defragance

Defragance

Use this skill after comparable sources have been found, or invoke find-comparables first if the sources are missing. The goal is to reuse what open source makes reusable, while extracting a sense of taste and converting it into choices the target project can actually carry.

Workflow

1. Frame the target and learner intent:
   • What is being built: software project, event, product, workflow, or other artifact.
   • Current stack, target stack, domain, users, deployment or delivery model, and hard constraints.
   • What the user wants to learn or re-learn while building.
2. Select evidence:
   • Use 3-5 high-fit comparable repositories or artifacts.
   • Include official language/framework guidance and durable written references when relevant.
   • Record exact source paths, docs, or examples for every practice worth borrowing.
3. Distill the fragrance:
   • Naming, boundaries, entrypoints, file layout, and dependency direction.
   • Error handling, configuration, state, logging, testing, docs, release, and operations habits.
   • What the work feels like: compact, explicit, boring, sharp, playful, layered, or deliberate.
   • What the comparable refuses to do, not just what it does.
4. Translate instead of clone:
   • For each practice, mark it as adopt, adapt, defer, or avoid.
   • Explain why the target can carry it, or why the comparable's scale makes it a bad fit.
   • Copy code, names, or structures only when the license and attribution expectations allow it.
   • Bind every recommendation to a likely local file, module, ritual, or next slice.
5. Build the learning loop:
   • Start with one tracer-bullet slice that exercises the borrowed style end to end.
   • Turn unfamiliar idioms into small exercises, review prompts, or code-reading checkpoints.
   • Ask the agent to explain language-specific choices before implementing them when the user is learning.
6. Produce working rules:
   • A short project-specific style guide.
   • A first-slice implementation plan.
   • AI collaboration rules that preserve the borrowed taste during future coding sessions.

Output

Always return:

• Fragrance brief: 5-10 principles that capture the target style.
• Evidence map: source paths or references behind the principles.
• Adoption map: adopt, adapt, defer, and avoid decisions.
• First slice: the smallest buildable path that teaches and proves the style.
• AI rules: prompts or constraints to keep future implementation aligned.

Guardrails

• Do not praise famous projects without exact evidence.
• Do not import enterprise-scale patterns unless the target has enterprise-scale forces.
• Do not flatten the result into generic best practices.
• For Go, prefer simple packages, clear ownership, explicit errors, table-driven tests, narrow interfaces near consumers, and official Go guidance over framework fashion.