learn by building

Mastering new technologies by reading documentation or watching videos is slow and passive. This skill forces you to learn by building, handling all the setup, boilerplate, and testing while leaving the core logic gaps for you to solve. You stay in the driver's seat of the implementation while the agent acts as an expert navigator and mentor.

What it does

• Structured learning follows a four-phase workflow from initial paper sketch to final architecture note.
• Tracer bullet generation creates the smallest possible end-to-end version of a system with core logic replaced by NotImplementedError gaps.
• Dynamic hint ladder provides support ranging from simple observation pointers to partial code examples based on your progress.
• Edge case testing generates challenges for concurrency, resource limits, and network failures to test your implementation.
• Knowledge capture synthesizes your explanations and transfer questions into a concise architecture note for future reference.

How it works

1. Define the topic by asking to learn a specific concept like a rate limiter, circuit breaker, or custom memory allocator.
2. Complete the sketch by answering questions about invariants, data flow, and potential failure modes to build the SPEC.md.
3. Fill the gaps in the generated skeleton code while the agent runs check commands and provides graduated hints.
4. Break the build by responding to specific challenges that test how your code handles stress and malformed inputs.

Frameworks & tools

The skill is language-agnostic. It automatically detects and uses existing project toolchains for Python, JavaScript, or other installed environments. It manages state via local SPEC.md and NOTE.md files within the project root.

Why this beats prompting it yourself

Generic prompts often just give you the answer, killing the learning process. This skill enforces a "driver/navigator" relationship where the agent is strictly prohibited from writing the core logic unless you explicitly hand over the keyboard, ensuring you actually internalize the mechanics of the code.

Use cases

• Learning internals like building a mini-Redux or a custom RPC framework from scratch.
• Onboarding developers to a complex internal pattern by having them implement a simplified version.
• Refactoring practice by stripping a feature to its skeleton and rebuilding it with better invariants.

Known limitations

Requires a working local interpreter for the chosen language. Core logic fixes that require entirely new concepts are moved to a "next topics" list rather than being taught mid-session.