Glossary

A document containing your unique insights, domain knowledge, and contrarian takes that AI doesn't have. Your competitive edge in any project.

A prompt structure: [what exists] + [what you want] + [constraints]. The most reliable pattern for effective prompts.

The limited amount of text AI can "see" at once. Includes your prompt, conversation history, and any referenced files. When full, AI starts forgetting earlier context.

A prompting technique where you explicitly state what AI should NOT do: no new dependencies, no abstractions, no changes outside scope.

When a long conversation pollutes AI's context window with failed attempts, causing it to contradict itself or repeat mistakes. Fix: start a fresh chat.

Stacking multiple security layers (validation, auth, authorization, encryption, logging) so that if one fails, others still protect the system.

Your execution roadmap. Tech stack decision, architecture overview, phased tasks, and risk mitigation. Written before coding starts.

A massive function that does everything. An antipattern AI frequently produces. Fix: ask for single-responsibility functions that each do one thing.

The landscape of your system that AI can't see: existing patterns, constraints, integrations, history, and future plans. You provide this; AI goes vertical from there.

Building features in small, reviewable steps instead of one big prompt. Each step is small enough to review, test, and catch mistakes before they compound.

When something breaks, go back to the original prompt with the error log and "what to avoid" instead of trying to patch the broken output. Fresh context beats polluted context.

A document specifying goals, non-goals, user stories, technical requirements, and success metrics. The contract between you and your AI assistant.

Context + Task + Constraints + Output Format. The four-part structure that every effective prompt follows.

Pointing AI at existing code as an example instead of describing patterns in words. "Follow the pattern in UserService.ts" beats "use a service pattern with dependency injection."

Three forces that solve any problem: research docs to understand, add logs to see what's happening, write tests to prove the fix. Each feeds the others in a feedback loop.

Create the file structure first, then fill in individual files with directed prompts. Prevents AI from making assumptions about architecture.

If you've tried to fix something 3 times and it's still broken, stop fixing forward. Restart with a fresh oneshot prompt containing everything you've learned.

For security-sensitive code: Pass 1 checks if it works correctly. Pass 2 checks if it can be exploited by an attacker.

A project-root file that gives AI a "senior developer" identity with your coding standards, security guardrails, and organizational conventions.

AI goes vertical (straight down from prompt to implementation). You control horizontal (where on the landscape AI aims). Bad output = you aimed the vertical power at the wrong horizontal point.