Nanocoder 1.23.0: Interactive Prompts, Scheduled Tasks, and Enhanced Stability

[will-lamerton]

We're very happy to announce Nanocoder 1.23.0 - a major release bringing powerful new features for automation, interactivity, and reliability. Whether you're running scheduled tasks or building more sophisticated AI workflows, this release delivers significant improvements.

Interactive User Prompts

The new ask_user tool enables your AI agent to pause and ask questions during a conversation. The model can now present selectable options and wait for your response before continuing. This transforms Nanocoder into a true interactive partner - perfect for decision-making workflows where the AI needs human input to proceed intelligently.

Agent: Should I create a new branch for this feature?
Options:
  1. Yes, create "feature/new-ui"
  2. Yes, create "bugfix/improve-performance"
  3. No, work on main

Your choice: 1

Scheduled Task Automation

Running the same tasks repeatedly? The new per-project cron scheduler is built for you. Store markdown prompts with YAML frontmatter in .nanocoder/schedules/, then manage them with the /schedule command:

/schedule create "Daily code review"
/schedule add "0 9 * * MON" "Review pull requests and test coverage"
/schedule start

Tasks run on a schedule with full run history, sequential job execution, and intelligent deduplication. Perfect for automated testing, report generation, or daily health checks.

Graceful Shutdown and Reliability

Behind the scenes, we've added a centralized ShutdownManager that coordinates cleanup of all services—VS Code integration, MCP servers, LSP managers—preventing orphaned processes and dangling connections. Combined with our V2 type-safe tool system overhaul and three-tiered defense against chaotic LLM outputs, Nanocoder now delivers significantly more stable behavior, especially with local LLMs.

Additional Improvements

• New file operations: delete_file, move_file, create_directory, copy_file tools for comprehensive file management
• Enhanced text input: Readline-style keybinds (Ctrl+W, Ctrl+U, Ctrl+K, Ctrl+A/E, Ctrl+B/F) for smoother CLI editing
• Context awareness: /context-max command and env variable for manual context length control on unlisted models
• Better searches: search_file_contents now supports include and path parameters for scoped queries
• Visual enhancements: Kanagawa theme, persistent context percentage display in mode indicator
• Unified commands: Refactored skills system into cleaner custom commands with optional metadata
• Robust parsing: XML parser optimizations and comprehensive JSON tool call parsing with self-correction

What's Under the Hood

This release includes 40+ changes improving stability, security, and developer experience. We've strengthened error handling, optimized bash output streaming for better responsiveness, fixed local Ollama server detection, and updated core dependencies for security and compatibility.