chainlesschain 0.40.2 → 0.40.3

package/src/lib/slot-filler.js

@@ -0,0 +1,465 @@
+/**
+ * CLI SlotFiller — parameter slot filling for agentic workflows
+ *
+ * Ported from desktop-app-vue/src/main/ai-engine/slot-filler.js
+ * Adapted for CLI with InteractionAdapter support for terminal and WebSocket modes.
+ *
+ * Flow: detect missing slots → infer from context → ask user → LLM inference (optional) → validate
+ */
+
+import { EventEmitter } from "events";
+
+/**
+ * Required slots per intent type — must be filled before execution.
+ */
+const REQUIRED_SLOTS = {
+  create_file: ["fileType", "path"],
+  edit_file: ["target"],
+  deploy: ["platform"],
+  refactor: ["scope"],
+  test: ["target"],
+  analyze: ["target"],
+  search: ["query"],
+  install: ["package"],
+  generate: ["type"],
+};
+
+/**
+ * Optional slots per intent type — filled opportunistically.
+ */
+const OPTIONAL_SLOTS = {
+  create_file: ["template", "content", "overwrite"],
+  edit_file: ["action", "position"],
+  deploy: ["env", "branch", "dryRun"],
+  refactor: ["strategy", "dryRun"],
+  test: ["framework", "coverage", "watch"],
+  analyze: ["outputFormat", "depth"],
+  search: ["directory", "fileType"],
+  install: ["version", "dev"],
+  generate: ["output", "template"],
+};
+
+/**
+ * Slot prompts — question text, type, and options for user interaction.
+ */
+const SLOT_PROMPTS = {
+  fileType: {
+    question: "What type of file do you want to create?",
+    type: "select",
+    options: [
+      { name: "JavaScript", value: "js" },
+      { name: "TypeScript", value: "ts" },
+      { name: "Python", value: "py" },
+      { name: "JSON", value: "json" },
+      { name: "Markdown", value: "md" },
+      { name: "Other", value: "other" },
+    ],
+  },
+  path: {
+    question: "Where should the file be created? (path)",
+    type: "input",
+  },
+  target: {
+    question: "Which file or directory should be targeted?",
+    type: "input",
+  },
+  platform: {
+    question: "Which platform are you deploying to?",
+    type: "select",
+    options: [
+      { name: "Docker", value: "docker" },
+      { name: "Vercel", value: "vercel" },
+      { name: "AWS", value: "aws" },
+      { name: "Local", value: "local" },
+    ],
+  },
+  scope: {
+    question: "What is the scope of the refactoring?",
+    type: "select",
+    options: [
+      { name: "Single file", value: "file" },
+      { name: "Directory", value: "directory" },
+      { name: "Module", value: "module" },
+      { name: "Full project", value: "project" },
+    ],
+  },
+  type: {
+    question: "What do you want to generate?",
+    type: "select",
+    options: [
+      { name: "Component", value: "component" },
+      { name: "Test", value: "test" },
+      { name: "API endpoint", value: "api" },
+      { name: "Config file", value: "config" },
+    ],
+  },
+  query: {
+    question: "What are you searching for?",
+    type: "input",
+  },
+  package: {
+    question: "Which package do you want to install?",
+    type: "input",
+  },
+  framework: {
+    question: "Which test framework?",
+    type: "select",
+    options: [
+      { name: "Vitest", value: "vitest" },
+      { name: "Jest", value: "jest" },
+      { name: "Mocha", value: "mocha" },
+    ],
+  },
+};
+
+/**
+ * File extension → fileType mapping
+ */
+const EXT_TO_FILE_TYPE = {
+  ".js": "js",
+  ".mjs": "js",
+  ".cjs": "js",
+  ".ts": "ts",
+  ".tsx": "ts",
+  ".py": "py",
+  ".json": "json",
+  ".md": "md",
+  ".html": "html",
+  ".css": "css",
+  ".vue": "vue",
+  ".yaml": "yaml",
+  ".yml": "yaml",
+};
+
+export class CLISlotFiller extends EventEmitter {
+  /**
+   * @param {object} options
+   * @param {function} [options.llmChat] - LLM chat function for inference
+   * @param {object} [options.db] - Database for history
+   * @param {import("./interaction-adapter.js").InteractionAdapter} options.interaction
+   */
+  constructor({ llmChat, db, interaction }) {
+    super();
+    this.llmChat = llmChat || null;
+    this.db = db || null;
+    this.interaction = interaction;
+  }
+
+  /**
+   * Main flow: detect missing required slots → infer → ask user → validate.
+   *
+   * @param {{ type: string, entities: object }} intent - parsed intent
+   * @param {object} context - project context (cwd, files, etc.)
