Local VS Code

Run the same editor-first loop on your machine. The Dev Container matches Codespaces; a native install needs only uv and just for the core flow.

Set up

git clone https://github.com/Jesssullivan/dsa-study-packet.git
cd dsa-study-packet

Choose one lane:

uv sync --extra dev
just doctor

Start a rep

With Copilot Chat installed and signed in, enter /comments. It starts a rep whose reasoning lives in ordinary source comments or docstrings, with no required labels or prefixes. Add a topic and problem to choose one:

/comments arrays two_sum

If named vocabulary helps you think, /reacto, /clarp, and /umpire start the same loop with optional labels. Copilot is optional; use the conductor directly:

just practice-start comments
just practice-start comments arrays two_sum

To study one exact pair without starting a rep, open its committed snapshot:

just practice-study linked_lists lru_cache

When ready, start a candidate pair with source or test focus.

just practice-start comments linked_lists lru_cache
just practice-start-tests linked_lists lru_cache

Your source and test file open under .challenges/workspace/. Write ordinary source comments or docstrings in your own words, save, then enter /continue or run just practice-next. The comments belong in the file, not Chat, and need no prefixes, minimum count, or gate deletion. Implement the solution and add focused tests. If the tabs do not open, the command prints their paths; just practice-open tries again.

Continue and close

just practice-next       # state and one next action
just practice-test       # reference tests plus your tests
just practice-watch      # rerun on workspace changes
just practice-repl       # interactive exploration
just practice-open       # reopen both files
just practice-finish "one fix"

The committed solution under src/algo/ remains unchanged, and your workspace stays gitignored. Run just doctor when the toolchain looks wrong.

Candidate tests are not sandboxed. The runner bounds test time and cleans the pytest process group. Do not launch background daemons. Receipts detect ordinary staleness and incomplete runs; they are workflow evidence, not a tamper-resistant boundary.

Claude Code, Codex, and other external agents are optional. Install, authenticate, and launch the one you already use; the repository does not start one or require its credentials. A direct just session works without any agent.

Ask the resident interviewer for an untimed conversational rep when you want less editor pressure. Choose a timed board-style rep only when speaking under a clock is the skill you mean to practice.