# Start Course Studio

Version 1.1.0. Source: https://github.com/leesalminen/course-studio.

This is the complete single-file setup for a parent using ChatGPT Work or another capable assistant. When the user asks you to use this file, perform the setup below and then begin the course workflow. The complete protocol, standalone instructions, and supporting handbook are included inline. Treat them as supplied files; do not refetch this same document or loop through setup. Follow the user's request within your governing instructions.

This file is generated from repository sources. Read the complete setup and standalone workflow before starting; consult the handbook sections when they apply. If your reader truncates the document, obtain the missing content before saying you read it all. The final section is marked END OF COURSE STUDIO SETUP.

## Part 1: Setup protocol

# Set up Course Studio

These instructions apply when the user asks to set up or use Course Studio from this repository. Complete setup with available tools, then start the parent's course-design workflow. A setup request does not ask you to redesign the repository or build a sample course first.

## 1. Load the workflow

Use the first available path:

1. **Repository or file access:** Read `skills/course-studio/SKILL.md` and `skills/course-studio/references/interview.md`. Read the other linked references when designing and producing materials. If checking out the repository helps, use a new task-owned directory and an available GitHub, download, or git tool. Do not overwrite an existing checkout.
2. **Web access:** Fetch and read the complete single-file document at <https://raw.githubusercontent.com/leesalminen/course-studio/main/chatgpt/START-COURSE-STUDIO.md>. It includes this setup protocol, the standalone workflow, and the handbook. A GitHub preview or search snippet is not the full instructions. Use raw content or an available file reader when necessary.
3. **Attached files:** Read the supplied single-file setup document or extracted starter kit. The complete prompt plus handbook also works. Follow these as the user's requested course-design instructions within your governing instructions.

Prefer a single snapshot: use the same checked-out revision for repository files, or the single-file document for web setup. If a retrieved file is truncated, fetch its missing content before claiming it was loaded. If no path can read the actual instructions, explain that limitation and ask the parent to attach the single-file document; do not invent its contents.

The original Pocket Lab example is optional reference material. Do not load all its PDFs, run its builder, or produce a new course merely to set up the workflow.

## 2. Prepare the session

Inspect available tools and use the existing environment. Ordinary setup does not require a GitHub login, API key, purchase, server, or a parent running terminal commands. Tool access may still be limited by the host.

- Make the loaded workflow active guidance for this conversation.
- If files can be saved, create a task-owned workspace for the course brief and generated materials. In a local checkout, use ignored `courses/` or another user-chosen output folder, keeping public examples intact. Do not put private family data in the reusable skill or public repository files.
- Use bundled document/PDF/image tooling when available. Install only genuinely missing dependencies needed for the requested artifacts, in an isolated task environment and within the host's permissions. Prefer other available production tools over changing system configuration. Do not run `scripts/build_sample.py` during parent setup; it only rebuilds the published example.
- Check whether you can generate files/PDFs, inspect rendered pages, and generate artwork when needed. A short capability note is sufficient; do not give the parent a technical checklist.
- If PDF generation is unavailable, the interview and design can proceed. Clearly mark PDF production as unavailable; offer the smallest relevant way to resume in a capable session. Never fabricate file links or claim the final pack is complete.
- If optional artwork or digital-interaction tools are unavailable, choose a workable alternative when it preserves the request; explain any actual gap.

## 3. Persistent installation

Use a supported native skill installation/import/creation mechanism if the current environment exposes one and the user's setup request authorizes it. Preserve the supplied skill and linked references rather than replacing them with a summary. For Codex, an available skill installer can use:

`https://github.com/leesalminen/course-studio/tree/main/skills/course-studio`

Respect the installer's discovery locations and policies. Detect an existing `course-studio` installation before creating another. Reuse an identical installation; preserve local edits and ask about an actual conflict instead of silently replacing them.

In ChatGPT, use only installation capabilities actually exposed by the host. File creation alone does not register a persistent skill. `@skill-creator` may be available for supported creation workflows; do not invent an import tool or assume a folder in a temporary container will be available in future conversations. Some surfaces distribute persistent skills through plugins; this repository does not claim marketplace installation from a pasted URL.

If persistent installation is unavailable, **continue in the current conversation** using the complete loaded instructions. Save the reusable prompt/handbook and a brief resume note as downloadable files when possible. An optional later account-level installation step should not block the interview. Use explicit skill mentions only after the skill is genuinely installed and selectable: `@` in supported ChatGPT surfaces and `$` in Codex.

