Coworking patterns
Plan-then-build, asking for confirmation, code review loops.
How to work with Claude as a colleague rather than a vending machine: when to plan, when to ask, when to let it just go, and how to keep the loop tight.
All coworking patterns questions (9)
-
How do I get Claude to react to a Sentry alert while I sleep?
Channels are MCP servers that push events into your already-running Claude Code session — webhooks from CI or Sentry, DMs from your phone, anything that can speak HTTP. Here's the webhook-receiver pattern end to end, plus the sender-gating footgun that bites people.
-
How do I let Claude take notes about my project without writing CLAUDE.md myself?
Auto memory is the second persistence mechanism alongside CLAUDE.md — but you don't write it, Claude does. Here's how to find it, audit it, and decide which mechanism a given fact belongs in.
-
How do I make Claude Code react to a webhook without losing events?
Channels deliver events only while a session is open: at-most-once, no retry. Run Claude in a persistent terminal or background process, design idempotent handlers, and surface failures via a sender that does its own retry, because the channel layer will not.
-
How do I make Claude open a PR every Monday morning?
Routines are saved Claude Code configurations that run on a schedule, on demand via HTTP, or in response to GitHub events — on Anthropic's cloud, not your laptop. This is everything you need to set one up and avoid the obvious footguns.
-
How do I make Claude plan before it edits code?
Press Shift+Tab in the CLI until the input shows plan mode, or write 'do not edit yet, write a plan'. Plan mode holds source edits until you approve; reads and shell commands still run (with normal prompts). Approve only plans that list exact files, an explicit out-of-scope list, and a verification command.
-
How do I see what's eating my Claude Code budget?
Run /usage in any session for the local dollar estimate; the Claude Console Usage page is authoritative. For teams, an admin workspace centralizes tracking; for org-wide attribution, large deployments often add LiteLLM in front. The biggest wins come from model choice, prompt-cache hit rate, and not running an idle agent team.
-
How do I start a plan locally, work on it remotely, and finish in my terminal?
Type /ultraplan in a Claude Code v2.1.91+ session to hand the planning task to a Claude Code on the web cloud session running in plan mode. The CLI shows a status indicator while you work; when the plan is ready, choose to execute in the cloud or teleport back to your terminal.
-
How do I wire Claude Code into my GitHub PRs without giving it my production secrets?
Install the Claude Code GitHub Action via /install-github-app, scope ANTHROPIC_API_KEY as a repo or environment secret, and pass only the secrets a PR job actually needs. The webhook is GitHub's; the work runs on GitHub's runners; the failure mode is mis-scoped secrets, not the action itself.
-
Which Claude code review tool should I actually use?
Three reviewers ship inside Claude: a single-pass local /review, a cloud-sandbox multi-agent /ultrareview, and a GitHub App that runs on every PR. They cost different amounts, run in different places, and answer different questions. Here's the picking guide and the per-tool footguns.