The Main Loop

At the heart of Systematic is a four-step loop adapted from battle-tested compound engineering principles:

This loop transforms your AI assistant from a simple code generator into a disciplined engineering collaborator. By following this cycle, you ensure that every change is researched, executed in isolation, rigorously reviewed, and documented for the future.

---

1. Plan

The plan phase turns an idea into a blueprint. Before a single line of code is written, the AI researches your codebase, gathers framework documentation, and identifies potential edge cases.

Skill: /ce:plan

What happens in Plan:

1. Requirement Analysis: The AI clarifies what you’re building and why.
2. Codebase Research: Using @repo-research-analyst, it identifies existing patterns and affected files.
3. External Research: Using @framework-docs-researcher, it consults live documentation for best practices.
4. Design Selection: It proposes an approach, listing files to change and potential risks.
5. Validation: You review the plan and tweak it until it’s ready for implementation.

---

2. Work

The work phase executes the approved plan. Instead of you typing, you monitor the AI as it implements the solution step-by-step in an isolated environment.

Skill: /ce:work

What happens in Work:

1. Isolation: The AI sets up a git worktree or branch to keep your main workspace clean.
2. Execution: It implements the plan incrementally, following the defined approach.
3. Verification: It runs tests, linters, and type checks after significant changes.
4. Handoff: Once complete, it creates a Pull Request (PR) for you to review.

---

3. Review

The review phase catches issues before they ship. Systematic launches multiple specialized agents in parallel to analyze your code from different perspectives.

Skill: /ce:review

What happens in Review:

• Multi-Agent Audit: Agents like @security-sentinel and @architecture-strategist scan for vulnerabilities and design flaws.
• Prioritization: Findings are categorized by severity (P1: Critical, P2: Important, P3: Minor).
• Resolution: The AI addresses findings automatically, fixing P1 issues first, then P2.
• Validation: Fixes are verified to ensure they address the feedback without introducing new regressions.

---

4. Compound

The compound phase turns today’s solutions into tomorrow’s advantages. By documenting solved problems, you build “institutional knowledge” that future AI agents can use.

Skill: /ce:compound

What happens in Compound:

1. Solution Capture: The AI asks what you learned during the implementation.
2. Categorization: It adds YAML frontmatter and tags for searchability.
3. Knowledge Base: The solution is added to your documentation (e.g., docs/solutions/).
4. Continuous Improvement: Next time a similar problem arises, agents will find this documentation and avoid repeating the same mistakes.

---

Quick Reference

Phase	Skill	Primary Agent/Skill
Plan	/ce:plan	@repo-research-analyst
Work	/ce:work	skill:git-worktree
Review	/ce:review	@architecture-strategist
Compound	/ce:compound	skill:compound-docs
Full Loop	/systematic:lfg	All of the above (Autonomous)

---

Next Steps

Now that you understand the loop, learn how Systematic integrates with your environment in the Architecture Guide.