@defai.digital/ax-cli 3.8.7 → 3.8.9

package/README.md

@@ -20,12 +20,11 @@
 
 - [🚀 Quick Start](#-quick-start)
 - [✨ Features](#-features)
-- [🎉 What's New](#-whats-new-in-v384)
+- [🎉 What's New](#-whats-new-in-v389)
 - [📦 Installation](#-installation)
 - [⚙️ Configuration](#️-configuration)
 - [🔒 Security](#-security--api-key-handling)
 - [🎯 Usage Examples](#-usage-examples)
-- [🌐 Web Search](#-web-search)
 - [🔌 MCP Integration](#-mcp-model-context-protocol)
 - [🧠 Project Memory](#-project-memory)
 - [🎯 Multi-Phase Planner](#-multi-phase-task-planner-v300)
@@ -100,19 +99,6 @@ ax-cli
   - **Verbose mode**: Full details for debugging
   - Press `Ctrl+O` to cycle between levels
   - Auto-expands errors with full details
-- **💬 Dual-Model Mode**: Use different models for chat vs coding
-  - Configure chat and coding models separately
-  - Manual model switching with `--chat-mode` flag
-  - Optimize cost and performance for different task types
-- **🌐 Web Search**: Real-time package search capabilities
-  - **Works out-of-the-box**: npm, PyPI, and crates.io package search (no API keys required)
-  - **Intelligent routing**: Automatically selects the best engine based on query intent
-    - JavaScript/Node.js packages → npm registry search
-    - Python packages → PyPI registry search
-    - Rust packages → crates.io registry search
-  - Results caching for faster responses (5 minute TTL)
-  - Support for package discovery, dependency research, and version information
-  - Session-based context for iterative package exploration
 - **🔄 Auto-Update**: Built-in update checker and installer
 - **🔒 Enterprise-Grade Security**: **FREE & Open Source**
   - **Command Injection Protection**: CVSS 9.8 CRITICAL fix - Safe command execution with whitelisting
@@ -164,6 +150,55 @@ AX CLI uses **industry-standard max tokens** based on research of leading AI cod
 
 [View all features →](docs/features.md)
 
+## 🎉 What's New in v3.8.9
+
+**Refactoring & Configuration Improvements** - Further centralization and code quality enhancements:
+
+### 🔧 Improvements
+
+- **Centralized file name constants** - Added `FILE_NAMES` constant object to eliminate hard-coded file names across the codebase
+- **Extended CONFIG_PATHS** - Added new path constants for CUSTOM_MD, INDEX_JSON, MEMORY_JSON, and user directories
+- **Unified path handling** - Replaced 15+ hard-coded `.ax-cli` path references with centralized constants
+- **Improved text editor tool** - Enhanced functionality and reliability
+- **Subagent types refinement** - Streamlined specialized agent type definitions
+- **Prompt improvements** - Enhanced default prompts configuration
+
+### ✅ Quality
+
+- TypeScript strict mode passes
+- Build succeeds
+- 76/77 test files pass
+- Zero breaking changes
+
+---
+
+## 🎉 What's New in v3.8.8
+
+**Code Quality & Maintainability Release** - Major refactoring for improved code consistency:
+
+### 🔧 Improvements
+
+- **Centralized configuration constants** - Added `CONFIG_DIR_NAME` constant to eliminate hard-coded `.ax-cli` strings across 15+ files
+- **Unified error handling** - Replaced inline `error instanceof Error ? error.message : String(error)` patterns with `extractErrorMessage()` utility across 20+ files
+- **Improved ES module consistency** - Converted inline `require()` calls to proper ES module imports
+- **Better home directory handling** - Replaced `process.env.HOME || process.env.USERPROFILE` patterns with Node.js `homedir()` function
+- **Removed examples folder** - Cleaned up unused example files
+
+### ✅ Quality
+
+- All tests passing
+- 98%+ test coverage maintained
+- Zero breaking changes
+- Improved code maintainability
+
+---
+
+## 🎉 What's New in v3.8.7
+
+**Critical Bug Fixes & Resilience Improvements** - Stability and reliability enhancements.
+
+---
+
 ## 🎉 What's New in v3.8.6
 
