checkpoint-approval

What problem does it solve?

Critical transitions in AI workflows (e.g., approving a plan, merging changes) require explicit user consent. Bypassing these checkpoints can lead to unintended consequences. This Skill enforces mandatory user approval, ensuring human oversight and preventing costly errors.

Core Features & Use Cases

• Clear Presentation: Presents the plan or changes to the user in a clear, structured format, explicitly requesting approval.
• Indefinite Waiting: Crucially, the skill stops and waits indefinitely for explicit affirmative user input, preventing accidental or silent progression.
• Approval Flag Creation: Upon receiving explicit approval, it creates a persistent flag file (user-approved-synthesis.flag or user-approved-changes.flag) as an audit trail.
• State Transition: Transitions the task to the next appropriate state (e.g., IMPLEMENTATION or COMPLETE) only after approval is secured and the flag is created.
• Use Case: After a complex implementation plan has been synthesized, use this Skill to present the plan to the user for approval. It will wait until the user explicitly says "approved" or "proceed," ensuring the user has full control over the project's direction before any code is written.

Quick Start

After the SYNTHESIS phase completes, invoke the skill for plan approval: TASK_NAME="implement-formatter-api" /workspace/main/.claude/scripts/checkpoint-approval.sh
--task "$TASK_NAME"
--type "plan"
--plan-file "/workspace/tasks/$TASK_NAME/task.md"