Overview

Executing Plans loads a pre-written implementation plan, reviews it critically, and executes tasks in batches with checkpoints for architect review between batches. It supports resuming from a saved state via .tasks.json and uses worktree isolation to protect the main branch.

How This Skill Works

Load persisted tasks to resume, verify or set up the worktree, then execute tasks in batches (default 3). After each batch, report results and wait for human feedback; when all tasks finish, complete the development branch.

When to Use It

• You have a written implementation plan on disk.
• You are executing in a separate or parallel session.
• You want human checkpoints between task batches.
• You need to resume a plan after an interruption (via .tasks.json).
• You require workspace isolation using git worktrees to protect the main branch.

Quick Start

1. Step 1: Load persisted tasks and verify the work environment.
2. Step 2: Execute tasks in batches (default 3) and report after each batch.
3. Step 3: Finish the branch after all tasks complete and await human review.

Best Practices

• Load and validate the plan before starting the run.
• Use a batch size of 3 by default, but adjust for task complexity.
• After each batch, generate a report and solicit architect feedback.
• Save progress frequently to .tasks.json to enable clean resume.
• Run with isolated worktrees to keep changes contained and reversible.

Example Use Cases

• Implementing a feature from a written plan stored on disk, executed in 3-task batches with architect review after each batch.
• Running a data migration plan in a separate session, resuming later via .tasks.json if interrupted.
• Refactoring work performed in a parallel session, with batch reviews before proceeding.
• Finishing the development branch after all plan tasks are completed, with final human approval.
• Onboarding a new developer by executing a pre-approved plan in isolated worktrees.

Frequently Asked Questions