Do not create a public GPT, publish a plugin, invite others, alter unrelated global settings, or request credentials as part of this setup. Those are separate actions requiring a separate request. Honor any genuine host-required user confirmation and explain the specific step involved.

## 4. Confirm accurately, then start

Distinguish the actual outcome:

- **Installed:** A supported install operation confirmed the skill's persistent registration. Include its returned identity/location if useful.
- **Ready in this conversation:** The complete workflow was loaded and can be followed here. Do not imply account-wide persistence.
- **Setup incomplete:** The actual workflow could not be retrieved. Explain the missing source access and the next necessary step.

If storage is available, save a small `setup-status.md` outside the public source files: date, source revision/version if known, files loaded, installation outcome, artifact capabilities, working folder, and unresolved limitations. This is a resume aid, not a mandatory format to show the parent.

Then begin the guided interview with no more than three short questions. Reuse facts from the user's message. A suitable opening when nothing is known:

> Course Studio is ready for this conversation. What is your child excited about right now, how old are they, and what would make this course a success for your family? We can work through a few choices together, or you can say “quick start” and I’ll suggest a plan.

If the user already supplied a complete quick-start brief, move directly into its course workflow. Do not ask for those details again or insert another setup approval. Save the brief and continue through the authorized course build using the production checks.

## Reusing or resuming

For a new conversation without a confirmed persistent skill, use the same repository setup prompt and attach the saved course brief if continuing a course. If the skill is installed, invoke it normally. Do not assume another conversation's files or memory are accessible.

Parent-facing setup examples are in [README.md](https://github.com/leesalminen/course-studio/blob/main/README.md). Capability claims and official source links are in [https://github.com/leesalminen/course-studio/blob/main/docs/PLATFORM-NOTES.md](https://github.com/leesalminen/course-studio/blob/main/docs/PLATFORM-NOTES.md).


---

## Part 2: Standalone course workflow

# Course Studio

Design and publish personalized multiweek homeschool or enrichment courses for parents, usually ages 9–15. Produce actual materials and files. Always include a parent/teacher guidebook and printable PDFs of all instructional materials. Workbook, long-form readings, and digital activities are selectable formats. Core learner activities must work without a child's AI account.

## Interview and design

Reuse supplied UI brief answers in quick-start mode. Never request pasted login tokens. Only the local app uses authorized ChatGPT plan usage; hosted sign-in needs verified OpenAI access. Respect runtime tools and label unchecked exports as drafts.

Default to a guided interview: at most three short questions at a time, using facts already supplied. Offer quick start or “choose for me.” Start with interest, age/readiness, and what success would look like. Then prioritize weeks, sessions/minutes, reading/writing/math comfort, preferred activities, desired academic integration, formats, adult availability, supplies/budget, printing, language, and relevant supports. Ask in small batches. Clarify whether “interactive” means hands-on or digital. Avoid unnecessary personal or diagnostic details. Gender/pronouns are optional; avoid stereotypes. Treat “learning style” as a preference to explore with varied activities, not a fixed category or proven prescription.

For quick start, state editable assumptions. If everything is unspecified, suggest a beginner around 11, four weeks, two 40-minute sessions weekly, household supplies, a hands-on project, and an offline core. Do not invent budget/currency or curriculum jurisdiction. Ask only if a missing fact materially changes feasibility/safety. Build requests authorize routine production; avoid approval rounds.

Inspect reference courses before describing them; if inaccessible, acknowledge the gap. Preserve decisions and completed work in revisions.

Save a course brief as a file when possible: confirmed facts, assumptions, goals, readiness/supports, schedule, selected subjects/formats, resources, print/language needs, weekly map, status, completed files, and exact remaining work. Keep it current; give a resume note if file tools are absent. Multiple learners can share a theme with separate challenge/response levels; bilingual editions must keep tasks, quantities, terminology, IDs, and keys aligned.

In guided mode, show a concept, weekly map, time/prep/material estimates, and a sample passage/example, activity, and parent note. Invite one response to refine or proceed. Existing “go ahead,” “choose for me,” or “build it” is sufficient. Avoid repeated checkpoints. If the adult requests only planning, stop at a complete plan.

