oh-my-claudecode v4.1.0: The Consolidation & Coordination Update
This major release introduces a fundamental overhaul of the agent architecture, streamlines skills and commands, and rolls out a powerful new Team Coordination system for distributed, resilient multi-agent workflows.
Breaking Changes
The previous tiered agent system (-low, -medium, -high suffixes) has been deprecated. Use unified agents with model parameter instead:
Task(subagent_type="oh-my-claudecode:architect", model="opus", ...)
Highlights
Agent Architecture Reform (#480, #481)
- Consolidated from 34 tiered agents to 28 unified, specialized agents
- New agents: debugger, verifier, style-reviewer, quality-reviewer, api-reviewer, performance-reviewer, dependency-expert, test-engineer, quality-strategist, product-manager, ux-researcher, information-architect, product-analyst
Advanced Team Coordination
- MCP Team Workers bridge daemon for distributed workflows
- Atomic task claiming with O_EXCL locks (replaces sleep+jitter)
- Security hardening: FD leak fix, path traversal prevention
- Resilience: hybrid orchestration, git worktrees, observability
System Prompt Rewrite for Claude Opus 4.6
- Rewritten following Anthropic's official prompting best practices
- XML behavioral tags, calm language, comprehensive tool/skill reference
Skill Consolidation (#470, #471)
- Merged: local-skills-setup -> skill, learn-about-omc -> help, ralplan/review -> plan
- Formalized commands vs skills architecture
Reliability Fixes
- Session state namespacing (#456)
- Compaction debounce (#453)
- Timeout-protected stdin (#459)
- MCP 429/rate-limit fallback (#469)
- Cross-session state leakage fix
- Ralplan native Plan Mode approval gate (#448, #463)
See CHANGELOG.md for full details.