oh-my-claudecode v4.0.0
Release Date: 2026-02-05
Breaking Changes
- Node.js 20 Required - Minimum Node.js version increased from 18 to 20. Users on Node.js 18 must upgrade before using this version. This aligns with Node.js LTS schedule (Node 18 EOL: April 2025).
MCP Breaking Changes
- MCP Response Contract (#424) -
ask_codexandask_gemininow return file paths instead of inline content.output_fileparameter is now required. Addedworking_directoryparameter for path resolution. - Prompt Parameter Deprecated (#421) - The
promptparameter forask_codexandask_geminiis removed. Useprompt_fileinstead, which writes prompts to files to avoid OS argument length limits. - Agent Role Required -
agent_roleparameter is now required for bothask_codexandask_geminitools
Refactored
- MCP Core Modules - Extracted shared Codex/Gemini business logic into
codex-core.tsandgemini-core.ts, eliminating code duplication between in-process SDK servers and standalone stdio servers - Slim Server Wrappers - Reduced
codex-server.ts,codex-standalone-server.ts,gemini-server.ts, andgemini-standalone-server.tsto thin transport wrappers (~40-70 lines each) - Build Scripts - Updated esbuild scripts with global npm module resolution banner for proper native module loading
Changed
- Codex/Gemini Agent Routing - Codex (
ask_codex) now recommended only for reviewer and planning agents (architect, planner, critic, analyst, code-reviewer, security-reviewer, tdd-guide). Gemini (ask_gemini) recommended only for design agents (designer, writer, vision). - CLAUDE.md Compacted - Reduced from 679 to 389 lines (43% smaller). All 33 agents, all skills, and all MCP tools with descriptions now inline.
- Agent Tiers Reference - Added External AI Consultation section with strict Codex/Gemini routing per agent domain
- Model Routing Enforcement (#384) - Advisory agents enforce read-only behavior; execution agents get full tool access
- Gemini Model Fallback Chain - gemini-3-pro-preview → gemini-3-flash-preview → gemini-2.5-pro → gemini-2.5-flash
Performance
- Levenshtein LRU Cache - Added LRU cache for Levenshtein distance calculations in skill matching
- Skill Metadata Cache - Added 30-second TTL cache for skill metadata
- Debounced State Writes - Subagent tracker now batches state file writes with 100ms debounce
- Pre-compiled Regex Patterns - Pre-compiled 15+ regex patterns in hot paths
- Parallel State File Reads - Pre-compact module now reads all state files in parallel
- Session Index Caching - Token tracker caches session indices with 5-minute TTL
- Space-Optimized Levenshtein - Algorithm now uses O(n) space instead of O(n²)
Added
- Job Management Tools (#420) -
wait_for_job,check_job_status,kill_job,list_jobsfor background Codex/Gemini job control - Version Drift Detection (#422) - Automatic detection and cleanup of stale plugin versions
- HUD Fine-Grained Control (#399) - Configurable sessionHealth display elements
- MCP Prompt File Parameters (#416) - New
prompt_fileandoutput_fileparameters - MCP Prompt Persistence Audit Trail - Prompts persisted to disk for debugging and replay
- Plugin-Scoped Codex/Gemini MCP Servers - Proper MCP server discovery
- MCP System Prompt Injection - Better agent context injection
- Built-in MCP Tools - State, notepad, and project memory tools
- Configurable Stop Hook Callbacks (#395)
- Swarm Aggressive Mode - Wave-based spawning and micro-task decomposition
- Codex/Gemini Timeout Increase - Increased to 1 hour for complex analysis
Fixed
- Per-Directory Debounce - Fixed race condition in subagent tracker
- TOCTOU Race Condition - Fixed time-of-check-time-of-use race in pre-compact
- True LRU Cache - Fixed Levenshtein cache to use true LRU eviction
- Session Index Off-by-One - Fixed token tracker session index calculation
- Cache Hit Rate Formula (#425) - Corrected calculation (was exceeding 100%)
- HUD Token Formatting (#417) - Fixed returning "0.999k" for values under 1000
- LSP goto_definition Null Safety (#417) - Fixed crash on
LocationLinkobjects - MCP Output File Suffix (#419) - Fixed writing to
.rawsuffix - MCP Stdin Prompt Piping - Fixed OS argument length limit issues
- HUD Semver Sorting (#373) - Fixed version comparison
- Security Hardening - Cross-platform compatibility and security hardening
- Codex/Gemini Role Allowlists - Fixed to match documented routing
Security Hardening
- Model parameter validation (prevents command injection)
- TOCTOU-safe output file writes (symlink protection)
- Cross-platform path validation (Windows support)
- PID validation and spawn registry
- workingDirectory trust boundary enforcement
- Non-blocking stdin in hooks
Install/Upgrade
npm install -g oh-my-claude-sisyphus@4.0.0Full Changelog
See CHANGELOG.md