Summary: The author spent six months using Claude Code knowing essentially one command — typing code and hoping — before discovering an ecosystem of productivity multipliers hiding in plain sight. The reframe: when the model seems to be the problem, the real fix is usually a better workflow or the right command. (Command behaviors below reflect the author's April 2026 experience; the Claude Code command set evolves, so confirm specifics against current docs.)

Setup commands — run once, benefit forever

  • /init — auto-generates a CLAUDE.md by reading the whole project structure and codebase: project summary, architecture decisions, coding conventions, key dependencies, setup instructions. Replaces ~45 minutes of manual documentation per project.
  • /memory — opens a persistent configuration that applies across every project and session. Set tone preferences, coding standards, and pet peeves once (e.g. "always suggest tests first," "prefer composition over inheritance," "never use any without explicit permission") and Claude remembers.
  • /pr_comments — pulls every GitHub pull-request comment into the session before you write a line, eliminating the GitHub ↔ Claude tab-switching that used to eat 30–40 minutes per PR.

Daily drivers

  • /btw — ask a side question (/btw what's the best way to handle refresh tokens?) without losing your place in the main task. Lets you hold multiple threads of thought without cognitive overload.
  • /compact — compresses a long conversation into a clean summary, preserving key decisions and code while freeing the context window. The alternative to panicking and starting over when a session is hundreds of messages deep.
  • ! command — run a shell command directly in Claude (! git status, ! npm run test, ! git add .) and the output lands right in the conversation. No tab-switching, no copy-paste.
  • /cost — a fuel gauge for the session: total cost, tokens used, context remaining — so you know when to /compact before you're stranded.

Power moves

  • /model — switch models mid-conversation. The author's pattern: Opus for the heavy thinking (e.g. designing a complex database migration), Haiku to generate the code fast, Sonnet for the review — all in one session without losing context.
  • /fast — toggle faster responses from the same model, optimized for speed over depth. Good for brainstorming, quick refactors, iterative debugging, and "just make it work" moments; toggle off for careful review.
  • /review — a systematic code-review pass covering bugs and edge cases, logic errors, code clarity, performance, and security, with line-referenced findings — not a casual rubber stamp.

Recovery commands — when things go wrong

  • /clear — wipes the conversation but keeps setup, tools, permissions, and project context intact. A soft reset, not a nuclear option, for when the conversation has gone in circles.
  • /doctor — checks setup health: API key validity, Node version, permissions, config files, terminal integration. The IT support ticket you resolve in seconds.
  • /terminal-setup — configures shell integration so Claude can actually see terminal output.
  • /help — lists every command with a short description; the starting point when lost and a good first stop for onboarding teammates.

The transformation

The difference between fighting the tool and shipping faster wasn't talent or experience — it was a handful of commands that turned Claude Code from fancy autocomplete into a development partner. The author's suggested adoption path is one command at a time: /init, then /memory, then /btw, /compact, !, /review, and finally experimenting with /model switching.