← Learn
Playbook · 10 steps · checked 5 October 2026 · 0 adopted

Claude Code

At most ten steps, in order of how much they help. Each step is from Anthropic's own pages, linked. Re-checked against those pages every week; anything that can't be confirmed is marked unverified rather than removed.

  1. 01

    Run /init, then cut

    /init writes a starter CLAUDE.md from the code. Keep only what Claude would get wrong without it: commands it cannot guess, style rules that differ from the defaults, gotchas. For each line ask "would removing this cause a mistake?". Bloated files get ignored. Run /context to check it loaded. (best practices)

  2. 02

    Give Claude a check it can run

    A test command, a build, a lint, or a screenshot to compare. Put the command in CLAUDE.md. Without a check, "looks done" is the only signal. (best practices)

  3. 03

    Pre-approve what you trust, block what you don't

    Use /permissions to allow the test and lint commands and deny reading .env. Use /sandbox if you want fewer prompts with OS-level limits. (best practices)

  4. 04

    Add the secret guard

    Add a git pre-commit hook that refuses commits containing keys (ask Claude to write one, or use a scanner such as gitleaks). Rules that must always hold belong in hooks, not in CLAUDE.md. (hooks)

  5. 05

    Switch tools on for this project only

    Add MCP servers with claude mcp add --scope project (written to .mcp.json) and plugins with claude plugin install <name>@<marketplace> --scope project (or local for just you). Keep user scope (every project) for the few things you use everywhere. This is what stops tools piling up. (plugin scopes) The starter kits on this site list good ones for each kind of project.

  6. 06

    Typed language? Add a code intelligence plugin

    Claude then sees type errors after its own edits and finds code by symbol: typescript-lsp, pyright-lsp, swift-lsp and others from the official marketplace. Install the language server first (see the plugin's README). Terminal sessions only; it does not run in cloud sessions. (code intelligence)

  7. 07

    Plan big changes, keep the context clean

    Plan mode (Shift+Tab) for anything touching several files: explore, plan, then build; skip it if the change fits in one sentence. For a big feature, have Claude interview you and write SPEC.md, then build it in a fresh session. /clear between unrelated tasks, and after two failed corrections. Subagents for research. (plan first, manage your session)

  8. 08

    Review in a fresh context

    Run /code-review on the diff before you call it done, so the agent that wrote it is not the one grading it. (best practices)

  9. 09

    Make it work in cloud sessions

    A cloud session (claude.ai/code, the phone app, routines) gets only what is in the repo: CLAUDE.md, .mcp.json (single-repository sessions only), and the repo's .claude/skills and agents, plus skills you switched on in your claude.ai account. It does not get your user skills, MCP servers added at local or user scope, or plugins (not even ones the repo's settings turn on). So: MCP at project scope (step 5), commit any skill the project needs into its own .claude/skills, and install dependencies with a SessionStart hook. Toolchains that are not pre-installed go in the cloud environment's setup script. (what carries over)

  10. 10

    Prune monthly

    /plugin (Installed tab, "Not used recently"), claude plugin details <name> for what one costs per session, /mcp for servers. Keep only the few you use in every project switched on everywhere; turn off claude.ai account skills you never use in the Claude app. (find unused plugins)

Picked from around the web, with a one-line takeaway. The originals are by their creators.

  1. instagram

    Some guy on GitHub just gave Claude eyes, so it can read anything on the internet ↗

    Charlie Hills (@charliehills) · still works (checked 8 Oct 2026)

    Introduces Agent Reach, a GitHub tool that lets Claude Code read more of the web, such as YouTube transcripts, RSS feeds, Reddit and LinkedIn, without paid API keys.

  2. instagram

    Claude Code just dropped Mods. Here are 3 ways to use them ↗

    Joshua Stevenson (@nocode.joshua) · still works (checked 7 Oct 2026)

    Shows three uses for Claude Code mods: sessions that message each other, a custom dashboard in the interface, and routing simple requests to a cheaper model.

Unverified

What Anthropic's own pages disagree on, or could not be confirmed.

  • /skill-doctor (skill cost and 7-day usage) is described in Claude Code's built-in reference but was not found on code.claude.com on 2026-10-04. Try it; if it is missing on your version, use /context instead.