agentic-workflow_skill

Overview

This skill packages a practical, battle-tested workflow for working with multiple AI agents in day-to-day development. It provides command cheats, keyboard shortcuts, session aliases, Git integration patterns, MCP (multi-agent) usage tips, and container-safe experimentation guidelines. The focus is on predictable outcomes: faster iteration, reliable session recovery, and smooth CI/CD interactions.

How this skill works

The skill describes agent-specific commands (Claude, Gemini, Codex), recommended keyboard shortcuts, and shell aliases to speed common tasks. It maps multi-agent orchestration patterns—plan, analyze, implement, execute—and shows how to route work to MCP tools like Playwright, Supabase, and Codex-CLI. Concrete Git and container workflows enable isolated worktrees, automated commit/PR generation, and safe sandboxing for risky experiments.

When to use it

• Optimizing daily AI-assisted development and code review loops
• Integrating AI agents into Git workflows, PR creation, and branch isolation
• Coordinating large-codebase analysis or high-token tasks via Gemini or MCP
• Managing long-running or recoverable agent sessions across terminals
• Running experiments in disposable containers to avoid local pollution

Best practices

• Use aliases (c, cc, cr, g, cx) and --continue/--resume flags to resume sessions quickly
• Limit active MCP servers and tools; disable unused ones to reduce noise and cost
• Work in git worktrees for isolated feature/hotfix sessions with independent AI contexts
• Prefer TDD pattern: add failing test, implement minimal fix, run tests, commit iteratively
• Keep a Quick Reference Card for essential commands (/clear, /context, /usage, /init)

Example use cases

• API design: Claude drafts spec → Gemini analyzes repo patterns → Claude writes code → Codex runs tests and builds
• PR review: checkout PR, ask agent to summarize changes, flag security/performance spots, propose patches
• Large-scale analysis: ask-gemini to scan entire src/ for architectural smells or duplication
• Emergency hotfix: create a git worktree, resume a dedicated agent session, run tests and deploy from container
• Automated commits/PRs: request AI to write commit messages and create draft PRs with change summaries