69 skills · 19 hooks · 4 contracts · 5 missions
The universal agent enhancement layer for AI coding CLIs.
Works with: Copilot CLI · Claude Code · OpenAI Codex · Gemini CLI · Cursor · Windsurf · OpenCode
Adds superpowers to your AI coding agents - regardless of which CLI you use. Combines the best patterns from oh-my-codex, oh-my-claudecode, oh-my-openagent, clawhip, and more into a single unified system.
Without oh-my-universal: Your AI agent writes code when asked. No planning, no review, no memory. Every session starts from scratch.
With oh-my-universal: Your AI agent plans before implementing, reviews its own code, remembers decisions across sessions, delegates to sub-agents, auto-fixes build failures, guards against dirty worktrees, produces handoffs between sessions, reasons deeply through complex decisions, preserves state through compaction, and self-improves from failures. Across ANY CLI tool you use.
- Auto-planning before implementation (plan -> critique -> execute)
- Code review on every significant change
- Memory/context persistence across sessions
- Multi-agent delegation for complex tasks
- Self-maintaining docs - agents update documentation as they work
- Autonomous pipelines - idea to validated code with zero intervention
- Screenshot-based UI review - visual feedback on UI changes
- Lifecycle hooks - intercept agent events to guard and extend behavior
- Discord notifications when long tasks complete
- Cross-project skills - use these skills from any project directory
# Clone the repo
git clone https://github.com/Tushar49/oh-my-universal.gitThen point your CLI at it:
# Copilot CLI
copilot
/add-dir /path/to/oh-my-universal
# Claude Code
claude --plugin-dir /path/to/oh-my-universal
# Cursor
# Copy .cursor/rules/skills.mdc to your project's .cursor/rules/
# Windsurf
# Copy .windsurfrules to your project root
# Gemini CLI / OpenAI Codex / OpenCode
# Reads AGENTS.md automatically - see docs/SETUP.mdThat's it. No runtime, no dependencies - just plain markdown files your agent reads as instructions.
After /add-dir, skills appear in Copilot CLI as:
/skills- 7 grouped skill categories (plan-and-build, investigate, quality, docs-and-memory, operations, meta, collaborate)- Individual prompts - top 15 skills (plan, review, ultrawork, trace, tdd, etc.) available directly
See docs/SETUP.md for detailed per-CLI setup.
| CLI | How it connects | Entry file | Status |
|---|---|---|---|
| Copilot CLI | /add-dir or copy .github/instructions/ |
.github/copilot-instructions.md |
✅ Ready |
| Claude Code | --plugin-dir or symlink .claude/skills/ |
CLAUDE.md |
✅ Ready |
| OpenAI Codex | AGENTS.md routes to skills/ | AGENTS.md |
✅ Ready |
| Gemini CLI | GEMINI.md / AGENTS.md routes to skills/ | GEMINI.md |
✅ Ready |
| OpenCode | Plugin or AGENTS.md | AGENTS.md |
✅ Ready |
| Cursor | Rules file | .cursor/rules/skills.mdc |
✅ Ready |
| Windsurf | Rules file | .windsurfrules |
✅ Ready |
skills/ → 69 reusable workflow patterns (plan, review, trace, etc.)
hooks/ → 19 lifecycle event handlers (auto-run at key moments)
contracts/ → 4 behavioral invariants (always enforced)
missions/ → 5 scoped task packages (specific goals with success criteria)
Skills are the core - markdown files that any AI CLI reads as instructions. Hooks make the agent proactive (auto-save memories, guard worktrees, verify deliverables). Contracts are rules that can never be violated (state precedence, terminal handoffs). Missions are copy-paste task specs for common automation goals.
oh-my-universal/
├── .github/
│ ├── copilot-instructions.md # Universal agent rules
│ ├── agents/ # Custom agents (planner, reviewer, etc.)