 **Patch Release** - Bug fixes and stability improvements:
@@ -738,374 +773,6 @@ The `doctor` command checks:
 - ✓ MCP server configuration
 - ✓ Dependencies (ripgrep, git)
 
-## 💬 Dual-Model Mode
-
-Use different models for chat vs coding tasks to optimize performance and cost:
-
-### Configuration
-
-**Option 1: Project Settings** (recommended for project-specific preferences)
-
-Add to `.ax-cli/settings.json` in your project directory:
-
-```json
-{
-  "dualModel": {
-    "enabled": true,
-    "chatModel": "grok-code-fast-1",
-    "codingModel": "glm-4.6"
-  }
-}
-```
-
-**Option 2: Environment Variables** (for temporary or CI/CD use)
-
-### Usage
-
-```bash
-# Use faster chat model for questions
-ax-cli --chat-mode -p "explain what this project does"
-
-# Use coding model (default) for implementation
-ax-cli -p "implement user authentication"
-
-# In interactive mode, default is coding model
-ax-cli
-```
-
-### Environment Variables
-
-```bash
-# Enable dual-model mode
-export AI_DUAL_MODEL_ENABLED=true
-export AI_CHAT_MODEL=grok-code-fast-1
-export AI_CODING_MODEL=glm-4.6
-
-ax-cli --chat-mode
-```
-
-**Benefits:**
-- 💰 **Cost savings**: Use faster/cheaper models for simple queries
-- ⚡ **Better performance**: Match model capability to task complexity
-- 🎯 **Manual control**: You decide when to use each model
-
-## 🌐 Web Search
-
-Search package registries for JavaScript, Python, and Rust packages with intelligent language detection and cross-registry comparison.
-
-### ✨ Works Out of the Box!
-
-**Package search is enabled by default** (no API key required):
-- **npm** - JavaScript/Node.js packages from npmjs.com
-- **PyPI** - Python packages from pypi.org
-- **crates.io** - Rust packages from crates.io
-
-### Quick Setup
-
-**No setup required!** Package search works immediately:
-- **npm** - JavaScript/Node.js packages work immediately
-- **PyPI** - Python packages work immediately
-- **crates.io** - Rust packages work immediately
-- Perfect for package discovery and dependency management
-- No API keys needed
-
-### Usage
-
-```bash
-# The AI will automatically use web search when needed
-ax-cli
-
-> "Find a React state management library"     # Uses npm search
-> "Search for axios npm package"              # Uses npm search
-> "Find a Python data analysis library"       # Uses PyPI search
-> "Search for tokio rust crate"               # Uses crates.io search
-```
-
-### How It Works
-
-- **Intelligent Routing**: Automatically selects the best search engine based on query type and language detection
-  - **JavaScript/Node.js packages** → npm registry search (always available, no API key)
-  - **Python packages** → PyPI registry search (always available, no API key)
-  - **Rust packages** → crates.io registry search (always available, no API key)
-  - **General/technical queries** → package search fallback
-
-- **Language Detection**: Automatically detects programming language from keywords
-  - Python keywords (pip, django, flask, pandas) → PyPI
-  - Rust keywords (cargo, crate, tokio, serde) → crates.io
-  - npm/package keywords → npm registry
-  - Multiple engines may be used in parallel for best results
-
-- **Automatic Caching**: Results cached for 5 minutes for faster subsequent queries
-
-- **LLM Integration**: The AI automatically uses package search for:
-  - Package discovery and dependency management
-  - Version compatibility checks
-  - Alternative package recommendations
-  - Package documentation and usage information
-
-### Features
-
-- **Multi-Registry Package Search**: Search across npm, PyPI, and crates.io simultaneously
-  - Package metadata, descriptions, and download statistics
-  - Version information and release dates
-  - No API keys or setup required
-- **Source Attribution**: All results include URLs and sources
-- **Parallel Search**: Multiple engines searched concurrently for comprehensive results
-
