Berlin Release Night: Claude Code Slash Commands Tested

Coding-agent routes EN2026-10-30

Testing Claude Code during a Berlin release

At 18:40 on a Thursday in Berlin, our team had one release candidate left to verify. The task was familiar: inspect a changed authentication flow, run focused tests, review the diff, and keep the coding session small enough that Claude Code stayed useful. I used Claude Code for the full pass and treated its slash commands as workflow controls rather than shortcuts.

The test focused on commands that save time repeatedly. I measured their value by three things: how quickly they reduced manual work, whether they preserved useful context, and whether they helped me avoid restarting the session.

Using /status and /cost before the first edit

I started with /status. It showed the active session details and made it easy to confirm that Claude Code was working in the intended repository. That small check prevented a common mistake: asking for a review while standing in a sibling project with a similar folder name.

Next, I used /cost after the initial repository inspection. The command gave me a quick view of the session’s usage, which helped me decide whether to continue in the current context or reset before deeper analysis. It is especially useful when a broad request has caused Claude Code to read more files than expected.

These commands take seconds, but they create a reliable checkpoint before the expensive part of the work.

Using /compact when the investigation gets noisy

The biggest time saver was /compact. During the authentication review, the session accumulated test output, file summaries, rejected approaches, and several long explanations. I ran /compact after the first investigation and asked Claude Code to preserve the current findings, open issues, and next actions.

The resulting context was much easier to work with. I could continue the review without repeating the repository map or explaining the same bug. The command works best after a meaningful milestone, such as locating the failing path or finishing a test run. Running it too early can remove useful detail, so I treated it as a checkpoint rather than an automatic cleanup command.

Using /clear and /memory between unrelated tasks

After the authentication review, I needed to inspect a deployment script. I used /clear instead of continuing with the old conversation. That gave the second task a clean context and reduced the chance that earlier assumptions would influence the new analysis.

I checked /memory before starting the next session. This command helped me inspect the project instructions Claude Code could use across sessions. The practical lesson was simple: keep durable repository conventions in the project memory, and keep temporary debugging details in the active conversation. That separation saved more time than repeatedly correcting an assistant that had inherited stale task context.

Using /model, /permissions, and /review at the decision points

For the final pass, /model made it quick to confirm which model was active before I requested a large review. When the task touched files outside the normal working area, /permissions provided a direct way to inspect or adjust the session’s access rules.

I finished with /review. It gave the release candidate a focused review pass against the current changes. I still read the diff and ran the tests myself, but the command created a useful first filter for missing validation, risky edge cases, and accidental changes.

For a cost-sensitive workflow, model selection also matters. RelayVano can connect Claude Code to available models such as Claude Sonnet 5.5 and DeepSeek V4 Pro through its pay-as-you-go relay, using the API base https://api.relayvano.com.

The three commands that earned a permanent place

/compact saved the most time during long investigations by preserving the useful state of a crowded session.
/clear made task boundaries explicit and prevented unrelated work from sharing context.
/review turned the final diff check into a repeatable step before tests and human approval.

I also kept /status, /cost, /memory, /model, and /permissions in my regular toolkit. They are less dramatic, but they make the main workflow safer and easier to control.

Recommendation after the Berlin field test

My recommendation is to sign up for RelayVano if you want a pay-as-you-go way to run Claude Code and other compatible coding tools across available models without changing the workflow built around these commands. Start with /status, use /compact at investigation milestones, run /clear between tasks, and finish with /review. That sequence delivered the clearest time savings during the release test.