│ └── prompts/ # Reusable workflow prompts
├── skills/ # 69 cross-project skills
├── hooks/ # 19 lifecycle hooks
├── contracts/ # 4 behavior contracts
├── missions/ # 5 scoped task missions
├── config/ # Configuration
├── docs/ # Setup guides, architecture
├── AGENTS.md # Codex/Gemini entry
├── GEMINI.md # Gemini CLI entry
├── CLAUDE.md # Claude Code entry
└── README.md
| Skill | Description |
|---|---|
| plan | Structured planning with self-critique before implementation |
| ultrawork | Full lifecycle: plan -> implement -> verify -> review -> commit |
| autopilot | Fully autonomous 5-phase pipeline: idea to validated working code |
| verify | Evidence-based change verification (tests + manual checks) |
| build-fix | Auto-fix build failures (3 attempts max, then escalate) |
| tdd | Test-driven development: Red -> Green -> Refactor |
| ralph | Persistent verify-and-fix loop until complete (max 10 iterations) |
| pipeline | User-defined sequential stage runner with rollback |
| ecomode | Token-efficient execution mode |
| Skill | Description |
|---|---|
| deep-dive | Two-stage investigation: causal tracing + Socratic questioning |
| trace | Evidence-driven debugging with structured hypothesis testing |
| ask | Query multiple AI models and cross-validate answers |
| architecture | Map and understand codebase structure |
| analyst | Requirements gap analysis - find missing specs before coding |
| autoresearch | Systematic multi-source research with cross-validation |
| external-context | Load external docs/URLs into agent context |
| debug | Hands-on step-through debugging workflow |
| deep-interview | Socratic intent-clarification loop with ambiguity scoring |
| scientist | Data analysis and hypothesis-driven investigation with statistical rigor |
| ultrathink | Deep reasoning for complex decisions with structured deliberation |
| Skill | Description |
|---|---|
| review | High-signal code review (bugs, security, logic only) |
| multi-model-review | 4-perspective review: bug hunter, security, maintainer, perf |
| security-review | Repo-wide security audit (threats, auth, deps, CVEs) |
| visual-verdict | Screenshot-based UI review with structured visual feedback |
| perf-audit | Performance profiling: identify bottlenecks, measure, optimize |
| pre-commit-check | Quality gate: verify + review + security before commit |
| designer | UI/UX specs from requirements - wireframes, components, tokens |
| ai-slop-cleaner | Strip AI cliches and fluff from output |
| ultraqa | Comprehensive QA with iterative fix cycles |
| deliverables | Validate all expected outputs exist and meet acceptance criteria |
| api-design | REST/GraphQL API design — endpoints, versioning, errors, pagination |
| critic | Structured multi-perspective review of plans, specs, and analysis |
| Skill | Description |
|---|---|
| doc-maintainer | Auto-update documentation after code changes |
| wiki | Auto-maintained project knowledge base (reduces token usage) |
| remember | Memory persistence across sessions (native tools -> .memory/ fallback) |
| writer-memory | Enhanced persistent context for writing projects (tone, style, narrative) |
| Skill | Description |
|---|---|
| release | Release management: changelog, version bump, git tag, publish |
| hooks | Pre/post hooks for agent operations - intercept lifecycle events |
| dirty-guard | Block dangerous git operations with uncommitted changes |
| run-tagging | Unique identity per agent session for audit trails |
| parity-check | Agent health and contract smoke tests |
| handoff | Structured state handoff between agent runs |
| hud | Heads-up display - compact status showing agent progress |
| cancel | Kill-switch to safely abort running operations |
| workflow-state | State machine for agent workflow transitions |
| container-sandbox | Run untrusted code in isolated container sandbox |
| worktree-sandbox | Git worktree parallel isolation for safe experimentation |
| git-master | Git workflow management (branches, PRs, conflicts) |
| cost-tracker | Track and report token/API costs per session |
| dependency-upgrade | Safe dependency upgrades with compatibility checks and rollback |
| compact-guard | State preservation through context compaction |
| permission-tuner | Denial pattern analysis and permission optimization |
| Skill | Description |
|---|---|
| self-improve | Analyze agent failures, improve skills to prevent recurrence |
| skillify | Create new skills from observed repeatable patterns |
| doctor | Project health check (fast + full modes) |
| mcp-setup | Auto-detect project type and configure MCP servers |
| session-manager | Manage, resume, and fork agent sessions with persistent state |
| session-protocol | End-of-session: docs + memory + verify + commit prep |
| mission-runner | Execute scoped task missions from missions/ directory |
| command-gen | Scaffold reusable commands and PRPs |
| config-sync | Sync agent rules across CLI frameworks |
| deepinit | Deep project initialization and scaffolding |
| rules-discovery | Auto-discover and surface project conventions and rules |
| web-clone | URL-driven website cloning with visual + functional verification |
| output-styles | Configurable output modes (verbose, compact, structured, etc.) |
| Skill | Description |
|---|---|
| team | Multi-agent delegation with serial fallback |
| repo-merge | Merge features from external repos (research -> matrix -> port) |
| notify | Notifications: console, system toast, Discord webhook |
| refactor | Detect code smells, apply improvements preserving behavior |
19 lifecycle hooks in hooks/ covering the complete agent lifecycle:
| Category | Hooks |
|---|---|
| Session | session-start, session-end |
| Tool | pre-tool, post-tool, tool-failure |
| Skill | skill-start, skill-end |
| Keyword | keyword-detector |
| Memory | memory-save, memory-load |
| Safety | permission-check, dirty-guard-hook, context-guard |
| Quality | verify-deliverables, code-simplifier-hook |
| Subagent | subagent-start, subagent-end |
| Context | pre-compact, context-inject |
See hooks/README.md for detailed documentation.
