checkpoint_skill

Overview

This skill provides meta-cognitive decision support that pauses execution, evaluates the current work context, and surfaces 2–5 clear next-step options to the user. It recommends one option with a short rationale to reduce risk, avoid rework, and keep momentum aligned with the user's goals.

How this skill works

When invoked, the skill silently assesses progress, quality, alignment, uncertainty, risk, and efficiency. It generates a curated set of distinct options (including at least one continue and one verify choice), selects a recommended action a senior engineer would take, and presents the choices via an AskUserQuestion-style prompt with short labels and descriptions.

When to use it

• After a meaningful milestone, feature, or fix completes and a decision is needed
• When requirements are ambiguous or multiple valid approaches exist
• If task scope or complexity increases beyond initial expectations
• When confidence is low or critical assumptions remain unverified
• Before finalizing, committing, or shipping to catch omissions early

Best practices

• Keep the assessment quick and focused—this should take seconds, not minutes
• Provide 2–5 sharply different options, not many subtle variants
• Always include at least one forward-motion option and one pause/verify option
• Make a clear recommendation and explain the risk-reduction rationale
• Trigger sparingly—reserve for real decision points to avoid interrupting flow

Example use cases

• A feature build finishes but used a few shortcuts; present options like refactor, add tests, or commit
• Multiple pending tasks compete for priority; propose a re-prioritize or continue-now choice
• Unclear requirements discovered mid-implementation; offer to clarify, run a systems audit, or pause
• Scope expanded with cross-module changes; suggest a systems audit, document changes, or proceed carefully
• Before final commit: offer test/verify, quick code review, or commit-and-iterate