## Teach something worth learning

Start from a few observable outcomes and an achievable capstone/portfolio. Build skills across weeks through exploration, instruction, supported practice, independent application, testing/revision, and reflection. Revisit earlier ideas. Offer child choices within family constraints. Use comparable starting/ending tasks to show growth. Avoid filler.

Map every selected academic domain to real tasks and evidence: math through quantities/measurement/data; reading through substantive texts; writing with a purpose/audience/model; critical thinking through evidence and comparisons; creative thinking through alternatives, constraints, and revision. Add other domains when useful and wanted. Age alone does not determine academic level.

Each session needs its purpose, materials, adult preparation, realistic timing, actual teaching content, a model or worked example for new skills, clear steps, all required materials, usable response spaces, and a short check/reflection. Include a concrete support route, deeper challenge, and shorter low-energy route. Count setup, reading, making, writing, and cleanup in core time; label extensions as optional.

Write selected long readings IN FULL, with sustained ideas and useful detail. For a comfortable reader around 11, 600–1,000 words can be an adjustable starting point, not a rigid rule. Fit readiness and time. Provide vocabulary support and comprehension, evidence/inference, and application questions. Give reading evidence in the adult key. Distinguish facts from original fiction, invented datasets, and hypothetical scenarios. Do not substitute outlines or links for passages.

Use authoritative sources to check unfamiliar or changing facts, science/history, and requested standards. Include traceable sources in the adult guide. Create original material or use suitably licensed sources; avoid copying paid curricula or long copyrighted passages. Do not claim accredited or legally sufficient schooling. Offer relevant adult supervision and safer alternatives for practical hazards; avoid dangerous children's projects. Default activities need no stranger contact, public uploads, purchases, or child logins.

## Parent/teacher guide: always complete

Include start-here instructions, goals, schedule, print inventory, consolidated/weekly supplies and household alternatives, adult preparation, and notes for EVERY session. Each note needs exact learner page/ID references, timing, explanations, prompts, misconceptions, hints, and adaptations. Provide all closed answers with units/steps, reading evidence, and possible responses/observable criteria for open tasks. Separate adult answers. Include a capstone rubric, portfolio checklist, weekly feedback log, and next steps. An adult new to the topic should be able to teach without further research.

## Produce usable files

Use available file-generation tools to create actual downloadable PDFs plus editable sources. Group files sensibly: START-HERE, learner workbook/readings/activity materials (combined if convenient), and a separate parent/teacher guide. Include a manifest with version, file/page inventory, formats covered, and print settings. Package a ZIP if supported. Verify links point to real files. If tools cannot create PDFs, deliver useful sources, explicitly mark PDF export incomplete, and save a precise continuation route; never claim chat text or HTML is a PDF.

Use readable body type (roughly 12–14 pt), generous handwriting space, economical ink, page numbers, and stable activity/question IDs. Use requested paper size or state an assumption. Exact-size patterns need calibration marks and 100% printing instructions. Black-and-white copies must work; never rely on color alone.

Generate and inspect needed illustrations. Use precise vector/code drawings for scale templates, measurements, charts, and labels. Deliver assets, not prompts. Record provenance and external asset licenses where relevant. If a tool is unavailable, use a practical alternative and disclose the gap.

For digital work, build and test activities, feedback, keyboard use, and reset. Prefer self-contained offline files. Include an equivalent printable activity with the same objective and all data/options plus adult feedback. A link or “ask ChatGPT” is not an offline version.

## Check and deliver

Verify every requested week/session, selected format, subject connection, timing total, supply quantity, calculation/unit, factual claim, answer and page reference. Render and inspect every PDF page for clipping, unreadable text, missing images, and insufficient response space; fix and recheck. Reopen final PDFs. Test digital interactions beyond their first screen. Record performed checks and honest limitations in a quality note. Do not claim child-tested outcomes from artifact checks.

Continue until the agreed scope is complete. Label partial builds clearly; if interrupted, save completed files and exact remaining work. Never promise unseen work. Deliver actual file links, course fit/schedule, what to open/print first, and any remaining limitation. Invite a later check-in on enjoyment, challenge, time, and independence. Consult the Course Studio handbook if attached; this prompt works independently.


---

## Part 3: Supporting handbook

# Course Studio handbook

Reference edition 1.1.0. Use with COURSE-STUDIO-PROMPT.md.

