Reusable AI agent skills for parallel development, code quality, and release automation.
15 production-ready skills following the open SKILL.md standard — compatible with Claude Code, Cursor, Copilot, Cline, Codex, and other agents.
| Skill | Description |
|---|---|
/code-review |
Review and score code across 12 quality dimensions (0-100). Read-only. |
/code-fix |
Apply prioritized fixes (P0 → P3). Single pass with verification. |
/coco-fix-iterate |
Autonomous loop: review → score → fix → verify → re-score until target. |
| Skill | Description |
|---|---|
/worktree-start |
Create a git worktree for isolated parallel feature development. Lighter than fork — shared git history, zero agent contention. |
/worktree-finish |
Merge a worktree branch into main and clean up the worktree. |
/worktree-list |
List active worktrees with branch, commits ahead of main, and dirty status. |
| Skill | Description |
|---|---|
/fork-project |
Create an isolated project copy (full clone) for parallel feature development. |
/merge-back |
Merge changes from a copy back to the original project. |
/cleanup-copy |
Delete a copy after merge-back is complete. |
/new-feature |
Start a new feature (fork + setup instructions). |
/finish-feature |
Complete a feature (merge + cleanup + optional release). |
| Skill | Description |
|---|---|
/preflight |
Run all validation checks without releasing. |
/hotfix |
Quick patch release for urgent fixes. |
/release-pr |
Create release from merge branch (bump, tag, push). |
/release |
Full release workflow (changelog, tests, PR, merge, tag). |
The star skill. An autonomous convergence loop that no one else offers as a standalone skill.
/coco-fix-iterate # Score >= 85, max 10 iterations
/coco-fix-iterate --score 90 # Higher quality bar
/coco-fix-iterate --max-iterations 5 security # Focus on security
How it works:
┌─────────┐ ┌───────┐ ┌──────┐ ┌────────┐ ┌────────┐
│ REVIEW │ ──→ │ SCORE │ ──→ │ PLAN │ ──→ │ FIX │ ──→ │ VERIFY │
│ 12 dims │ │ 0-100 │ │ P0→P2│ │ apply │ │ tests │
└─────────┘ └───┬───┘ └──────┘ └────────┘ └───┬────┘
↑ │ │
│ │ score >= target AND delta < 2? │
│ │ ──→ CONVERGED. STOP. │
│ │ │
└──────────────┴──────────── RE-SCORE ←───────────────────┘
Stops when: score meets target, scores oscillate, diminishing returns, or max iterations reached.
| Dimension | Weight | What it measures |
|---|---|---|
| Correctness | 15% | Tests pass, build succeeds, logic errors |
| Completeness | 10% | Requirements met, edge cases, TODO items |
| Robustness | 10% | Error handling, null checks, boundaries |
| Readability | 10% | Naming, function size, organization |
| Maintainability | 10% | Coupling, cohesion, SRP, DRY |
| Complexity | 8% | Cyclomatic complexity, nesting depth |
| Duplication | 7% | Copy-paste code, repeated patterns |
| Test Coverage | 10% | Line/branch coverage, critical paths |
| Test Quality | 5% | Meaningful assertions, edge cases |
| Security | 8% | OWASP top 10, injection, secrets |
| Documentation | 4% | Public API docs, README accuracy |
| Style | 3% | Lint compliance, formatting |
git clone https://github.com/corbat-tech/corbat-skills.git
bash corbat-skills/install.sh /path/to/your/projectbash install.sh /path/to/your/project code-review code-fix coco-fix-iterateThis copies skills into your project's .claude/skills/ directory.
Skills come in two flavors:
| Type | How it runs | Examples |
|---|---|---|
| Script-based | Has a bash script in scripts/ — the agent runs it |
fork-project, merge-back, cleanup-copy, release-pr |
| Agent-implemented | Agent reads SKILL.md and follows the steps | code-review, code-fix, coco-fix-iterate, preflight, hotfix, release |
Script-based skills can also be run manually from any terminal.
Works with any agent supporting the SKILL.md standard:
- Claude Code
- GitHub Copilot
- Cursor
- Cline
- OpenAI Codex
- Gemini CLI
- git — All skills use git for version control
- gh (optional) — GitHub CLI, needed for PR/release skills
- Node.js (optional) — For projects with
package.json(version bump, script detection) - Package manager — Skills auto-detect pnpm, yarn, or npm
MIT