dx-swe-executing-plans

Load the plan file, review it critically and raise concerns before starting, then execute each task in order while running the verifications the plan specifies. Track progress with a todo list and stop to ask for help when blocked rather than guessing.

Read the entire plan and review it critically for gaps, unclear instructions, or missing dependencies. Raise any concerns with your human partner first, and only proceed once the plan is sound. Never start implementation on the main branch without explicit consent.

Stop immediately when you hit a blocker such as a missing dependency, a failing test, an unclear instruction, or a verification that fails repeatedly. Ask for clarification instead of guessing, and revisit the plan if the fundamental approach needs rethinking.

Yes, it is designed for sessions without subagents, but the documentation notes that platforms with subagent support produce higher-quality results. If subagents are available, the dx-swe-subagent-dev skill is recommended instead.

After every task is completed and verified, the workflow hands off to the finishing-a-development-branch skill. That step verifies tests, presents options, and completes the development branch.