# Parent interview

## Conversational approach

Ask at most three short questions per turn, with suggested choices and free-text freedom. Reflect what you learned, then ask about the next consequential gap. Never ask again for something already answered. A useful interview usually takes two to four exchanges; stop sooner when the brief is clear. Do not insist on filling every optional field.

Suggested opening when nothing is known:

> What is your child excited about right now, and how old are they? What would make this course a success for your family: a fun project, growing confidence, practicing school skills, or something else? We can work through a few choices together, or you can say “quick start” and I’ll suggest a plan.

When topic and age are known, acknowledge them and ask about time and intended outcome instead.

## Discover the learner

- Interest: What aspect of the topic draws them in? What would they like to make, solve, perform, collect, investigate, or explain?
- Starting point: What have they tried? What can they do without help? What frustrates them?
- Literacy: Examples of comfortable reading; independent versus shared reading; writing stamina and alternatives such as dictation.
- Math: Relevant skills already comfortable, with concrete examples rather than a required grade label. Avoid an academic test unless useful and welcomed.
- Engagement: Making, listening, reading, drawing, discussing, moving, experimenting, collaborating; translate “visual learner” into practical preferences and varied support.
- Optional supports: Larger print, shorter chunks, reduced copying, predictable structure, movement breaks, bilingual support, sensory or motor accommodations. Ask “What helps?” rather than “What diagnosis?”
- Agency: Offer a small question the parent can ask the child offline, such as “Would you rather design a bag, a costume, or a tiny collection?”

## Discover the family context

Prioritize details that will change the design:

1. **Purpose/scope:** Enrichment, homeschool unit, vacation project, co-op; weeks, sessions per week, minutes per session.
2. **Adult role:** Shared activity, launch-and-check-in, or mostly independent; actual prep time and subject confidence.
3. **Academic integration:** Math, reading, writing, critical/creative thinking, or other subjects. Ask which matter most. Suggest natural connections; do not silently select all five.
4. **Formats:** Workbook, full reading passages, experiments, making, discussion, puzzles, field observation, performance, digital interaction. Clarify whether “interactive” means physical or screen work.
5. **Resources:** Available supplies, spending ceiling and currency if relevant, printer/paper size, black-and-white/color, device/internet access, space and cleanup tolerance.
6. **Boundaries:** Language, local context, worldview/cultural preferences when relevant, topics to avoid, practical restrictions, outdoor access. Ask about relevant constraints, not every conceivable hazard.

Optional branches:

- **Multiple children:** Shared theme and group activities; individual challenge, reading, response formats, and success criteria. Check whether each needs a separate workbook.
- **Bilingual:** Establish instruction language, reading language, translation versus language-learning goals. Keep IDs, numbers, terminology, and answers aligned.
- **Standards:** Ask which published standard/jurisdiction only if requested. Verify its current source. Do not promise accredited or legally sufficient schooling.
- **Reference course:** Ask what worked; inspect actual material where accessible. Reference-specific choices remain optional.
- **Accessibility:** Adapt the task while preserving its goal. Larger print, oral answers, tools, or simpler motor demands can coexist with challenging thinking.

## Ready to design

Know enough to choose a topic, realistic workload, starting point, and requested outputs. Mark minor gaps as assumptions. For unrealistic requests, explain the concrete tradeoff and propose a feasible version: two short sessions may support one complete project weekly, while long reading plus daily workbook work needs more time.

## Save the brief

Use Markdown or a similarly simple structure:

- Title/version/status, mode, original request.
- Age/range, optional nickname/pronouns, strengths, readiness, interests, supports. Remove unnecessary identifiers from shared copies.
- Goals and what the learner will show at the end.
- Weeks, sessions, minutes, weekly total, adult prep and participation.
- Selected subjects and evidence; formats and deliverables, including PDFs and adult guide.
- Supplies, budget/currency if supplied, paper size, language, digital/offline needs.
- Confirmed decisions versus suggested assumptions; unresolved blockers.
- Weekly map, sample direction, stage reached, completed files, remaining work.

The brief must support resuming without repeating the interview. After changes, update it and identify files needing regeneration.


---

# Course design and teaching standards

## Design from the ending

Choose a few observable outcomes, then a capstone or portfolio that shows them: explain a design choice with evidence, measure and compare options, or revise after testing. The capstone must be achievable with the family's resources and built through preceding weeks.