-### Manual Usage
-
-While the AI uses web search automatically, you can also request it explicitly:
-
-```bash
-# In interactive mode
-> "search npm for a markdown parser library"
-> "search PyPI for a web scraping package"
-> "search crates.io for async runtime"
-
-# Headless mode
-ax-cli -p "search for react-query npm package"
-ax-cli -p "find a Python FastAPI alternative"
-```
-
-### Web Search Session
-
-AX CLI maintains intelligent search context across your conversation, enabling natural follow-up questions and iterative refinement:
-
-**Session Continuity:**
-- Package search results are preserved in conversation context
-- Ask follow-up questions about packages without re-searching
-- Reference previous search results naturally
-- Session context includes package URLs, metadata, versions, and download stats
-
-**Example Session:**
-
-```bash
-ax-cli
-
-# Initial package search
-> "search npm for a state management library"
-🔍 Searching npm registry...
-Found 5 packages:
-
-1. **zustand** (2.5M weekly downloads)
-   Small, fast and scalable state-management
-   Latest: v4.4.7 | Size: 1.2KB gzipped
-
-2. **redux** (8.1M weekly downloads)
-   Predictable state container for JavaScript apps
-   Latest: v5.0.0 | Size: 6.2KB (core only)
-
-3. **mobx** (1.2M weekly downloads)
-   Simple, scalable state management
-   Latest: v6.12.0 | Size: 16KB
-   ...
-
-# Natural follow-up (uses cached context from npm search)
-> "which one has the smallest bundle size?"
-Based on the npm search results:
-- ✅ zustand: 1.2KB (gzipped) - Smallest
-- jotai: 2.9KB (gzipped)
-- redux: 6.2KB (core only)
-- mobx: 16KB
-
-# Version and compatibility check
-> "what's the latest version of zustand and does it support React 18?"
-Package: zustand v4.4.7 (latest)
-✅ Full React 18 support with concurrent features
-✅ TypeScript 5.0+ support
-📅 Last published: 2 weeks ago
-
-# Installation guide
-> "show me how to install and use zustand"
-Installation:
-npm install zustand
-
-Basic usage:
-[Provides code example from npm documentation]
-```
-
-**Context-Aware Features:**
-
-1. **Result Caching**: Package search results stay in memory for the session
-   - 5-minute cache for identical queries
-   - Instant responses for follow-up questions about packages
-   - No repeated API calls to registries
-
-2. **Multi-Turn Package Refinement**:
-   ```bash
-   > "search npm for a react table library"
-   Found: tanstack-table, react-table, ag-grid-react, mui-x-data-grid
-
-   > "which ones have TypeScript support?"
-   All 4 packages support TypeScript:
-   - @tanstack/react-table: Full TS rewrite
-   - react-table (deprecated, use @tanstack)
-   - ag-grid-react: TypeScript included
-   - @mui/x-data-grid: Full TS support
-
-   > "which has the best documentation?"
-   Based on npm stats and GitHub stars:
-   - @tanstack/react-table: Excellent docs, 24K stars
-
-   > "install that one"
-   npm install @tanstack/react-table
-   ```
-
-3. **Cross-Registry Context**:
-   ```bash
-   > "search for data validation libraries"
-   Searching npm, PyPI, and crates.io...
-
-   npm: zod, yup, joi, ajv
-   PyPI: pydantic, marshmallow, cerberus
-   crates.io: serde, validator
-
-   > "compare the JavaScript and Python options"
-   **JavaScript (npm):**
-   - zod: 3.5M/week, TypeScript-first, 30KB
-   - yup: 5.2M/week, Schema builder, 45KB
-
-   **Python (PyPI):**
-   - pydantic: 50M/month, Type hints, fast
-   - marshmallow: 8M/month, Schema validation
-
-   > "which is fastest?"
-   - JavaScript: zod (TypeScript inference, zero-cost)
-   - Python: pydantic (uses Rust core, 20x faster than marshmallow)
-   ```
-
-4. **Package Comparison Tables**:
-   ```bash
-   > "search npm for http client libraries"
-   Found: axios, node-fetch, got, ky, superagent
-
-   > "create a comparison table"
-
-   | Package     | Weekly DLs | Size    | Last Update | Browser | Node |
-   |-------------|------------|---------|-------------|---------|------|