+   * @returns {Promise<{ entities: object, validation: object, filledSlots: string[], missingRequired: string[] }>}
+   */
+  async fillSlots(intent, context = {}) {
+    const intentType = intent.type || "unknown";
+    const entities = { ...(intent.entities || {}) };
+    const filledSlots = [];
+
+    const requiredSlots = REQUIRED_SLOTS[intentType] || [];
+    const optionalSlots = OPTIONAL_SLOTS[intentType] || [];
+
+    // Step 1: Infer from context
+    for (const slot of [...requiredSlots, ...optionalSlots]) {
+      if (entities[slot]) continue; // Already filled
+
+      const inferred = this.inferFromContext(slot, context);
+      if (inferred !== null) {
+        entities[slot] = inferred;
+        filledSlots.push(slot);
+        this.emit("slot-inferred", {
+          slot,
+          value: inferred,
+          source: "context",
+        });
+      }
+    }
+
+    // Step 2: Ask user for missing required slots
+    for (const slot of requiredSlots) {
+      if (entities[slot]) continue;
+
+      try {
+        const value = await this.askUser(slot);
+        if (value) {
+          entities[slot] = value;
+          filledSlots.push(slot);
+          this.emit("slot-filled", { slot, value, source: "user" });
+        }
+      } catch (_err) {
+        // User cancelled or timeout — leave slot empty
+      }
+    }
+
+    // Step 3: LLM inference for remaining optional slots
+    if (this.llmChat) {
+      const missingOptional = optionalSlots.filter((s) => !entities[s]);
+      if (missingOptional.length > 0) {
+        try {
+          const inferred = await this.inferWithLLM(
+            missingOptional,
+            context,
+            entities,
+            intentType,
+          );
+          for (const [slot, value] of Object.entries(inferred)) {
+            if (value && !entities[slot]) {
+              entities[slot] = value;
+              filledSlots.push(slot);
+              this.emit("slot-inferred", { slot, value, source: "llm" });
+            }
+          }
+        } catch (_err) {
+          // LLM inference failure is non-critical
+        }
+      }
+    }
+
+    // Step 4: Learn from user preferences
+    if (this.db) {
+      for (const slot of requiredSlots) {
+        if (!entities[slot]) {
+          try {
+            const preference = await this.learnUserPreference(intentType, slot);
+            if (preference) {
+              entities[slot] = preference;
+              filledSlots.push(slot);
+              this.emit("slot-inferred", {
+                slot,
+                value: preference,
+                source: "preference",
+              });
+            }
+          } catch (_err) {
+            // Non-critical
+          }
+        }
+      }
+    }
+
+    const validation = this.validateSlots(intentType, entities);
+
+    return {
+      entities,
+      validation,
+      filledSlots,
+      missingRequired: validation.missingRequired,
+    };
+  }
+
+  /**
+   * Infer a slot value from context (rules-based).
+   */
+  inferFromContext(slotName, context) {
+    switch (slotName) {
+      case "fileType": {
+        if (context.currentFile) {
+          const ext = context.currentFile.match(/\.[^.]+$/)?.[0];
+          if (ext && EXT_TO_FILE_TYPE[ext]) {
+            return EXT_TO_FILE_TYPE[ext];
+          }
+        }
+        return null;
+      }
+
+      case "target": {
+        if (context.currentFile) return context.currentFile;
+        if (context.selectedText) return context.selectedText;
+        return null;
+      }
+
+      case "path": {
+        if (context.cwd) return context.cwd;
+        return null;
+      }
+
+      case "platform": {
+        if (context.hasDockerfile) return "docker";
+        if (context.hasVercelConfig) return "vercel";
+        return null;
+      }
+
+      case "framework": {
+        if (context.hasVitest) return "vitest";
+        if (context.hasJest) return "jest";
+        return null;
+      }
+
+      case "directory": {
+        return context.cwd || null;
+      }
+
+      default:
+        return null;
+    }
+  }
+
+  /**
+   * Ask the user to fill a slot via the interaction adapter.
+   */
+  async askUser(slotName) {
+    const prompt = SLOT_PROMPTS[slotName];
+    if (!prompt) {
+      return this.interaction.askInput(
+        `Please provide a value for "${slotName}":`,
+      );
+    }
+
+    if (prompt.type === "select" && prompt.options) {
+      return this.interaction.askSelect(prompt.question, prompt.options);
+    }
+
+    return this.interaction.askInput(prompt.question);
+  }
+
+  /**
+   * Use LLM to infer optional slot values (low temperature).
+   */
+  async inferWithLLM(slots, context, currentEntities, intentType) {
+    if (!this.llmChat) return {};
+
+    const prompt = `Given the following context, infer reasonable values for these parameters.
+
+Intent type: ${intentType}
+Already known: ${JSON.stringify(currentEntities)}
+Working directory: ${context.cwd || "unknown"}
+Parameters to infer: ${slots.join(", ")}
+
+Respond with a JSON object mapping parameter names to inferred values.
+Only include parameters where you have high confidence. Use null for uncertain ones.