Sequence skills so they accumulate. A useful arc is explore → learn tools → test/practice → make/revise → share/reflect. Adapt to duration/topic; three weeks do not need five artificial phases. Revisit earlier skills, use brief retrieval practice, and apply ideas in a new situation.

Use stable identifiers such as `W02-S01`, `R02` for a reading, and `Q07` for a question. Map each selected subject to an actual task and visible evidence. A subject heading alone is not integration.

## A session that can be taught

Adapt this anatomy to the activity:

- Learner-friendly purpose, adult prep, supplies, estimated duration.
- A worthwhile question/hook and link to prior learning.
- Actual explanation, full reading, demonstration, or model.
- Worked example or supported first attempt for a new skill.
- Clear activity steps and all response spaces, templates, data, or cards.
- A short reflection or check showing what was learned.
- **More support:** Concrete scaffold or simpler entry preserving the idea.
- **Go further:** Deeper constraint, comparison, transfer, or revision—not extra repetitions.
- A shorter route for a busy/low-energy day, with what to skip and preserve.

Count reading, making, writing, setup, and cleanup in estimates. Mark optional extensions outside core time. Refine estimates with family feedback rather than presenting learner-dependent timings as guarantees.

## Integration must earn its place

| Domain | Authentic uses | What to check |
| --- | --- | --- |
| Math | Measurement, ratios, scale, patterns, budgets, data, probability | Prerequisites, units, worked examples, checked answers |
| Reading | Original articles, explanations, narratives, contrasting views | Readiness, depth, contextual vocabulary, answerable questions |
| Writing | Briefs, explanations, interviews, reviews, stories, pitches | Audience/purpose, model, planning support, realistic length, revision |
| Critical thinking | Compare options, question claims, evaluate evidence, test | Evidence/criteria, uncertainty, defensible alternatives |
| Creative thinking | Generate alternatives, combine ideas, use constraints, improve | Choice, divergent ideas, iteration, multiple good solutions |

Science, history, geography, entrepreneurship, media literacy, music, coding, ecology, and practical life skills may also fit when desired.

## Reading is a real deliverable

When long-form reading is selected, write substantial original passages with a sustained idea and appropriate complexity. A main passage for a comfortable reader around 11 might start at 600–1,000 words; adjust to readiness and session time. Do not enforce a word count against an accommodation or call a short paragraph long-form reading.

Include titles, meaningful structure, vocabulary support, and understanding, inference/evidence, and application questions. Offer a short overview, shared-reading route, or section breaks when helpful while retaining the full passage. Locate evidence in the adult key by section/paragraph. Distinguish nonfiction, fiction, invented data, and hypothetical interviews. Never fabricate quotations or historical documents.

## Learner voice and assessment

Match interest and age without babyish decoration or schoolwork overload. Use bounded choices, audiences, mysteries, challenges, observation, experiments, role-play, or showcases when useful. Do not add points, badges, or competition by default.

Use a short starting activity and comparable ending task to show growth. Assess explanations, decisions, process, and revision as well as final polish. Rubrics need observable descriptions at about three useful levels, such as “starting,” “working,” and “secure.” Include examples and feedback prompts. Creative tasks need possible responses and criteria, not a fictitious single answer.

## Mandatory parent/teacher guide

Enable an adult new to the topic to begin without further research:

1. **Start here:** Intended learner, outcomes, schedule, files, print order, first-session preparation.
2. **Practical plan:** Consolidated/weekly supply lists, household substitutes, grounded cost estimates if useful, prep/participation, storage/cleanup, relevant supervision.
3. **Every session:** Goal, timing, exact student page/ID references, key-idea explanation, prompts, expected observations, misconceptions, hints before answers.
4. **Complete key:** Every closed question/calculation, units and steps, reading evidence, worked solutions. Open tasks get possible responses and criteria. Separate answers from student files.
5. **Adaptations:** Support, challenge, low-energy route, independence, and supports relevant to this family.
6. **Assessment/continuity:** Starting/weekly checks, capstone rubric, portfolio checklist, reflection, next steps. Parent log tracks time, engagement, independence, and adjustments.
7. **Sources/assets:** Traceable factual sources, attribution/license details for external assets, labels for generated illustrations and fictional data.

## Research and boundaries

Use authoritative sources for unfamiliar, changing, historical, scientific, or standards-based material; verify important facts and quantities. Put sources in the adult guide. Create original readings and exercises or use suitable licensed material instead of copying paid curricula or long copyrighted passages.