-   | axios       | 48M        | 11.5KB  | 2 weeks ago | ✅      | ✅   |
-   | node-fetch  | 35M        | 4.5KB   | 3 months    | ❌      | ✅   |
-   | got         | 23M        | 15KB    | 1 week ago  | ❌      | ✅   |
-   | ky          | 1.2M       | 12KB    | 2 weeks ago | ✅      | ✅   |
-
-   > "which is best for Node.js backend with retry logic?"
-   Recommendation: **got**
-   - Built-in retry with exponential backoff
-   - HTTP/2 support
-   - Request cancellation
-   - Promise & stream support
-   ```
-
-**Session Management:**
-
-- **Session Duration**: Active for entire interactive session
-- **History Integration**: Search results included in `--continue` sessions
-- **Memory Commands**:
-  ```bash
-  /clear    # Clears search context and conversation
-  /exit     # Ends session (context lost)
-  ```
-- **Persistent Context**: Use with `--continue` to maintain search context across sessions
-
-**Best Practices:**
-
-1. **Start Broad, Refine Iteratively**:
-   ```bash
-   > "search npm for testing libraries"
-   Found: jest, vitest, mocha, jasmine, playwright, cypress
-
-   > "focus on those for integration testing"
-   Integration testing: playwright, cypress, vitest (has browser mode)
-
-   > "which has TypeScript support?"
-   All 3 have TypeScript:
-   - playwright: Native TS
-   - cypress: Full TS support
-   - vitest: Native TS (Vite-powered)
-
-   > "show setup for playwright"
-   npm install -D @playwright/test
-   [Provides example config and test]
-   ```
-
-2. **Leverage Context for Framework Comparisons**:
-   ```bash
-   > "search npm for react vue svelte packages"
-   Found core packages with download stats:
-   - react: 22M/week
-   - vue: 5.1M/week
-   - svelte: 850K/week
-
-   > "compare their package ecosystems"
-   **React:** 180K+ packages
-   **Vue:** 45K+ packages
-   **Svelte:** 8K+ packages
-
-   > "which has better TypeScript support?"
-   All have excellent TS support:
-   - React: @types/react (20M/week)
-   - Vue: Built-in TS (Vue 3+)
-   - Svelte: svelte-check + TypeScript plugin
-   ```
-
-3. **Version Compatibility Checks**:
-   ```bash
-   > "search npm for next auth package"
-   Found: next-auth (8M/week, v4.24.5)
-
-   > "does it work with Next.js 15?"
-   ⚠️  Compatibility:
-   - next-auth v4: Next.js 12-14
-   - For Next.js 15: Use NextAuth.js v5 (beta)
-
-   > "show me the v5 package"
-   Package: next-auth@beta (v5.0.0-beta.4)
-   ✅ Next.js 15 compatible
-   [Installation and migration guide]
-   ```
-
-4. **Combine Search with Development Tasks**:
-   ```bash
-   > "search npm for a markdown parser library"
-   Found: marked, remark, markdown-it, showdown
-
-   > "which is fastest and most secure?"
-   Recommendation: **marked**
-   - 13M/week downloads
-   - Fast (built-in sanitization)
-   - Active maintenance
-
-   > "install marked and show me basic usage"
-   Installing: npm install marked
-   [Generates code example with marked usage]
-
-   > "add it to my project"
-   [Creates/updates relevant files with implementation]
-   ```
-
-**Performance Tips:**
-
-- **First search**: 1-3 seconds (registry API call)
-- **Follow-up questions**: Instant (uses cached package data)
-- **Cache duration**: 5 minutes per query
-- **Parallel searches**: Multiple registries searched concurrently for cross-language queries
-- **Offline work**: Use `--continue` to preserve search context across sessions
-
-### Troubleshooting
-
-**No results found**
-- Package registries (npm, PyPI, crates.io) are always available
-- Try refining your search query
-- Check your internet connection
-
-**Rate limit errors**
-- Package registry searches are rate-limited by the registry providers
-- Results are cached to minimize API calls
-
-**Slow searches**
-- Results are cached after first search (5 minute TTL)
-- Subsequent identical queries will be instant
-
 ## 🔌 MCP (Model Context Protocol)
 
