Context Engine Starter

Overview

AI tools like Claude and Gemini are capable of substantially accelerating design work — feedback, briefs, decisions, synthesis, copy review. But most designers were using them in a way that left most of that capability untouched. Every session started from scratch. The designer would re-explain the design system, re-describe their role, re-establish project constraints. The AI gave generic output because it had generic context.

The missing piece wasn't a better prompt. It was a context layer: a structured library of markdown files that a designer builds once and feeds to the AI before asking anything. A designer who gives Claude their who-i-am.md, a relevant playbook, and a project brief gets output calibrated to how they actually work. A designer who pastes a raw question gets output calibrated to the average designer.

The context-engine-starter is that layer, packaged for any designer to set up in 30 minutes.

What I Built

The personal workspace — a templatized yourname-context-engine/ directory:

• who-i-am.md — the gateway file. Role, strengths, working style, values, constraints, links to playbooks. The more specific it is, the sharper the AI output.
• playbooks/ — recurring workflows documented in a standard template. Playbooks are the highest-leverage context files: they teach the AI not just who you are but how you actually work.
• A synthesis hierarchy: session-summaries/ → daily-summaries/ → weekly-summaries/ → monthly-summaries/. Each level rolls up the one below it, so a monthly summary is a one-read strategic picture of the past 30 days of AI-assisted work.
• activity-log/ — running log of AI interactions and time saved.

The Spring Health reference engine — a production-grade shared context directory designers inherit rather than build. It contains the full Verdant design system documentation (component mapping, layout tokens, navigation specs, usage patterns, 10 foundational files on color, typography, shape, and accessibility), seven UX writing guidelines, and brand standards. When everyone feeds the same reference engine to their AI tools, the AI's design system knowledge is consistent across the team — because it comes from the same files.

The incremental profile ritual. Filling out a profile template in one sitting is daunting. The kit solves this with a daily question: say "ask me today's question" in any session inside the folder, and Clea (the chief-of-staff agent defined at the project root) reads who-i-am.md, finds the most useful gap, and asks one focused question — usually multiple choice so the answer takes seconds. Past questions are tracked so you never get the same one twice. Within two weeks, you have a rich, accurate profile earned through real reflection.

Five slash commands for session continuity:

• /add-session-summary — writes a structured summary at the end of each session to feed at the start of the next
• /add-activity-log — synthesizes what was asked, what was produced, estimates time saved
• /add-daily-summary → /add-weekly-summary → /add-monthly-summary — the rollup hierarchy

AI models don't remember between conversations. The synthesis hierarchy is the behavioral fix — and /add-session-summary is the habit that makes it work.

The Design Principles Behind It

Three decisions shaped the kit:

Ship a complete reference example, not just a template. A blank template asks designers to imagine what "good" looks like. The Spring Health context engine is good — designers can see what depth and specificity look like in practice, then build their personal engine to the same standard.

Make the first 30 minutes viable, not aspirational. Two files are enough to see a real difference: a filled-in who-i-am.md and one playbook. The guide explicitly says "rough context is better than no context" and "start small." The temptation to ship a comprehensive system that nobody completes is real; the discipline is shipping one that actually gets used.

Model-agnostic by design. The kit works with Claude today, Gemini tomorrow, and travels with the designer to their next job. When the team's primary model changes, the context files don't. The investment in building a context engine doesn't depreciate.

Outcome

Any designer on the team can now start an AI session with full context — who they are, how they work, the Verdant design system, Spring Health's writing voice — already loaded. The Spring Health reference engine doubles as living documentation of the design system and writing standards, useful as a reference even outside AI sessions.

The portable idea: a well-maintained context file outperforms a well-crafted prompt. Build the library once; stop re-explaining yourself.