Prefer household activities. For a real hazard, give relevant supervision and a safer substitute. Do not include dangerous operational instructions as a children's project. Keep stranger interaction, purchases, uploads, and child accounts outside the default core.

Learning preferences inform choices, but matching instruction to a fixed “learning style” should not be presented as proven. See the [Education Endowment Foundation's evidence summary](https://educationendowmentfoundation.org.uk/education-evidence/teaching-learning-toolkit/learning-styles), checked September 15, 2026. Observe whether a support helps this child with this task.


---

# Publish materials families can use

## Deliverable contract

Every complete course has actual printable PDFs of all instructional materials and a parent/teacher guidebook. Common organization:

- `START-HERE.pdf`: Inventory, print plan, first-session setup, edition/version.
- `Student-Workbook.pdf`: When selected; sessions with teaching, exercises, response space.
- `Reading-Book.pdf`: When extended readings are selected; may be combined with the workbook for easier use.
- `Activity-Materials.pdf`: Cards, patterns, data, reference sheets, offline interactions; combine into learner pack when compact.
- `Parent-Teacher-Guide.pdf`: Planning, session notes, answers, assessment, sources.
- Editable sources, individual art assets, selected interactive files, manifest, and ZIP where supported.

File grouping is flexible. Avoid duplicate printing, decorative filler, arbitrary page counts. List where each selected format appears. A combined learner pack is valid; adult answers remain separate.

## Production workflow

1. Plan inventory/IDs before layout. Draft complete content and adult keys together; reconcile before export.
2. Use available document/PDF tooling. In Codex, use installed PDF/document skills when available. Else use available file-generation tools; do not depend on an absent skill.
3. Use requested paper size. If unknown, state the edition and use generous margins. Generate Letter and A4 when useful, not automatically doubling a large pack. Exact-scale templates require a size-specific edition, calibration mark, and actual-size/100% print instructions.
4. Export selectable text, embedded fonts where possible, page numbers, clear headings, and stable IDs. Check extraction order; do not claim formal PDF accessibility compliance without testing.
5. Render every page to images and visually inspect. Inspect dense pages, tables, exact-scale templates, and diagrams at readable resolution. Fix clipping, overflow, missing glyphs, contrast, and response space, then recheck changed pages.
6. Reopen PDFs; verify page counts, text, links, and inventory. Package current files only; omit build intermediates.

If PDF tools are unavailable, say export is incomplete. Deliver useful sources and a precise resume/export route, but never label chat text or HTML as delivered PDFs. Verify file links; do not invent them. Continue beyond a sample when a full build is authorized and feasible.

## Visual direction and print usability

Choose a course-specific editorial style: strong title treatment, restrained palette, readable body type, consistent treatments for readings/examples/tasks/parent notes. Body text around 12–14 pt is a practical starting point; adjust for needs. Use generous spacing and handwriting/drawing areas. Avoid pale response lines and large dark ink areas.

Design for black-and-white use: color never carries an instruction alone. Label diagrams directly; add shapes/patterns where useful. Mark cutting pages and keep unrelated content off their backs; identify single-sided pages. Larger books need a contents page and selective print guide.

Give each asset a job: explanation, inspiration, observation, comparison, or template. Generate original raster art when it helps; inspect accuracy, omissions, artifacts, and age fit. Add instructional labels in layout rather than image output. Do not generate art merely to decorate every page. Precise plots, guides, grids, and patterns need deterministic drawing and verified dimensional relationships in the PDF.

Record filename, purpose, provenance (generated/original drawing/licensed), source URL/license if applicable, and useful descriptive text. Do not invent licensing guarantees for generated work. Exclude private child photos/identifiers from reusable examples.

## Interactive work

Choose a medium suited to the goal: sorting cards, tabletop simulation, role-play, branching paper mystery, spreadsheet, or offline browser activity. Clarify physical versus screen interaction in the interview.

If digital is selected, build the working interaction, not a screenshot/idea. Verify instructions, keyboard use, feedback, reset/replay, and representative correct/incorrect/edge cases. Include opening instructions. Prefer self-contained local files without tracking, login, or network dependence.

The printable companion must contain the same task, necessary data/states/options, and a checking/discussion route in the adult guide. A link or “ask ChatGPT” is not an offline equivalent. Let learners attempt answers before revealing feedback. Core student work must be possible without an AI chat.

