---
name: manage-quest-curriculum
description: Maintain an evidence-based Starlight Academy quest curriculum and facilitator plan. Use when adding or revising quests, mapping prerequisites to artifact outcomes, connecting specialist studios, planning a cohort, or deciding what learning evidence may enter a portfolio.
---

# Manage the quest curriculum

Use the existing curriculum owner and current source revision. This skill
prepares versioned content and facilitator records using available authorized
tools; it does not create an LMS, schedule, learner account or grading service.
Read `../run-learning-quest/references/quests.json` for the current local catalog
and `../run-learning-quest/references/practice-record.md` for progress semantics.

## Curate a path that produces useful work

Choose one learner outcome and a sequence of artifact dependencies. Start from
the learner's existing work; do not force completed material to be repeated.
Use evidence of prior work as placement context, not as automatic certification.
For each quest specify the decision, prerequisite artifacts, first attempt,
worked example, source contract, specialist method, allowed tools, acceptance
checks, transfer case and support owner. Count neither attendance nor model
verbosity as capability evidence.

Keep three lanes explicit:

| Lane | Evidence | Consequence |
| --- | --- | --- |
| Practice | Learner-owned attempt, assistance, critique and transfer | Suggested next practice |
| Facilitated review | Named human reviews exact artifact and rubric | Human feedback recorded with stated scope |
| Operational delivery | Tested release, destination read-back, rights and support | Product release under its own authority |

Certificates and account-backed progress remain separate future services until
authenticated identity, assessment, appeal, revocation and publication contracts
are implemented. Do not invent a credential to make the path feel complete.

## Compose specialists without duplicating their roles

Use Starlight Agent Designer for durable agent boundaries; Plugin Forge for
portable packages; Product Studios for one canonical deliverable; Living Design
for companion journeys and typed agent maps; Excellence for evidence-based
critique; Second Brain/Memory for selected-source continuity; Agentic Income for
accepted paid work. Resolve installed skill names before execution. Keep owned
income assets separate from services and DPI capital research; never label an
unproven business as passive or profitable.

Map sources by job: GitHub owns versioned curriculum/software; Drive holds
approved research and source masters; Figma owns editable design components;
Vercel serves approved web experiences; existing Cloudflare jobs provide only
their observed runtime behavior; Stripe records payment events through actual
verified integration. These are design roles, not claims of live connections.
Reuse the project's shared identity, artifact and commerce contracts. A domain,
subdomain or embed does not create a new source of truth or trust boundary.

Separate the learning experience from digital checkout. Keep purchases on the
owned website and preserve useful public learning without a purchase. Do not
use a ChatGPT quest as an indirect upsell or invent a paid prerequisite.

## Run a bounded maintenance cycle

1. Inspect current catalog, learner feedback voluntarily supplied for this
   purpose, source changes and unresolved defects. Do not search private learner
   data without a defined scope.
2. Propose the smallest content change and a held-out transfer test. Give each
   quest stable identity and edition; breaking artifact/rubric changes increment
   the edition and retain the previous record parser or explicit migration.
3. Prepare changes in the owning Git branch. Avoid storing learner records,
   private source text, keys or payment data in curriculum source.
4. Test links, packaged resources, example correctness, failed/missing-tool
   behavior, accessibility and the exact new source revision. Use an independent
   critic for a release-affecting change. Label unrun checks pending.
5. Publish only through the product's approved release flow. Record curriculum
   revision, deployment, plugin release and host smoke separately. A successful
   package save does not mean a website deployed or a learner completed a quest.

Return a compact curriculum change record: audience and outcome, changed quest
IDs/editions, artifact dependencies, teaching and delivery owners, source
revisions, actual checks, release states, missing integrations and next review.
Never create a recurring automation merely because curriculum should stay
current; use an actual scheduler only when the user asks for recurring work.