 Extend AX CLI with MCP servers for additional capabilities:
@@ -1304,15 +971,6 @@ AX CLI implements enterprise-grade architecture with:
 ### v3.6.1 (2025-11-22)
 
 **🔧 Improvements:**
-- **Web Search Simplification**: Removed Tavily AI dependency, focusing entirely on package registries
-  - Streamlined to npm, PyPI, and crates.io package search only
-  - No API keys required for web search functionality
-  - Reduced dependencies and simplified architecture
-- **Documentation Overhaul**: Completely updated web search documentation
-  - 200+ lines updated with package-focused examples
-  - 15+ new realistic examples showing npm, PyPI, and crates.io workflows
-  - Comprehensive session examples for package discovery and comparison
-  - Best practices for cross-registry searches and version compatibility checks
 - **Smart Paste Auto-Collapse**: Intelligent handling of large text inputs
   - Automatic collapse of 20+ line pastes for better readability
   - Press Ctrl+P to expand/collapse pasted content
@@ -1434,7 +1092,6 @@ AX CLI implements enterprise-grade architecture with:
 - Multi-phase task planner with automatic complexity detection
 - Enhanced MCP integration with production-ready templates
 - Project memory system with intelligent context caching
-- Web search capabilities with npm, PyPI, and crates.io package registries
 - Advanced code analysis tools (dependency, security, metrics)
 
 ## 📄 License

package/config-defaults/prompts.yaml

@@ -2,17 +2,30 @@
 # AI assistant instructions and guidelines for AX CLI
 
 system_prompt:
-  identity: "You are AX CLI Assistant, an enterprise-grade AI coding agent specialized in software development, file operations, and intelligent automation. You excel at understanding context, making efficient tool choices, and executing tasks with precision."
+  # Professional, humble identity - no marketing jargon
+  identity: "You are AX CLI, an AI coding assistant that runs in your terminal. You help developers with software engineering tasks using the tools available to you. Be direct, accurate, and efficient."
+
+  # Professional objectivity - prioritize truth over user validation
+  professional_objectivity:
+    title: "PROFESSIONAL OBJECTIVITY"
+    content: |
+      Prioritize technical accuracy and truthfulness over validating the user's beliefs:
+      • Focus on facts and problem-solving, not emotional validation
+      • No unnecessary superlatives, praise, or agreement
+      • If the user is wrong, respectfully correct them with evidence
+      • Don't say "You're absolutely right" or "Great question!" - just answer
+      • Investigate claims before confirming them
+      • Respectful correction is more valuable than false agreement
 
   core_principles:
     title: "CORE OPERATING PRINCIPLES"
     guidelines:
-      - "Speed & Efficiency: Minimize tool calls, maximize output per interaction"
+      - "Technical Accuracy: Be correct, not agreeable. Correct mistakes respectfully."
+      - "Brevity: Keep responses under 4 lines unless detail is explicitly requested."
       - "Context Awareness: Understand project structure before acting"
       - "Precision: Make exact, targeted changes - never broad strokes"
       - "Self-Correction: Validate your work and fix issues immediately"
-      - "Minimal Verbosity: Show results, not commentary"
-      - "Smart Tool Usage: Answer simple questions directly. Only use tools when they provide necessary context for accurate responses."
+      - "Smart Tool Usage: Answer simple questions directly. Only use tools when necessary."
 
   tools_header: "Available Tools:"
 