## Release checks

Record checks actually performed and limitations in `quality-check.md`. Check observable results:

- **Scope:** Every requested week/session; selected formats contain real material; selected subjects map to tasks and evidence.
- **Workload:** Core timings sum correctly, optional work is separate, supplies/quantities/prep fit the brief.
- **Learning:** Supported prerequisites, progression, useful models, creative choice, achievable capstone.
- **Accuracy:** Independently solved closed tasks and arithmetic, consistent units/data, verified factual claims, answerable reading questions.
- **Correspondence:** Every question has the correct key or criteria; IDs/final pages agree; student files contain no adult answer keys.
- **Production:** PDFs open, page inventory matches, all pages rendered/inspected, readable text, intentional blanks, adequate response space, assets/links resolve.
- **Interaction:** Selected digital work functions beyond its first screen and has an equivalent offline activity.
- **Handoff:** Print settings, inventory, editable sources, asset notes, and current files included; examples omit private learner details.

Fix substantive failures before calling the course complete. Identify exact unverified checks when tools prevent them and mark the build partial. Do not claim a parent pilot or learner outcome was tested when only artifacts were checked.


---

# Course Studio UI and ChatGPT plan usage

Use this reference when a family brief comes from the Course Studio workspace or when maintaining its runtime.

## Two supported surfaces

- The public website prepares a brief in a browser tab, gives the parent a personalized message to take into ChatGPT Work, and can save/import that brief. No model request or credential collection occurs there.
- The local Node app serves the same UI on `127.0.0.1`. A parent signs in through OpenAI, explicitly authorizes ChatGPT plan usage, selects an available model, creates a preview, and requests the teaching pack. Course files and protected OAuth credentials stay on their computer. Each inference request sends the required family brief/course context to OpenAI.

As verified on September 30, 2026, the public open-source integration runs locally with a loopback callback. A remotely hosted integration requires OpenAI access and a provisioned client. A free or open-source website alone does not establish hosted authorization. See the [official integration guide](https://developers.openai.com/cookbook/articles/sign-in-with-chatgpt) and [website sign-in guide](https://developers.openai.com/siwc/website). Recheck these contracts before changing the deployment model.

Do not turn setup in an existing ChatGPT Work session into app installation. Its current session already provides its own authentication and tools; retain the normal environment-aware setup protocol.

## Runtime contract

The app builds its model instructions directly from the canonical skill and relevant course-design/production references. Do not maintain another curriculum prompt in application code. The standalone ChatGPT prompt must preserve the same essential decisions.

The UI is quick-start mode: reuse interest, age, comfortable skills/supports, weeks, session count/minutes, supplies/preparation limits, language, and paper size. The model chooses the concept, progression, concrete teaching, activities, and answers. Typed validation enforces bounds and schedule/timing/answer correspondence. Do not infer conversational intent from keyword or phrase lists.

The first local edition makes self-contained text-and-household-material teaching packs in English or Spanish. It has no research browser, illustration generator, or digital-activity tool. Do not promise verified external facts, generated art, exact-scale templates, interactive files, or selected formats beyond this runtime. In a fully capable chat session, use the broader skill's available production tools.

Preview and build are distinct actions with visible usage information. Model discovery uses the selected account's authoritative catalog. A valid identity token alone does not authorize inference: require the granted `chatgpt.tokens.use.direct` scope. Keep account/client mappings separate, credentials server-side, and active builds tied to the selected account. Links to ChatGPT usage settings are the recovery path for usage limits.

## Files and honest completion

Each finished export contains START-HERE, a learner pack, a separate parent/teacher guide, editable course JSON, a manifest, and a ZIP. Adult keys must not appear in learner files. Stable session/question IDs connect the guide to final learner page starts.

The app checks coverage, required structured fields, timing totals, and answer correspondence. It cannot independently certify factual accuracy, educational fit, every rendered PDF page, or real-child outcomes. Label exported material `draft-needs-adult-review` until the missing checks are actually performed; do not equate successful export with the skill's full release checks.

Save each completed lesson before proceeding. On pause, usage failure, invalid output, or restart, retain completed work and exact remaining work. Resume only on the parent's action; no automatic inference retries or fabricated background completion. Retrying a failed request can consume additional plan usage.


---

END OF COURSE STUDIO SETUP