Agent Skills for AI agents. A collection of development workflow skills born from real-world experience building software with AI coding agents.

These skills capture hard-won patterns — the kind of things you learn after watching AI agents produce PRs that look good but aren't merge-ready. Instead of repeating the same review feedback across sessions, encode it once as a skill and let every future session benefit.

AI agents are great at generating code, but PRs are rarely merge-ready on the first try. Pattern violations, inconsistent error handling, missing edge cases — these slip through because the agent doesn't review its own work against the codebase's actual conventions.

This skill builds an iterative review cycle into the development workflow. The agent reviews its own PR using multiple independent reviewers, posts findings to the PR, fixes issues, and re-reviews until the verdict is APPROVE. By the time you see the PR, it's already been through N rounds of automated review.

Issue → Research → Plan → Implement → PR → Review ⟲ Fix → APPROVE → Merge
                                            ↑________________↩ (repeat until clean)

6 phases: Issue Research → Plan + Worktree → Implementation → Commit + PR → Review Cycle → Merge + Cleanup

Supports multiple reviewer strategies (agent, codex exec, claude -p), works with GitHub and GitLab, and composes with TDD, BDD, and other methodology skills.

Installs to all supported agents (Codex, Claude Code, Cursor, Cline, Gemini CLI, Amp, and 30+ more):

npx skills add yvictor/skills --all

claude plugins marketplace add yvictor/skills
claude plugins install dev-workflow@yvictor-skills

git clone https://github.com/Yvictor/skills.git
cp -r skills/skills/<skill-name> ~/.your-agent/skills/