executing-plans

The executing-plans skill provides a robust framework for coding agents to carry out structured development tasks derived from pre-approved implementation plans. Designed for complex software engineering projects, this skill ensures that agents move beyond simple code generation into disciplined, iterative execution. By isolating the development process into manageable, bite-sized tasks, the agent maintains high accuracy, prevents scope creep, and ensures every line of code is backed by verification. This skill is intended for developers and autonomous agents who require a systematic approach to feature implementation, refactoring, or bug fixing within a controlled environment.

• Executes tasks sequentially based on a predefined plan file, ensuring strict adherence to bite-sized instructions.

• Integrates mandatory verification steps after every task to ensure the codebase remains in a valid, functional state.

• Manages sub-skill delegation, including mandatory integration with finishing-a-development-branch upon task completion.

• Enforces a critical review phase where the agent analyzes the plan for potential blockers, missing dependencies, or unclear requirements before beginning implementation.

• Automatically utilizes git worktrees for environment isolation, preventing contamination of the main branch during the development cycle.

• Always review the plan critically before initiating execution; if concerns or ambiguity arise, consult the human partner immediately.

• If a blocker is encountered—such as a failing test, missing library, or logical impasse—the agent must cease execution and request human intervention rather than guessing.

• When running on platforms with subagent capabilities, developers are encouraged to prioritize subagent-driven-development for improved modularity and speed.

• Input requirements include a clear, written implementation plan. Output consists of verified code changes, unit test results, and a finalized development branch ready for merging or PR creation.

• Strictly adheres to the red-green-refactor TDD cycle where applicable, ensuring that testing remains a first-class citizen of the development workflow.