@@ -23,6 +36,8 @@ system_prompt:
       description: "Create NEW files only. Never use on existing files."
     - name: str_replace_editor
       description: "Edit existing files with precise string replacement. Primary editing tool."
+    - name: multi_edit
+      description: "Make multiple edits to a single file atomically. Use instead of multiple str_replace_editor calls."
     - name: bash
       description: "Execute shell commands for git, npm, testing, searching, and system operations"
     - name: search
@@ -63,8 +78,26 @@ system_prompt:
         - "NEVER create_file on existing files - instant data loss"
         - "ALWAYS view_file before str_replace_editor to ensure context"
         - "Use search when you don't know exact file locations"
-        - "Use bash for git operations, package management, testing"
         - "Create todo lists for tasks with 3+ distinct steps"
+        - "CONSOLIDATE EDITS: When making multiple changes to the same file, identify ALL changes first, then make ONE str_replace_editor call with a larger old_str block that captures the entire region"
+        - "Anti-pattern: Multiple str_replace_editor calls to same file in sequence - always consolidate into single call"
+        - "For function rewrites: Include entire function in old_str and replace completely rather than making line-by-line edits"
+
+    tool_preferences:
+      title: "TOOL PREFERENCE HIERARCHY"
+      content: |
+        PREFER dedicated tools over bash equivalents:
+        • search tool > bash grep/rg/find
+        • view_file > bash cat/head/tail
+        • str_replace_editor > bash sed/awk
+        • create_file > bash echo/cat with redirection
+
+        USE bash ONLY for:
+        • git operations (status, diff, commit, push, pull)
+        • Package management (npm, pip, cargo)
+        • Running tests (npm test, pytest, cargo test)
+        • Build commands (npm run build, make)
+        • System commands that have no tool equivalent
 
     search_and_discovery:
       title: "EFFICIENT CODEBASE EXPLORATION"
@@ -110,13 +143,15 @@ system_prompt:
     response_optimization:
       title: "RESPONSE STYLE"
       guidelines:
-        - "NO pleasantries: Skip 'Great!', 'Thanks!', 'Sure thing!'"
-        - "Show, don't tell: Display results, not descriptions"
-        - "Explain complex decisions only when necessary"
+        - "BREVITY: Keep responses under 4 lines unless user explicitly asks for detail"
+        - "One-word answers are preferred when appropriate"
+        - "NO preamble: Never start with 'Sure!', 'Great!', 'I'll help you with that', 'Let me...'"
+        - "NO summaries: Don't summarize what you just did unless asked"
+        - "Show results, not descriptions of what you're about to do"
         - "Errors: state problem → show fix → done"
-        - "Success: confirm completion → show key changes"
-        - "Simple questions: Answer directly without unnecessary tool calls"
-        - "If you find yourself calling the same tool multiple times with no progress, STOP and provide a direct answer"
+        - "Success: confirm completion briefly"
+        - "Simple questions: Answer directly without tool calls"
+        - "If calling same tool repeatedly with no progress, STOP and answer directly"
 
     confirmation_behavior:
       title: "USER CONFIRMATION FLOW"
@@ -140,6 +175,8 @@ system_prompt:
         - "Cache understanding: Don't re-read files unnecessarily"
         - "Use regex in search for pattern matching"
         - "Leverage bash for bulk file operations"
+        - "Edit consolidation: Multiple edits to same file? Include entire function/block in old_str and replace in one call"
+        - "Extensive rewrite (>50% of file changing): Consider create_file to overwrite rather than many str_replace_editor calls"
 
     error_handling:
       title: "ERROR HANDLING & VALIDATION"
@@ -150,7 +187,54 @@ system_prompt:
         - "Auto-fix obvious errors (typos, syntax) immediately"
         - "Report non-obvious errors with context"
 
