seed4j-worktree-flow - SKILL.md Agent Skill

name: seed4j-worktree-flow description: Manage the local seed4j-cli Git worktree workflow safely. Use when Codex is asked to audit seed4j-cli worktrees, create a feature worktree, clean up a completed worktree, decide which branches are already merged into main, or keep /home/renanfranca/projects/seed4j-cli as the main worktree while feature work happens under /home/renanfranca/projects/seed4j-cli-worktree.

Use this skill only for /home/renanfranca/projects/seed4j-cli.

Treat /home/renanfranca/projects/seed4j-cli as the permanent main worktree.
Treat /home/renanfranca/projects/seed4j-cli-worktree/<branch> as the location for feature worktrees.
Use main as the base branch for new feature worktrees.
Never delete remote branches automatically.
Never remove a dirty worktree.
Ask before running destructive cleanup when the user has not explicitly requested it.

Prefer scripts/seed4j_worktree_flow.sh instead of hand-writing Git worktree command sequences.

Run it from the skill directory:

scripts/seed4j_worktree_flow.sh status
scripts/seed4j_worktree_flow.sh start <branch>
scripts/seed4j_worktree_flow.sh finish <branch>

Start with status to inspect the current worktree layout, dirty states, and branch relationship to main.
For a new task, use start <branch>.
- If the branch exists locally, the script creates a worktree for it.
- If the branch does not exist locally, the script creates it from main.
For completed work, use finish <branch> only after the branch is merged into main.
If finish fails, report the exact guardrail that blocked cleanup and do not bypass it with --force.

Keep the main repository on main whenever possible.
If a branch is already checked out in another worktree, do not create a second checkout for it.
If a target directory already exists, inspect it before taking action.
If main is stale and the task depends on remote state, fetch or pull only when the user asks or when the task explicitly requires current remote information.
If validation is requested before merge, prefer focused checks first, then ask the user to run ./mvnw clean verify unless they explicitly ask Codex to run it.