Capstone brief: Karim ships a real PR

Six weeks ago, Karim was prompting "write a Python function to dedupe a list" and getting tutorials. By the end of this capstone, he'll have shipped a 200-line PR to a real open-source repo using only the prompt patterns from this course. So will you.

This module's flow:

The capstone is structured as four artifacts:

1. A picked issue. A real "good first issue" or small enhancement on a repo you actually use.
2. A code-explainer system prompt. A reusable system prompt for understanding any unfamiliar function in the codebase.
3. A solution prompt sequence. The actual prompts you used — codegen skeleton, debug, refactor, review — to produce the PR.
4. The PR itself. Description written by prompt. Conventional commit. Tests included. Self-review passed.

The capstone is graded on a rubric, not a binary "did it merge" — that's outside your control once the PR is open. The rubric in lesson 4 measures the quality of your prompting, which is what this course teaches.

Why a 200-line cap? Because a 200-line PR is the largest size where the four-block skeleton from module 1 still fits in a single iteration. Beyond that, you start needing the agentic patterns from module 5 (Claude Code planning), which deserve their own course. 200 lines is also about the size of a real "feature" PR in most production codebases — a fix, a small endpoint, a small UI component.

What counts as a real PR? Three criteria:

Criterion	Why
Public repository (GitHub, GitLab, etc.)	Forces real conventions and review
You did not write it before discovering this course	Isolates the new skill
You used at least 4 different prompt patterns from the course	Proves the skeleton generalises

Pick something achievable. A typo fix doesn't count — there's no codegen, no review, no test. A library you don't understand doesn't count — you'll burn weeks reading code instead of practising prompts. The sweet spot is a small feature or a small bug fix on a repo you use and understand.

The end-to-end capstone pipeline you'll build:

The four modules so far have given you the building blocks:

• Module 1 — the codegen skeleton for the new code you'll write
• Module 2 — the debugging prompts for when your first attempt is wrong
• Module 3 — the refactor + PR description prompts for shipping
• Module 4 — the review prompts to self-review before pushing
• Module 5 — the tool-specific shapes (Cursor, Aider, conventional commits)

This module wires them together. The next three lessons walk through (1) picking the right issue, (2) building the code-explainer system prompt that helps you understand the existing code, and (3) the rubric you'll grade yourself against.

Next up: how to pick an issue Karim won't regret. :::