-
Notifications
You must be signed in to change notification settings - Fork 0
FAQ
Questions that come up while working through the curriculum, with answers. Grows organically — when an intern hits something confusing, the answer goes here.
If your question isn't answered, check the relevant chapter first, then open a discussion.
Check the Tool status page. Every external tool the curriculum depends on (SpecKit, Claude Code, Docker, Next.js, etc.) has a row with the last-verified date and current status.
If a tool isn't on that page and you suspect it's deprecated, open an issue and we'll verify and add it.
Don't take "X is deprecated" as gospel without verification. Open source projects get falsely declared dead all the time. The Tool status page is the canonical answer.
That's intentional. The longer story is in Chapter 02 / priming-patterns.md, but the short answer:
Summaries waste tokens AND lie. Every word Claude generates summarizing your CLAUDE.md is a word that could have been spent on real work, and the summary is always lossy — important rules get dropped or rephrased into something weaker. The terse confirmation tells you the read happened. That's all you need.
Because /compact poisons context. The war story is in principles/context-hygiene.md. TL;DR: a compacted summary kept "tests live alongside components" as a fact long after the project moved tests to a centralized directory. Claude wasted three message cycles patching files that didn't exist anymore.
/clear + /prep is cheap, fast, and clean. Use it.
Only for the 10 /speckit.* commands. The other 20 starter-kit commands (priming, git, quality, testing, session, reading) work without SpecKit being initialized.
If you want the SpecKit loop from Chapter 04, run this once per project:
uvx --from git+https://github.com/github/spec-kit.git@v0.5.0 specify initCheck Tool status for the current recommended SpecKit version before installing.
hello-world-ai is intentionally minimal. It's a starter for learning the workflow, not a production template. It exists so Chapter 01's exercise can be "clone, boot, ship a commit in 15 minutes" instead of "spend an hour bootstrapping a project."
For a production-grade Next.js template with auth, database, payments, etc., look at ScriptHammer — TurtleWolfe's actual production PWA template that the curriculum's principles are derived from.
Yes, but with one caveat:
-
Docker-first commands (
docker compose up, etc.) work fine in WSL2 - iOS/native mobile development (the ScanDo project) does NOT work on WSL — that needs macOS or EAS Cloud
- The curriculum itself is platform-agnostic markdown
WSL2 + Docker Desktop is TurtleWolfe's actual dev setup, so the workflow is tested on it.
The Day-1 starter (hello-world-ai) is Next.js because that matches the curriculum author's primary stack. The principles generalize:
- Docker-first applies to any stack
-
Secrets in
.envapplies to any stack - CLAUDE.md applies to any project type
- 5-file component scaffolding has Python/Go/Rust analogs (see principles/five-file-components.md)
The slash commands themselves are mostly stack-agnostic. /commit, /test, /code-review, /ship detect the package manager and test runner from package.json / pyproject.toml / go.mod / Cargo.toml.
If you want a Python or Go starter analogous to hello-world-ai, build one and PR it back — that's a valuable contribution.
Almost never. Chapter 05's when-this-helps.md goes into detail, but the short version: Chapter 04 is the daily move; Chapter 05 is the signature move. Most TSD interns will never need Chapter 05. Read it for the concepts, set it up only when you have a specific need (parallel wireframes, council governance, multi-stage pipelines).