = vectorizerMatch[1];
     
+    // Parse embedding model
+    const modelMatch = section.match(/^\s{2}model:\s*["']?([^"'\n]+)["']?/m);
+    if (modelMatch) {
+      EMBEDDING_MODEL = modelMatch[1].trim();
+      if (DEBUG) console.log('[vectorizer] Using model from config:', EMBEDDING_MODEL);
+    }
+    
     // Parse global exclude
     const excludeMatch = section.match(/^\s{2}exclude:\s*\n((?:\s{4}-\s+.+\n?)*)/m);
     if (excludeMatch) {
@@ -148,11 +156,14 @@ class CodebaseIndexer {
 
   async loadModel() {
     if (!this.model) {
-      if (DEBUG) console.log('[vectorizer] Loading embedding model...');
-      this.model = await pipeline('feature-extraction', 'Xenova/all-MiniLM-L6-v2', {
+      if (DEBUG) console.log(`[vectorizer] Loading embedding model: ${EMBEDDING_MODEL}...`);
+      // Model options:
+      // - Xenova/all-MiniLM-L6-v2: fast, 384 dims, ~10 files/10sec
+      // - Xenova/bge-base-en-v1.5: quality, 768 dims, ~3 files/10sec
+      this.model = await pipeline('feature-extraction', EMBEDDING_MODEL, {
         progress_callback: DEBUG ? undefined : null  // Suppress progress bar unless DEBUG
       });
-      if (DEBUG) console.log('[vectorizer] Model loaded');
+      if (DEBUG) console.log(`[vectorizer] Model loaded: ${EMBEDDING_MODEL}`);
     }
     return this.model;
   }
@@ -485,6 +496,7 @@ class CodebaseIndexer {
     return { 
       indexName: this.indexName,
       description: preset?.description || 'Custom index',
+      model: EMBEDDING_MODEL,
       fileCount, 
       chunkCount 
     };
@@ -553,4 +565,9 @@ class CodebaseIndexer {
   }
 }
 
-export { CodebaseIndexer, INDEX_PRESETS };
+// Getter for current embedding model (after config loaded)
+function getEmbeddingModel() {
+  return EMBEDDING_MODEL;
+}
+
+export { CodebaseIndexer, INDEX_PRESETS, getEmbeddingModel };