People & Blogs

How to use Claude Goal. Best practices, how it works, and why you need it!

by Don’t Sleep On AI

Share:

✨ Temas principales

  • Qué hace GoalsEl comando /Goals permite que Claude Code trabaje de forma autónoma hasta completar una tarea. Una sesión independiente de Claude revisa los resultados para comprobar que se cumplieron los criterios, en lugar de depender solo de la afirmación de la sesión principal. 0:30
  • Preparar Claude CodeActualiza Claude Code a la versión más reciente y activa el modo automático o el modo de omisión de permisos para evitar interrupciones mientras trabaja. 1:31
  • Planificación detalladaAntes de usar Goals, convierte las ideas del proyecto en fases y tareas independientes, y guarda el plan en un sistema de memoria a largo plazo para que Claude pueda consultarlo. 2:34
  • Tareas y criterios de verificaciónPide las dos tareas siguientes del plan junto con criterios para confirmar que cada una funciona. Limita cada lista de tarea y verificación a 3.500 caracteres, ya que Goals admite hasta 4.000. 3:04
  • Revisar el rumbo del proyectoDespués de completar algunas tareas, pide a Claude que revise el plan, las conversaciones recientes y los registros de sesión para comprobar que sigue avanzando hacia los objetivos generales. 4:39

🧠 Conclusiones clave

  • La revisión independiente importaLa revisión adversarial ayuda a detectar tareas incompletas o errores que Claude podría pasar por alto al afirmar que ya terminó. Claude sigue trabajando hasta que la sesión independiente considera que se cumplió el objetivo. 1:01
  • Avanza en lotes pequeñosEl creador recomienda pedir dos tareas a la vez, o como máximo tres, porque pueden aparecer errores y bloqueos que cambien el orden del plan. 4:07
  • La memoria mantiene el contextoUn sistema de memoria a largo plazo ayuda a Claude a recordar el plan, el propósito del proyecto y los entregables, y reduce el riesgo de desviarse o tener que refactorizar más adelante. 5:42

📚 Lecciones aprendidas

  • Deja que el plan evolucioneA medida que aparecen nuevos requisitos y problemas, revisa y ajusta el orden de las tareas en vez de seguir a ciegas el plan original. 5:11
  • Repite el procesoSolicita las siguientes tareas con sus criterios de verificación y vuelve a ejecutarlas con /Goals para avanzar con comprobaciones continuas. 6:13

🏁 Conclusión y próximos pasos

  • Flujo de trabajo recomendadoPlanifica el proyecto con detalle, guarda el plan en la memoria, entrega a /Goals pocas tareas cada vez y revisa periódicamente si el trabajo sigue alineado con los objetivos principales. 6:13

Transcript excerpt

0:00 Hey Claude Code users, here's the best tip you'll receive today on coding with Claude. Anthropic just released a new goal slash command in Claude Code and it will significantly improve your AI coding if you start using it properly. I'll explain exactly why you want to use it and I'll show you exactly how to use it and I promise this will be a tool you will happily use daily because it's about to save you so much time and catch bugs in your coding before they ever exist. So, really quick, what is goals? This is a new command that gives Claude Code the ability to autonomously operate

0:30 on completing your coding tasks and Claude will keep working on a task until the goal of that task is achieved. In order to know if Claude achieved that task goal, he'll run a separate independent Claude session who will review the results to ensure the goal was met rather than simply having your primary Claude say that it's done. And guys, this is massive and solves a major problem which is Claude hallucinating that he finished the task or worse, Claude just wants to be lazy and never actually completes the task at all even

🔒 The full, searchable transcript is available with Pro.

🔒 Unlock Premium Features

This is a premium feature. Upgrade to unlock unlimited Q&A, transcripts, mindmaps, and translations.

Questions & Answers

Common questions about this video

What does the /goal command do in Claude Code?

It lets Claude Code work autonomously on a coding task until the task’s goal is achieved. A separate Claude session reviews the results to check whether the goal was met, and Claude keeps working until that independent review says it is. 0:30

Why does the video recommend using the /goal command?

The independent review can catch cases where Claude claims a task is complete when it is not. Claude also keeps working to complete and verify the goal instead of stopping prematurely. 1:01

What should you do before using /goal in Claude Code?

Update Claude Code to the latest version, then exit and log back in. Also enable auto mode or bypass permissions so Claude can work without repeatedly asking for approval. 1:31

What information should you provide for a goal?

Give Claude a clear task, such as what to build, fix, debug, or refactor, along with specific criteria for verifying that the task is complete. 2:01

How does the video recommend preparing tasks and verification criteria?

First work with Claude on a detailed project plan that breaks brainstorming ideas into phases and independent tasks, and save that plan in long-term memory. Then ask Claude for the next two tasks with verification criteria, keeping each task-and-verification list under 3,500 characters. 2:34

Why ask Claude for only two tasks at a time?

As Claude develops the project, it may discover bugs, blockers, or configuration needs that change the best task order. The recommendation is to provide two tasks at a time, or at most three, so the plan can adapt. 4:07

How can you keep Claude aligned with the project's high-level goals?

After every few completed tasks, ask Claude to review the plan and recent project conversations or session logs in your Obsidian memory system, and check whether it is still on track to achieve the project’s deliverables. 4:39

Why does the speaker say a long-term memory system is important?

It helps Claude remember the details and rationale of the project plan and its goals. Without it, the speaker says Claude may not accurately remember what the project is meant to achieve. 5:42

🔒 Unlock Premium Features

Access to Chat is a premium feature. Upgrade now to unlock unlimited studying tools.

🔒 Unlock Premium Features

Access to Mindmap is a premium feature. Upgrade now to unlock unlimited studying tools.

🔒 Unlock Premium Features

Access to Translation is a premium feature. Upgrade now to unlock unlimited studying tools.

Get unlimited summaries, Q&A, transcripts and more with Pro

Upgrade to Pro

Suggestions

🔒 Unlock Premium Features

Access to AI Suggestions is a premium feature. Upgrade now to unlock unlimited studying tools.