4 formal behavior contracts in contracts/ that all skills and hooks MUST respect:
| Contract | What it enforces |
|---|---|
state-precedence |
Session > project > global state; rollback on failure; forbidden transitions |
terminal-handoff |
Every skill must end in a terminal state (finished/blocked/failed/cancelled/needs-input) |
team-mutation |
Claim-and-version protocol for parallel sub-agent work |
quality-gate |
Build + test + security + docs gates before commit |
See contracts/README.md for details.
Missions are scoped task specifications - self-contained work packages with clear success criteria. While skills define HOW to work, missions define WHAT to do.
See missions/README.md for full documentation. See docs/SANDBOX_SPEC.md for the sandbox evaluation system.
| Mission | What it does |
|---|---|
security-hardening |
Audit and fix security vulnerabilities |
test-coverage-boost |
Increase test coverage by 15%+ |
performance-baseline |
Establish perf baselines, fix bottlenecks |
docs-from-scratch |
Generate comprehensive project docs |
dependency-cleanup |
Remove unused deps, patch CVEs |
Create your own: copy missions/_template/ and customize.
- Copilot CLI:
.github/instructions/skills.instructions.md - Claude Code:
.claude/skills/oh-my-universal/SKILL.md - Codex/OpenCode:
AGENTS.md - Gemini CLI:
GEMINI.md - Cursor:
.cursor/rules/skills.mdc - Windsurf:
.windsurfrules - Setup guide:
docs/SETUP.md
Built by combining the best patterns from:
| Source | Stars | What we took |
|---|---|---|
| oh-my-codex | - | plan, ultrawork, team, build-fix, tdd, dirty-guard, run-tagging, parity-check, handoff, state model, contracts |
| oh-my-claudecode | - | remember, team, session hooks, skill routing, autopilot, deep-dive, ask, trace, visual-verdict, wiki, writer-memory, release, self-improve, hud, session-manager, mcp-setup, hooks |
| oh-my-openagent | - | architecture (Atlas), review (Momus), doc-maintainer (Librarian) |
| clawhip | - | notify (Discord), memory offload |
| planning-with-files | 19K | Persistent markdown planning approach |
| caveman | 38K | Compact token-light skill design |
| claude-forge | 659 | Security hooks, pre-commit gates, refactor patterns |
| everything-claude-code | 160K | Skill taxonomy, multi-perspective review |
| awesome-claude-code | - | status-line patterns (claude-pace, claude-powerline), worktree isolation |
| career-ops | - | repo-merge workflow (real-world tested) |
Q: Do I need to install anything? A: No. Just point your CLI at this directory. Skills are plain markdown - no runtime, no dependencies.
Q: Does this work with my existing project?
A: Yes. Use /add-dir (Copilot) or --plugin-dir (Claude) to add skills to any project.
Q: Can I use only some skills? A: Yes. Skills are independent. Use what you need, ignore the rest.
Q: How is this different from CLAUDE.md instructions? A: CLAUDE.md is one file for one tool. oh-my-universal is 69 skills + 19 hooks + 4 contracts that work across 7 different AI CLIs simultaneously.
- Fork the repo
- Create a skill:
cp missions/_template/ missions/my-feature/orcp skills/plan.md skills/my-skill.md - Follow the skill format (see any existing skill)
- Update the adapter files: run
config-syncskill or manually add to all CLI adapters - Submit a PR
See CONTRIBUTING.md for detailed guidelines.
If this saves you time, consider giving it a ⭐