jeo_skill

Overview

This skill orchestrates end-to-end AI agent workflows: plan, execute, verify, and clean up. It integrates planning (ralph + plannotator), execution (team/ bmad), browser verification (agent-browser), and automatic worktree cleanup. Multiple agent platforms are supported including Claude, Codex, Gemini, and OpenCode.

How this skill works

You start by generating a plan with ralph and perform a visual review using plannotator in a blocking foreground session for user approval. Execution runs via omc team mode when available or falls back to BMAD workflows. Browser-based features are verified with agent-browser snapshots and element checks. After tasks finish, a worktree cleanup script prunes stale git worktrees and stores state in .omc/state/jeo-state.json.

When to use it

• When you need a repeatable, reviewable plan-before-execute workflow for AI-driven changes.
• When parallel multi-agent execution (omc team) can speed up implementation.
• When tasks include UI or browser interactions that require automated verification.
• When you want automatic cleanup of git worktrees and persisted execution state.
• When running across different agent platforms: Claude, Codex, Gemini, or OpenCode.

Best practices

• Always run plannotator in foreground (no &), require explicit Approve before executing.
• Prefer omc team mode on Claude Code for staged, parallel pipelines; use BMAD as fallback.
• Keep plan documents (plan.md) and approved plans under .omc/plans for traceability.
• Run worktree-cleanup.sh immediately after completion to avoid branch/worktree clutter.
• Add blocking plannotator instructions to platform hooks (Gemini/Gemini.md) to ensure feedback is applied in-turn.

Example use cases

• Implementing a feature with an AI-written implementation and validating UI changes via browser snapshots.
• Coordinating multi-agent code refactors using omc team pipeline to parallelize tasks.
• Running a CI-like agent workflow on a platform without team support by using BMAD fallback.
• Reviewing and approving complex change plans interactively with plannotator before any code runs.
• Automating git worktree pruning after multiple agent-created branches to keep the repository clean.