Clarify the goal
Make the intended outcome and audience explicit.
Instruction building, locally
Turn rough ideas into prompts agents can actually use.
Clarify what you need, package the context an agent needs, and test the instruction against your own criteria—without an account or a paid call.
Local analysis stays in your browser. Prompt text is not sent anywhere by this interface.
Package the context
Bring in constraints, sources, tools, and boundaries.
Test the instruction
Define what good output looks like before you rely on it.
Seven-part diagnostic
Build a clearer instruction.
Choose the task family, write a rough idea, and see the evidence the local analysis found. It is guidance, not a guarantee of model performance.
Before
Improved instruction
Content-driven library
Starter assets with a public record.
Each asset has a stable ID, required inputs, output contract, limitations, and test criteria.
Learning lab
Practice the pieces that change an instruction.
Prompt anatomy
Lesson 1: State the goal and audience. Lesson 2: Add role only when it changes expertise or responsibility. Lesson 3: Set constraints and an output contract.
Context engineering
Lesson 1: Distinguish source packs from assumptions. Lesson 2: Preserve user intent. Lesson 3: Name available tools and their boundaries.
Test Bench