HarmonyFidelisHarmonyFidelis
Login
AI foundationsHow LLMs workLLM skillsEvaluationGlossaryBuilder

Aura · Learn to instruct AI

Turn an intention into a useful, verifiable instruction.

Choose a goal, add the context that matters and select a working method. Aura assembles a portable prompt locally in your browser—without an account, API call or saved history.

Build an instructionBrowse free templates
  • Local builder
  • No account required
  • 12 free templates
  • ChatGPT · Claude · portable

A simple contract before a clever prompt

Useful instructions make the task, boundaries and definition of success observable. Model choice comes after that foundation.

  1. 01

    Frame the job

    State one concrete objective and provide only the context needed to act.

  2. 02

    Set boundaries

    Specify constraints, expected format and the criteria a good answer must satisfy.

  3. 03

    Inspect the result

    Ask for assumptions, evidence, uncertainties and checks you can actually review.

Instruction configurator

The configurator runs in this page. Aura does not send or store the text entered in its fields. Review the generated instruction before sharing it with any service.

Working method

Generated instruction

Editable fields remain on the left; the preview updates instantly.

Describe your objective to generate the instruction.

Aura processes these fields in your browser and does not send or store their content. Site-wide privacy controls still apply.

12 · Load template

Twelve free starting points

Load a template, replace the bracketed details and adapt the checks to your real stakes. Every template remains editable and provider-neutral.

Explain, then quiz

Teach a concept at the right level and check understanding.

Source-based synthesis

Combine supplied sources without hiding disagreements.

Compare documents

Map similarities, differences and consequences.

Decision matrix

Compare bounded options against explicit priorities.

Project plan

Turn an outcome into dependencies and verifiable milestones.

Evidence-aware rewrite

Revise text while preserving meaning and checking claims.

Debug a failure

Discriminate causes before proposing a fix.

Claim–evidence audit

Test whether a conclusion is supported by its evidence.

Reproducible data analysis

Define metrics and checks before interpreting a dataset.

Meeting to actions

Extract decisions, owners and unresolved points.

Bounded creative options

Explore distinct directions without endless ideation.

Adaptive learning plan

Build a measurable study loop with retrieval and feedback.

Methods without magic words

These methods organize work and evaluation. They do not guarantee truth: quality still depends on context, source quality, model capability and your review.

robust practice

Direct and specific

A clear goal, relevant context and explicit output constraints are the most portable starting point.

More text is not automatically better. Remove background that cannot change the answer.

robust practice

Decompose then assemble

Break a complex task into bounded deliverables and verify their interfaces before combining them.

Chain of Thought (CoT) often means a detailed reasoning trace. Aura asks instead for verifiable sub-tasks or a concise justification, never a transcript of private reasoning.

robust practice

Evidence and verification

Ask for claims, sources, assumptions, uncertainty and explicit checks rather than an authoritative tone.

Citations can be wrong or fabricated. Open important sources and verify them independently.

workflow

Bounded Kaizen loop

Improve one measurable gap at a time, compare with the prior version and stop at a stated threshold.

Without a metric and stop rule, repeated rewriting can add cost without adding quality.

research idea

Bounded alternative search

Generate a few distinct options, score them against a rubric and keep the strongest path. This adapts tree-search ideas into a practical workflow.

Tree of Thoughts is an orchestration approach studied in specific settings, not a universal phrase that makes every model reason better.

Use with judgment

  • Do not paste secrets, personal data or confidential documents into a third-party assistant.
  • Treat fluent output as a proposal, not proof. Verify important facts, calculations and citations.
  • Ask for assumptions, evidence, criteria and concise checks—not a detailed hidden reasoning transcript.
  • Keep exploration bounded: a small option set, a comparison rule and a stop condition.

For medical, legal, financial, safety or other high-impact decisions, use qualified professional review and authoritative current sources. A generated instruction does not transfer responsibility.

Official guidance and primary research

Aura’s recommendations are tied to public sources. Provider guidance can change with models, so check the current documentation when precision matters.

OpenAI · Prompt engineering best practicesOfficial ChatGPT guidance on clear, specific instructions, context and iterative refinement.Open sourceOpenAI · Reasoning best practicesOfficial guidance favoring direct prompts, clear constraints and observable goals over requests for internal reasoning.Open sourceAnthropic · Claude prompting best practicesOfficial guidance on clarity, examples, output format, sequencing and XML structure.Open sourceTree of Thoughts paperPrimary paper presenting search over multiple candidate paths with evaluation and backtracking.Open source