-  closing: "Operate with precision, speed, and intelligence. You are a professional coding agent - act like one."
+    code_conventions:
+      title: "CODE STYLE MIMICRY"
+      content: |
+        IMPORTANT: Before writing any code, examine the existing codebase:
+        1. Indentation style (tabs vs spaces, 2 vs 4 spaces)
+        2. Quote style (single vs double quotes)
+        3. Semicolon usage
+        4. Naming conventions (camelCase, snake_case, PascalCase)
+        5. Import organization patterns
+        6. Comment style and density
+
+        Rules:
+        • Match existing conventions EXACTLY - never introduce new styles
+        • Don't add libraries/dependencies without explicit approval
+        • NEVER add comments unless logic is non-obvious or user requests them
+        • Don't add type annotations to code that doesn't have them
+        • Don't "improve" code style unless explicitly asked
+
+    git_safety:
+      title: "GIT SAFETY PROTOCOL"
+      content: |
+        CRITICAL GIT RULES:
+        • NEVER commit unless user explicitly requests ("commit this", "make a commit")
+        • NEVER push unless user explicitly requests
+        • NEVER amend commits unless explicitly asked
+        • NEVER force push
+        • NEVER run git commands with -i flag (interactive mode not supported)
+
+        Before committing:
+        1. Run git status to see changes
+        2. Run git diff to review what will be committed
+        3. Show user what will be committed
+        4. Use meaningful commit messages (not "update files")
+
+        If user hasn't asked to commit, just make the changes - they will commit when ready.
+
+    post_change_validation:
+      title: "POST-CHANGE VALIDATION"
+      content: |
+        After making code changes, validate your work:
+        1. If project has lint command (npm run lint, ruff, etc.) - run it
+        2. If project has typecheck (npm run typecheck, mypy, tsc) - run it
+        3. If you modified code with existing tests - run relevant tests
+        4. If validation fails, fix the issues before reporting completion
+
+        Don't skip validation to save time - catching errors early is faster.
+
+  closing: "Be direct. Be accurate. Be brief."
 
 custom_instructions_prefix: "\n\n=== CUSTOM PROJECT INSTRUCTIONS ===\n"
 custom_instructions_suffix: "\n=== END CUSTOM INSTRUCTIONS ===\n\nFollow custom instructions while adhering to core principles above."

package/config-defaults/settings.yaml

@@ -6,6 +6,7 @@ agent:
   default_timeout: 360000  # 6 minutes in milliseconds
   default_max_tokens: 8192
   max_recent_tool_calls: 20  # Keep only last 20 unique tool calls
+  max_messages: 500  # Maximum number of messages to keep in chat history
 
   # Loop Detection (ENABLED BY DEFAULT)
   # Detects when the AI repeats the same tool call multiple times
@@ -90,6 +91,8 @@ token:
 cache:
   default_max_size: 1000
   default_ttl: 300000  # 5 minutes
+  tool_args_cache_max_size: 500  # Max entries for tool call arguments cache
+  tool_args_cache_prune_count: 100  # Number of entries to remove when cache is full
 
 performance:
   debounce_delay: 300  # milliseconds

package/dist/agent/llm-agent.d.ts

@@ -57,7 +57,6 @@ export declare class LLMAgent extends EventEmitter {
     private bashOutput;
     private todoTool;
     private search;
-    private webSearch;
     private _architectureTool?;
     private _validationTool?;
     private chatHistory;
@@ -172,13 +171,23 @@ export declare class LLMAgent extends EventEmitter {
     private get validationTool();
     /**
      * Detect if a tool call is repetitive (likely causing a loop)
-     * Returns true if the same tool with similar arguments was called multiple times recently
+     * Uses the intelligent LoopDetector which provides:
+     * - Tool-specific thresholds (file ops get higher limits)
+     * - Progress-based detection (tracks success/failure)
+     * - Cycle pattern detection (A→B→A→B loops)
      */
     private isRepetitiveToolCall;
+    /** Last loop detection result for error messages */
+    private lastLoopResult?;
     /**
      * Reset the tool call tracking (called at start of new user message)
      */
     private resetToolCallTracking;
+    /**
+     * Generate a helpful warning message when a loop is detected
+     * Uses the lastLoopResult for context-aware suggestions
+     */
+    private getLoopWarningMessage;
     /**
      * Check if a request should trigger multi-phase planning
      */