# Tips

Source: https://www.storeframe.io/docs/get-started/claude-code/tips

Patterns for productive Claude Code work on StoreFrame environments.

Patterns that pay off when working with Claude Code in StoreFrame environments.

![Claude Code session in the Hub Console tab showing a multi-step investigation workflow](https://www.storeframe.io/assets/docs/get-started/claude-code/claude-console-tab-showing-multi-step-investigation-workflow.png)

## Use the pre-loaded context

Don't explain Magento basics to Claude — it knows them. Skip the background and jump straight to the specific:

* Instead of: "Look at the Magento exception log to find errors. Magento stores logs in `var/log`."
* Write: "Find the most-common stack trace in `exception.log` from the last hour."

## Bound the scope of long-running tasks

Claude will refactor an entire module if you ask. Bounding scope keeps changes reviewable:

* Instead of: "Improve the checkout module."
* Write: "In `app/code/Vendor/Checkout/Plugin/Discount.php`, the `apply` method uses nested conditionals. Refactor to guard clauses — no behavior changes."

## Ask for a plan before any code is touched

For multi-step work (migrations, refactors, debugging chains), ask Claude to write a plan you can review first:

* "Plan a migration that adds an index on `sales_order.created_at` and a follow-up script to verify the index is used. Don't write any code yet — just the steps."

Once the plan looks right, ask Claude to execute step by step.

## Query the database directly

Claude can run MariaDB queries via the local socket. Common patterns:

* "Show me the 10 most recent orders with status `pending_payment`"
* "Count orders by status for the past 30 days"
* "Find SKUs where `qty = 0` but `is_in_stock = 1`"

## Commit and open a PR from Claude

If your environment has a configured git remote, Claude can handle the commit and PR:

* "Commit the changes you just made with a descriptive message, push to a new branch named after the issue, and open a pull request."

Claude generates the branch name, commit message, and PR description.

## Recover from unwanted changes

If Claude did something you didn't intend:

* `git diff` — see what changed
* `git checkout -- <file>` — discard uncommitted changes to a specific file
* `git revert <sha>` — undo a committed change cleanly without rewriting history

For operations that already ran and can't be undone by git (a cache flush, an index rebuild, a queue consumer trigger), ask Claude to re-run the correct version.

## Multi-tab workflows

Open multiple Console tabs for parallel work. A common split:

* Tab 1 (Claude) — investigating or fixing the bug
* Tab 2 (Bash) — `tail -F var/log/exception.log` so new errors appear live
* Tab 3 (Bash) — `mysql` or `docker exec -it mariadb mysql` open for ad-hoc queries

Naming tabs after what they're for — `investigate`, `tail-logs`, `mysql` — makes switching fast and keeps [/remote-control](/docs/get-started/claude-code/generic-commands/remote-control) entries legible on other devices.

## When to use Codeserver instead

Switch to Codeserver (VS Code in the browser, launched from the environment header on desktop) when:

* You need to read or edit many files in parallel with syntax highlighting and IDE navigation
* You're doing sustained authoring rather than diagnosis
* You want Claude Code inside a full editor environment (it's available in the Codeserver integrated terminal too)

The Console tab and Codeserver share the same VM — you can have both open at once.
