learning-builder

Learning Builder

Own The Following Job

• Turn a vague learning request into one clear learner-specific tutorial brief.
• Gather the minimum learner context before doing deep research.
• Prefer authoritative sources over blogs, reposts, or anonymous summaries.
• Assemble one tutorial packet with explicit citations, exercises, and next steps.
• Export the tutorial to docx and pdf when requested.
• Offer an optional personalized webpage only after the tutorial content is approved.

Inputs

Expect one or more of:

• the topic or skill the learner wants to master
• the learner's current level, role, and target outcome
• a time budget, learning style, or deadline
• required or forbidden sources and domains
• existing notes, outlines, bookmarks, or source lists
• an explicit request for docx, pdf, or webpage output

Do Not Route Here

• requests that only need a short answer or explanation
• generic research with no tutorial or study guide deliverable
• unsourced opinion writing or marketing copy
• document export requests where the tutorial is already written and no learning workflow is needed
• pure webpage design work with no learning packet behind it

Default Workflow

1. Start with the sharp intake in reports/intent-dialogue.md. Ask only the questions that change scope, sourcing, or output shape.
2. Convert the answers into the schema in input/learner_profile_template.json. If key inputs are missing, list the blockers before researching.
3. Read references/authority-research.md and build an authority-first source list. Prefer primary sources and record why each source is trusted.
4. Read references/tutorial-assembly.md and draft one tutorial in markdown. Keep the tutorial aligned to the learner profile, not to the source order.
5. If docx or pdf output is requested, read references/export-pipeline.md and use scripts/export_tutorial.py. In this workspace, prefer markdown as the canonical source.
6. If the user wants a personalized webpage after reviewing the tutorial, read references/webpage-extension.md and generate that as a second deliverable, not as a substitute for the tutorial packet.
7. Validate the final artifact set:
   • the learner goal is explicit
   • key claims are backed by cited authority sources
   • exercises and next steps match the learner level
   • requested output files were actually produced
   • optional webpage work did not replace the core tutorial deliverable

Output Contract

The normal output set is:

• one learner-specific tutorial in markdown
• one compact source appendix with URLs, source dates, and trust rationale
• one docx export when requested
• one pdf export when requested
• one optional webpage plan or static learning page only after user confirmation

Validation Checklist

• The brief captures topic, learner state, target outcome, and constraints.
• The research set is authority-first and not padded with weak sources.
• The tutorial is personalized enough that a different learner profile would change the content.
• The export pipeline starts from markdown instead of maintaining separate source documents by hand.
• The webpage branch is clearly optional.

Reference Map

• Read reports/intent-dialogue.md before doing deep work.
• Read reports/reference-scan.md to understand the initial borrow plan.
• Read references/authority-research.md for source selection rules.
• Read references/tutorial-assembly.md for the tutorial content contract.
• Read references/export-pipeline.md for docx and pdf generation.
• Read references/webpage-extension.md before offering the personalized webpage branch.
• Inspect evals/trigger_cases.json when tightening route boundaries.