+Keep values concise (single words or short strings).`;
+
+    try {
+      const response = await this.llmChat([
+        {
+          role: "system",
+          content:
+            "You are a parameter inference assistant. Respond only with valid JSON.",
+        },
+        { role: "user", content: prompt },
+      ]);
+
+      const content = response?.message?.content || response?.content || "";
+      const jsonMatch = content.match(/\{[\s\S]*\}/);
+      if (jsonMatch) {
+        const parsed = JSON.parse(jsonMatch[0]);
+        // Filter out null values
+        const result = {};
+        for (const [key, value] of Object.entries(parsed)) {
+          if (value !== null && value !== undefined && slots.includes(key)) {
+            result[key] = String(value);
+          }
+        }
+        return result;
+      }
+    } catch (_err) {
+      // Non-critical
+    }
+
+    return {};
+  }
+
+  /**
+   * Validate that all required slots are filled.
+   */
+  validateSlots(intentType, entities) {
+    const required = REQUIRED_SLOTS[intentType] || [];
+    const missingRequired = required.filter((s) => !entities[s]);
+    const total = required.length;
+    const filled = total - missingRequired.length;
+
+    return {
+      valid: missingRequired.length === 0,
+      missingRequired,
+      completeness: total > 0 ? Math.round((filled / total) * 100) : 100,
+    };
+  }
+
+  /**
+   * Learn user preferences from past slot-filling history.
+   */
+  async learnUserPreference(intentType, slot) {
+    if (!this.db) return null;
+
+    try {
+      this._ensureHistoryTable();
+      const rows = this.db
+        .prepare(
+          `SELECT slot_value FROM slot_filling_history
+           WHERE intent_type = ? AND slot_name = ?
+           ORDER BY created_at DESC LIMIT 10`,
+        )
+        .all(intentType, slot);
+
+      if (rows.length < 2) return null;
+
+      // Find most common value
+      const counts = {};
+      for (const row of rows) {
+        counts[row.slot_value] = (counts[row.slot_value] || 0) + 1;
+      }
+      const sorted = Object.entries(counts).sort((a, b) => b[1] - a[1]);
+      if (sorted[0] && sorted[0][1] >= 2) {
+        return sorted[0][0];
+      }
+    } catch (_err) {
+      // Non-critical
+    }
+
+    return null;
+  }
+
+  /**
+   * Record slot filling history for preference learning.
+   */
+  recordHistory(intentType, entities) {
+    if (!this.db) return;
+
+    try {
+      this._ensureHistoryTable();
+      const stmt = this.db.prepare(
+        `INSERT INTO slot_filling_history (intent_type, slot_name, slot_value, created_at)
+         VALUES (?, ?, ?, datetime('now'))`,
+      );
+      for (const [slot, value] of Object.entries(entities)) {
+        if (value) {
+          stmt.run(intentType, slot, String(value));
+        }
+      }
+    } catch (_err) {
+      // Non-critical
+    }
+  }
+
+  _ensureHistoryTable() {
+    if (this._tableCreated) return;
+    this.db.exec(`
+      CREATE TABLE IF NOT EXISTS slot_filling_history (
+        id INTEGER PRIMARY KEY AUTOINCREMENT,
+        intent_type TEXT NOT NULL,
+        slot_name TEXT NOT NULL,
+        slot_value TEXT NOT NULL,
+        created_at TEXT DEFAULT (datetime('now'))
+      )
+    `);
+    this._tableCreated = true;
+  }
+
+  /**
+   * Get slot definitions for an intent type.
+   */
+  static getSlotDefinitions(intentType) {
+    return {
+      required: REQUIRED_SLOTS[intentType] || [],
+      optional: OPTIONAL_SLOTS[intentType] || [],
+    };
+  }
+
+  /**
+   * Get all supported intent types.
+   */
+  static getSupportedIntents() {
+    return Object.keys(REQUIRED_SLOTS);
+  }
+}

package/src/lib/task-model-selector.js

@@ -24,24 +24,24 @@ export const TaskType = {
  */
 const TASK_MODEL_MAP = {
   [TaskType.CHAT]: {
-    volcengine: "doubao-seed-1-6-flash-250828",
+    volcengine: "doubao-seed-1-6-251015",
     openai: "gpt-4o-mini",
     anthropic: "claude-sonnet-4-6",
     deepseek: "deepseek-chat",
     dashscope: "qwen-plus",
     gemini: "gemini-2.0-flash",
     mistral: "mistral-medium-latest",
-    ollama: "qwen2:7b",
+    ollama: "qwen2.5:7b",
   },
   [TaskType.CODE]: {
-    volcengine: "doubao-seed-code",
+    volcengine: "doubao-seed-1-6-251015",
     openai: "gpt-4o",
     anthropic: "claude-sonnet-4-6",
     deepseek: "deepseek-coder",
     dashscope: "qwen-max",
     gemini: "gemini-2.0-pro",
     mistral: "mistral-large-latest",
-    ollama: "codellama:7b",
+    ollama: "qwen2.5-coder:14b",
   },
   [TaskType.REASONING]: {
     volcengine: "doubao-seed-1-6-251015",
@@ -51,7 +51,7 @@ const TASK_MODEL_MAP = {
     dashscope: "qwen-max",
     gemini: "gemini-2.0-pro",
     mistral: "mistral-large-latest",
-    ollama: "qwen2:7b",
+    ollama: "qwen2.5:14b",
   },
   [TaskType.FAST]: {
     volcengine: "doubao-seed-1-6-lite-251015",