Summary

The Execute skill is a high-level workflow guide that directs agents through implementing approved tasks to completion. It provides a canonical execution framework with emphasis on verification, incremental changes, and clear deliverables. The skill outlines when to work in parallel vs. serially, prioritizes deletion over addition, and enforces quality gates like test passage and evidence collection before marking work complete.

Detected Capabilities

Trigger Keywords

Phrases that agents use to match this skill to user intent.

Use Cases

• Build small-to-medium features end-to-end with verification
• Coordinate multi-unit feature work across parallel executor agents
• Implement bug fixes or refactorings while keeping diffs minimal and reversible
• Manage long-running or unbounded task lists with checkpoints and completion criteria
• Transition from planning/design to working implementation with quality gates

Quality Notes

• Strengths: Clear, actionable workflow with explicit rules and scale guidance. Emphasis on verification, small reversible changes, and completion gates is sound engineering practice. The skill correctly separates concerns (authoring vs. approval, feature vs. refactoring) and discourages anti-patterns (workarounds, self-approval, placeholder TODOs).
• Strengths: Scoped as a meta-skill that routes to other agents (executor, team, review) — clarifies when to delegate vs. when to act directly. Mentions durable task lists for long-running work, showing awareness of real-world constraints.
• Weaknesses: No concrete examples (e.g., how to verify? what does 'evidence' look like?). The skill would benefit from 1-2 worked examples showing a small task broken into units and executed in parallel, plus example verification outputs.
• Weaknesses: No guidance on error recovery or rollback. What happens if a unit fails mid-task? Should the agent halt other parallel units, or continue and report later? This is addressed only implicitly via 'report what changed and what remains.'
• Weaknesses: References other skills (executor, team, review, verify) but provides no criteria for choosing among them. An agent might not know when to use 'team' vs. 'executor', or what 'review' expects as input.
• Weaknesses: No handling of dependency discovery. The skill says 'break into independent units' but provides no method for detecting or surfacing hidden dependencies. A naive agent might parallelize work that should be sequential.