Learning path map
Detailed content
✈️ CODA & Claude Code
The professional framework for coding with AI: separate the IDE (the fighter jet) from the model (the pilot), and work through the five stages — Configure, Outline, Deploy, Automation, Review.
The IDE (AntiGravity, Cursor) is the plane; the model (Claude Code, Gemini) is the pilot. You switch pilots depending on the mission.
Understanding this separation gives you freedom: the skill isn’t tied to a tool or a model.
Tool vs. model · pilot swap · preflight checklist.
A context file (claude.md / Gemini.md) in EVERY project: tech stack, directories, commands and — most importantly — success criteria + how to test.
It's the backbone of the project, a living document that helps the AI get things right without you having to repeat everything.
Context · success criteria · living document.
Define the problem and success BEFORE writing code. Back-and-forth conversation, clarifying questions, “ultrathink” for deep reasoning.
Spending ~95% of the time on the plan avoids rework and code that solves the wrong problem.
Plan Mode · specification · ultrathink.
Run multiple agents at the same time (5–7), from the terminal or IDE. You can even use web agents connected to a repo, running with your laptop closed.
Parallelizing multiplies your output — several problems move forward at once.
Parallelism · web agents · throughput.
Create specialized subagents, one per subtask (e.g., title, intro, script). Each has its own context window.
Splitting work among subagents prevents context overflow and keeps each task focused.
Subagent · context window · specialization.
Explicit success criteria (in claude.md): runs without errors, tests pass, no lint warnings, fulfills the request, changes are explainable. Ask for a review as a "skeptical senior engineer".
Without clear criteria, "done" becomes a matter of opinion. Verification is what separates amateurs from professionals.
Success criteria · verification · skeptical review.
Sometimes a short prompt works better than a super-detailed one. Try the short one first and see if the AI gets it right on its own.
Too many instructions can confuse the model. Simplicity saves time and tokens.
Brevity · iteration · letting AI infer.
🧰 Skills & Prompts Library
Seven ready-to-download artifacts to adapt: cloning and conversion skills, a debug prompt, an MCP config, a claude.md template, a QA checklist, and the prompt collection.
A protocol that rebrands a template: logo, copy, brands, prices, reviews, and color.
Reuses a validated structure for a new niche in minutes.
Template · rebrand · replacement checklist.
Converts a framework-based website into a single self-contained HTML file, ready to paste into GoHighLevel or WordPress.
Lets you deliver where the client already is, without starting over from scratch.
Self-contained · inline · portability.
A structured template (XML) for debugging with AI: generates predictions, investigates, finds the root cause, and gives step-by-step instructions. (Third-party open-source tool.)
Turns “it’s showing an error” into a disciplined troubleshooting process.
Prediction · root cause · step-by-step.
Ready-to-use MCP configuration (n8n, Notion, Pinecone, Supabase, Vercel, Fireflies) with key placeholders.
Connects your agent to external tools without setting everything up by hand.
MCP · servers · keys/placeholders.
The CODA context file template: tech stack, directories, commands, success criteria + how to test.
It's the starting point for the "C" in Configure in any new project.
Context · stack · success + testing.
A phased QA and stress-testing protocol that classifies each finding as Critical / Warning / OK.
Ensures you publish only after checking what matters — before "taking off".
QA · severity · pre-publication.
A collection: problem→SOP, Plan Mode, research→research.md, design inspiration, review/verify, screenshot-based troubleshooting, connecting MCP, env var, and “less is more.”
It’s your Swiss Army knife: copy, paste, and adapt it for any CODA step.
Reuse · SOP · living library.