← Volver a las guías

Claude Code Shortcuts: A Practical Developer Guide

Use Claude Code as a Decision Tool

Claude Code is most effective when you give it a clear job, a bounded context, and a way to verify the result. The biggest productivity gains usually come from a few repeatable commands and habits, not from writing longer prompts. Use this guide to decide which shortcut fits the task in front of you.

1. Start with the Right Context

At the beginning of a repository session, run /init if the project does not yet have a useful Claude Code memory file. Then ask Claude to inspect the relevant directories, package configuration, test commands, and contribution rules before changing anything. For a focused task, name the files or subsystem explicitly. For example: Review the authentication middleware and its tests, then list likely failure points before editing.

Do not load the entire repository by default. Narrow context produces faster responses, lower token usage, and fewer irrelevant edits. When Claude starts making assumptions, use /clear to begin a clean session, or restate the current goal and constraints before continuing.

2. Choose the Model for the Job

Use /model to inspect or change the available model selection. A practical rule is to reserve Opus for difficult architecture, unfamiliar code, and multi-step debugging; use Sonnet for most implementation and review work; and use Haiku for quick searches, small transformations, and routine explanations. The fastest model is not always the cheapest overall if it produces a patch that needs repeated correction.

When API cost matters, ask for a plan before implementation. A short planning pass can prevent a long, incorrect edit cycle. After approval, request the smallest patch that satisfies the plan.

3. Control Context Instead of Starting Over

Long sessions accumulate stale assumptions. Use /compact when the conversation is getting large but the task is still active. Before compacting, ask Claude to preserve the current objective, changed files, test results, unresolved questions, and important constraints. This gives the compressed context useful information instead of a vague summary.

Use /cost when available to inspect session usage, and use /status to check the active model, permissions, and connection details. These commands are especially useful when a task unexpectedly becomes expensive or behaves differently from a previous session.

4. Use Permission Modes Deliberately

For investigation, keep permissions restrictive and let Claude propose commands and edits. For a trusted, repetitive workflow, use Shift+Tab to cycle through the available permission or approval modes in your installed version. Do not enable broad auto-approval in a repository containing production credentials, deployment scripts, or destructive database commands.

Use Ctrl+C to cancel a running action, Ctrl+L to clear the terminal display, and the up and down arrow keys to reuse earlier prompts. If an edit is surprising, stop immediately, inspect the diff, and ask Claude to explain the change before accepting more work.

5. Make Verification Part of Every Prompt

Every implementation request should include a verification step. Ask Claude to run the narrowest relevant test first, then formatting, linting, type checking, or the full suite as appropriate. A useful request is: Implement the parser change, update affected tests, run the focused test file, and report any remaining failures without hiding them.

For reviews, ask for findings ordered by severity with file and line references. Then request tests for each confirmed issue. This turns Claude Code from a patch generator into a review and feedback loop.

6. Reduce API Cost Without Reducing Quality

Pay-as-you-go access is often a better fit than a fixed subscription for irregular development work. 59API provides Claude Opus, Sonnet, Haiku, and Fable access through https://api.59api.com, alongside GPT models, while remaining compatible with Claude Code, Codex, and OpenAI SDKs. Developers can configure Claude Code with the 59API base URL and account key according to the relay's setup instructions, then choose the model that matches each task.

Because 59API uses native official-quality models rather than downgraded substitutes, the cost decision does not require accepting an intentionally weaker model. Its low-cost, usage-based pricing and referral rebate can make experimentation and bursty workloads more economical. Developers comparing relay options can sign up for 59API and test the same Claude Code workflow with measured usage.

Claude Code Checklist

• Before editing: identify the exact files, constraints, and acceptance criteria.
• Before a long task: choose the model with /model and ask for a plan.
• During the session: use /compact when context grows and /clear when assumptions become polluted.
• Before approving commands: check permissions and inspect destructive operations.
• After every patch: review the diff and run focused tests, linting, or type checks.
• When costs rise: check /cost, switch routine work to a smaller model, and keep prompts scoped.