Download Latest Version v1.2.75 source code.tar.gz (455.4 kB)
Email in envelope

Get an email when there's a new version of The Pope Bot

Home / v1.2.72
Name Modified Size InfoDownloads / Week
Parent folder
README.md 2026-03-02 6.0 kB
v1.2.72 source code.tar.gz 2026-03-02 277.0 kB
v1.2.72 source code.zip 2026-03-02 394.8 kB
Totals: 3 Items   677.8 kB 1

Changelog: thepopebot v1.2.72

🚀 Major New Features

Claude Code Integration

  • Added support for Claude Code CLI as an alternative agent backend alongside Pi
  • Claude Pro/Max subscription support - Use OAuth tokens instead of API credits for agent jobs
  • Live coding workspaces - Launch Claude Code in Docker containers with WebSocket terminal access
  • Code mode in chat - Select GitHub repos/branches and launch workspaces directly from conversations
  • New database tables - Added claude_workspaces table with container tracking
  • Enhanced UI - Code chats show terminal icon and link to live workspace

Unified Skills System

  • Consolidated skills architecture - Single skills/ directory shared by Pi and Claude Code
  • Active skill management - Symlink activation via skills/active/ (replaces .pi/skills/)
  • Enhanced skill discovery - New get_skill_building_guide and get_skill_details tools
  • Skill inventory - Shows both active and available-but-inactive skills with descriptions
  • Updated skill format - Uses project-root-relative paths (e.g., skills/brave-search/search.js)

🔧 Core Improvements

CLI Enhancements

  • New upgrade command - One-step upgrade with git sync, build, push, and Docker restart
  • Piped input support - CLI commands can read values from stdin (e.g., echo "val" | thepopebot set-var KEY)
  • Better error handling - Upgrade command detects conflicts and provides recovery instructions
  • Version targeting - Support for @beta, @rc, and specific version upgrades

Setup Wizard Improvements

  • Separate chat vs agent LLM configuration - Choose different models for interactive chat vs job execution
  • Claude subscription flow - Guided OAuth token setup for Claude Pro/Max users
  • Simplified installation - Removed automatic gh CLI installation, clearer prerequisites
  • Enhanced Telegram setup - Better step-by-step guidance with visual progress indicators

Job System Upgrades

  • Job titles - LLM-generated descriptive titles for jobs instead of UUIDs
  • Enhanced job config - Single job.config.json with title, description, and metadata
  • Better commit messages - Human-readable "🤖 Agent Job: [title]" format
  • Job summary improvements - More structured output with steps, successes, and struggles

🎨 UI/UX Enhancements

Web Interface

  • Pull request dashboard - New /pull-requests page showing open PRs with auto-refresh
  • Chat title generation - Structured output for better, more descriptive chat titles
  • Code mode toggle - Repo/branch picker in chat interface for launching workspaces
  • Enhanced sidebar - Pull request count badge, better navigation
  • Workspace linking - Code chats show terminal icon and link to live workspace

Terminal Experience

  • WebSocket proxy - Secure WebSocket tunneling to Claude Code containers
  • Enhanced terminal - Xterm.js with proper theming and reconnection support
  • Docker socket access - Event handler can launch containers via Docker API

🛠️ Technical Improvements

Architecture Changes

  • Renamed config files - EVENT_HANDLER.mdJOB_PLANNING.md, AGENT.mdJOB_AGENT.md
  • New code planning system - Separate CODE_PLANNING.md for workspace conversations
  • Dual agent system - Separate getAgent() (job planning) and getCodeAgent() (workspace) instances
  • WebSocket hack - Workaround for Next.js WebSocket handler interference (see docs/HACKS.md)

Security & Stability

  • Better secret handling - Fixed eval injection in Docker entrypoint with proper jq @sh escaping
  • Enhanced authentication - Auto-set AUTH_URL from APP_URL for proper redirects
  • Docker security - Run containers as non-root users where possible
  • Input validation - Structured output for chat titles to prevent thinking-token leaks

Documentation

  • New guides - CLAUDE_CODE_VS_PI.md, HOW_TO_BUILD_SKILLS.md, RUNNING_DIFFERENT_MODELS.md
  • Updated references - Consolidated skill documentation, removed outdated Pi-specific guides
  • Hacks documentation - Documented Next.js WebSocket workaround for future maintenance

📦 Infrastructure Updates

Docker Images

  • New container types - job-pi-coding-agent, job-claude-code, and claude-workspace images
  • Updated base images - Node.js 22, Ubuntu 24.04, latest dependencies
  • Better Chrome handling - Proper Puppeteer installation and Chrome sandboxing

GitHub Actions

  • Workflow naming - "Agent Job" instead of "Run thepopebot Job"
  • Agent backend selection - AGENT_BACKEND variable controls Pi vs Claude Code
  • Enhanced release notes - Better beta/stable differentiation in publish workflow

Database

  • New migrations - Added claude_workspaces table with container tracking
  • Schema updates - Added claude_workspace_id to chats table for workspace linking
  • Auto-migration - Drizzle migrations apply automatically on server startup

🐛 Bug Fixes & Minor Changes

  • Fixed secret setting - CLI now properly handles multi-line secrets via stdin
  • Better error messages - Job summarization shows more context on failures
  • Template file handling - .template suffix convention for gitignore/CLAUDE.md files
  • Local development - Enhanced bin/local.sh for faster development cycles
  • Dependency updates - Latest versions of LangChain, Drizzle, and other core packages
  • Removed deprecated files - Cleaned up old Pi-specific documentation and unused skills

Technical Context: This release represents a major architectural shift from Pi-only to multi-backend agent execution. The skills system was completely redesigned to be provider-agnostic, and the chat interface now supports both conversational planning and live coding workspaces. The upgrade process is now fully automated with conflict detection and recovery.

Source: README.md, updated 2026-03-02