context-engineering

Managing context loss in software engineering requires a phased workflow that captures facts and decisions into structured documentation before coding. This approach breaks work into Research, Design, Plan, and Implement cycles to preserve traceability across teams.

The best way to structure complex refactoring is a four-phase cycle: Research gathers facts, Design defines architecture, Plan outlines work, and Implement executes. This enforces independent phase deliverables and peer reviews for predictable delivery.

To start the Research phase for a new feature, provide a task from your tasks.json file and trigger the workflow. The phase will guide your team to collect facts and generate structured research notes before moving to design.

Yes, you can use this phased workflow for bug fixes, new features, and large-scale refactors. It applies a disciplined context management process to ensure quality checks and structured documentation across various software engineering tasks.

The documentation artifacts generated during the phases include research notes, design decisions, and implementation plans. These structured deliverables ensure traceability and enforce quality gates before progressing to the next phase.

Yes, you need mandatory frontmatter in SKILL.md to enforce guardrails for the workflow. This requirement ensures independent phase deliverables, peer reviews, and quality gates are properly documented and maintained throughout the process.