brainmaxxing is published by Julia Muiruri from juliamuiruri4/brainmaxxing, pinned to 8a8a102ff3ab8736f015764207cdb7e68fcd5ecf. Review the source before installing.
Agent skills that quiz students on software engineering concepts grounded in their actual codebase. Designed for early-career developers who co-create projects with AI and need to verify their conceptual understanding.
The problem
AI coding agents make it easy to build sophisticated software without fully understanding the concepts behind it. BrainMaxxing helps students verify whether they actually understand what's in their project — or if they're just running code they can't explain.
How it works
Invoke
/brainmaxin your projectThe agent analyzes your codebase and detects which knowledge domains are implemented
You pick a domain (API Design, Database Design, Testing, etc.)
The agent quizzes you with 5 questions grounded in your actual code
Rooted in Bloom's Taxonomy, questions escalate through cognitive levels:
- Explain (Remember/Understand) → describe what this code does and why it exists
- Predict (Analyze) → trace behavior through code without running it
- Refactor (Apply/Evaluate) → identify tradeoffs and improve what's there
- Debug (Analyze/Evaluate) → spot a real issue and reason about its consequences
Each answer is scored on a 0–3 rubric
Evaluation and scoring mirrors the progession above:
- 0 → no recognition
- 1 → names it but can't apply it
- 2 → explains and reasons about change
- 3 → evaluates tradeoffs and could teach it.
When done, "compile report" to get your competency summary with personalized study recommendations
Skills
| Skill | What it quizzes |
|---|---|
/brainmax |
Orchestrator — detects domains, routes to quizzes, compiles report |
/api-design |
REST/GraphQL/gRPC design, HTTP semantics, middleware |
/database-design |
Schema design, normalization, queries, migrations |
/system-architecture |
Service boundaries, communication, infrastructure |
/implementation-patterns |
Design patterns, abstractions, code organization |
/testing-strategy |
Test design, mocking, coverage, test pyramid |
/security-fundamentals |
Auth, validation, encryption, OWASP concepts |
/devops-and-ci-cd |
Pipelines, deployment, infrastructure as code |
/error-handling-and-resilience |
Retries, circuit breakers, graceful degradation |
/requirements-and-scope |
Requirements engineering, scope management |
/domain-modeling |
DDD concepts, aggregates, business rules |
/ui-and-frontend |
Component architecture, state, rendering, accessibility |
/observability |
Logging, metrics, tracing, health checks |
Installation
Install the complete BrainMaxxing plugin to get all skills and the interactive BrainMax Canvas:
copilot plugin install juliamuiruri4/brainmaxxing
Then start Copilot in the project you want to assess and invoke /brainmax.
Skills-only fallback
Clients without Copilot plugin or Canvas support can install the same quiz skills without the Canvas:
npx skills add juliamuiruri4/brainmaxxing
Invoke /brainmax after installation. The complete quiz flow remains available in chat.
Maintaining bundled references
Each installed skill is self-contained. Runtime reference files live inside the individual skill directories so npx skills add works without needing a repository-level shared/ directory.
shared/ remains the canonical authoring source for shared quiz/orchestrator reference material. After editing a file in shared/, sync the bundled copies and verify they match:
./scripts/sync-skill-references.sh
./scripts/sync-skill-references.sh --check
Design principles
- Read-only: Skills analyze code, never edit it
- Grounded: Every question references specific code in the student's project
- Honest: Never fabricates bugs, issues, or code that doesn't exist
- One-at-a-time: Questions are asked sequentially, not batched
- Rubric-scored: All answers scored on a consistent 0–3 rubric
Canvas extension
The BrainMax Canvas is included with the plugin and launches when /brainmax publishes its detected domains. Students can select a domain, read each question, submit freeform answers, follow their score, and review the compiled report from the canvas. Answers are relayed into the Copilot session; question generation and rubric scoring remain with the domain skills.
If the host does not open the canvas when set_domains is invoked, open BrainMax once from the Copilot App. The current quiz state will then stream into it automatically. Skills-only installations continue entirely in chat.
License
MIT