In a recent LinkedIn post, Ruben Hassid explores a sophisticated 11-step framework for crafting prompts for the advanced AI model Claude 5. Hassid, a proponent of leveraging AI for complex tasks, outlines a structured approach designed to maximize the AI’s capabilities, moving beyond simple queries to a more collaborative and effective interaction model. He emphasizes that the key to unlocking Claude 5’s potential lies in understanding its advanced features and guiding its operation with precision.
Hassid begins by detailing the foundational elements of his prompt structure, stressing the importance of starting with the ‘why’ rather than the ‘what’. According to Hassid, Claude 5 excels at connecting disparate pieces of information to achieve a larger goal.
“Start with why, NOT what. Claude 5 connects the dots. ‘I’m working on [goal] for [who it’s for]. They need [what the output enables]. With that in mind: [task].'”
This initial step, as Ruben Hassid explains, sets the strategic direction for the AI, ensuring its efforts are aligned with the user’s overarching objectives. He then moves to the crucial aspect of providing context, advocating for the use of ‘Context Files’ to upload expertise rather than embedding lengthy explanations within the prompt itself.
Leveraging Expertise and Setting Expectations
Hassid highlights that Claude 5’s ability to process external files, referred to as ‘Context Files’, is a game-changer. This allows users to “upload their expertise” and avoid lengthy, repetitive explanations within the prompt. The AI is instructed to read these files thoroughly, effectively using them as its knowledge base.
“Read these files completely before responding: [filename .md] – [what it contains]. The file is the brain. This part never changes.”
Furthermore, Ruben Hassid points to the value of providing a clear ‘Reference’ example. He argues that demonstrating desired output with a concrete example is far more effective than providing numerous instructions. This visual or textual guide helps Claude 5 understand the target quality and format.
Challenging AI with Complex Problems
A significant portion of Hassid’s framework focuses on challenging the AI with difficult tasks. He introduces the concept of ‘Effort’, categorizing problems as ‘routine’, ‘hard’, or ‘hardest-unsolved’.
The Importance of ‘Hardest-Unsolved’ Problems
Ruben Hassid argues that teams often undersell AI models like Claude 5 by testing them on simple tasks. Instead, he urges users to:
“This is a [routine / hard / hardest-unsolved] problem. Scope it like it’s at the top of your range. Teams testing Claude 5 on easy tasks undersell it. Give it your hardest problem.”
This approach, according to Hassid, is essential for truly evaluating and developing the AI’s advanced problem-solving capabilities.
Advanced Control and Collaboration
Hassid also details mechanisms for controlling Claude 5’s output and fostering a more collaborative environment. The ‘Act’ directive, for instance, encourages the AI to act decisively once it has sufficient information, avoiding unnecessary deliberation or re-evaluation of user decisions.
The ‘Scope’ directive is introduced to manage Claude 5’s tendency to ‘over-deliver’. Hassid advises users to instruct the AI to “Do the simplest thing that works well. No extra features, refactors, or abstractions.” This ensures the AI remains focused and efficient.
Delegation and Evidence-Based Reporting
Ruben Hassid highlights the potential for ‘Delegation’ by splitting tasks across multiple ‘subagents’, transforming the AI from a single chatbot into a ‘team lead’. This enables parallel processing and more complex workflows.
To combat fabricated progress reports, Hassid introduces the ‘Evidence’ directive. This requires the AI to audit all claims against tool results before reporting, ensuring accuracy and transparency. As Hassid notes, this nearly eliminated fabricated status updates during Anthropic’s testing.
Memory, Checkpoints, and Reporting
The framework concludes with ‘Memory’, ‘Checkpoint’, and ‘Report’ directives. ‘Memory’ involves recording learnings in separate files to build a persistent knowledge base for the AI. ‘Checkpoint’ allows users to define specific conditions under which the AI should pause, preventing premature or delayed task completion.
Finally, the ‘Report’ directive emphasizes starting communications with a clear outcome or TLDR (Too Long; Didn’t Read), ensuring that the most critical information is immediately accessible.
Hassid provides a pathway for readers to access his full prompt template and associated .md files via how-to-ai.guide, encouraging them to subscribe for free to receive a welcome email containing download instructions.
📝 About This Content
This article is based on insights shared by Ruben Hassid on LinkedIn.
📅 Originally posted on June 11, 2026 | View original post on LinkedIn →