Authoring SCORM Courses — a Claude Agent Skill

A Claude Agent Skill that teaches an AI client how to author high-quality, interactive,
SCORM-compliant e-learning courses with the edumints SCORM MCP server.

🌐 Languages: English · Türkçe · Español · Русский · 简体中文 · Azərbaycanca · Қазақша · Кыргызча

Open-source, developed by the edumints.com platform. Open to contribution.

What this is

A Claude Agent Skill is a folder of
instructions that an AI client loads on demand. This skill gives the model the instructional-design
judgment and the exact tool recipes to turn a request like "make a 6-minute course on phishing"
into a polished SCORM package: clear objectives, varied screen types, aligned assessment, on-brand
theming, a slide-stage player with timed reveal, media, and the build → preview → feedback loop.

The skill is the author's playbook; the scorm-mcp server is the assembler.

Structure (progressive disclosure)

authoring-scorm-courses/
├── SKILL.md                         # entry point: workflow + quality bar
├── references/
│   ├── anti-slop.md                 # anti-slop discipline: training read + parametric dials (read FIRST)
│   ├── pre-flight.md                # MANDATORY pre-build quality-gate matrix
│   ├── core/                        # Layer 1 — method-independent core rules (+ the Layer-0 selector)
│   │   ├── method-selector.md       # Layer 0 — outcome type + dials → method pack(s) + overlay(s)
│   │   ├── evidence-binding.md      # every scored question binds to an in-course evidence source (K1–K6)
│   │   ├── alignment.md             # objective→question→evidence mapping + warn threshold (H1–H3)
│   │   ├── feedback-anatomy.md      # 3 mandatory feedback elements — floor rule (G1–G3)
│   │   └── scoring-timing.md        # formative/summative + "no score before evidence" (Z1–Z3)
│   ├── eval/
│   │   └── blind-test.md            # blind-test protocol (≥ 1/2 pass threshold) — gate for new screen types
│   ├── pedagogy/                    # Layer 3 — method packs (C series)
│   │   ├── _SCHEMA.md               # pack front-matter contract (evidence_phase(s) REQUIRED) + validation command
│   │   ├── _STUB-dogrusal.md        # schema-validation example: linear flow (not a pack)
│   │   ├── _STUB-dongulu.md         # schema-validation example: cyclic flow (not a pack)
│   │   ├── rosenshine-di.md         # C1 — Direct Instruction: model-first, guided→independent practice
│   │   ├── merrill-fpi.md           # C2 — Merrill First Principles: task-centered activation→demonstration→application→integration
│   │   ├── 5e-inquiry.md            # C3 — BSCS 5E inquiry cycle: explore-first (requires the exploration screen type)
│   │   ├── 4cid.md                  # C4 — 4C/ID complex-skill training: whole tasks, simple→complex, fading support
│   │   ├── mastery-learning.md      # C5 — Bloom mastery learning: unit → formative threshold → correctives loop → summative
│   │   ├── productive-failure.md    # C6 — Kapur productive failure: unscored struggle → consolidation (requires exploration; PK floor 4)
│   │   ├── pbl-case.md              # C7 — Barrows case/problem-based learning: case file = evidence artifact family (high PK)
│   │   ├── kolb-experiential.md     # C8 — Kolb experiential cycle: concrete experience → reflection → concepts → active experimentation (attitudes)
│   │   ├── sim-drill.md             # C9 — simulation drill: model run → unscored try-mode → debrief → part-task loop → scored scenario
│   │   ├── gagne-9.md               # C10 — Gagné's nine events (compliance/mandatory training; documented fallback default)
│   │   ├── cognitive-apprenticeship.md  # C11 — Collins/Brown/Newman: expert think-aloud model → coaching → fading → articulation → reflection → exploration
│   │   └── retrieval-spaced.md      # C12 — retrieval practice + spacing (refresher-only; evidence = the re-exposure reference artifact)
│   ├── overlays/                    # Layer 2 — method-orthogonal overlays (D series)
│   │   ├── _FRAMEWORK.md            # overlay file format + pack-independence rule + conflict format
│   │   ├── cognitive-load.md        # D1 — cognitive-load management: segmenting/pre-training/modality/coherence/redundancy/signaling → screen decisions
│   │   ├── udl.md                   # D2 — UDL (CAST 3.0): multiply representations of the SAME evidence source; response-format options; honest audio/caption limits
│   │   ├── arcs.md                  # D3 — Keller ARCS: attention/relevance/confidence/satisfaction as STRUCTURAL decisions (not tone); no decorative gamification
│   │   ├── expertise-adaptive.md    # D4 — Kalyuga expertise reversal: PK → support dose (worked_example fading), expert paths via visible_if; "skippable = SUPPORT, never EVIDENCE"
│   │   ├── assessment-alignment.md  # D5 — Bloom-revised/SOLO level ↔ question-type mapping; "a recall question cannot measure an apply objective"; score-weight distribution
│   │   └── accessibility.md         # D6 — WCAG 2.2 AA authoring-time decisions on top of the platform's honest conformance statement (alt-text quality, keyboard-safe types, learner-controlled timers)
│   ├── migration-v1-to-v2.md        # v1→v2 migration guide: breaking changes + recipes + Pattern A→rosenshine-di mapping + the 3-demo playbook
│   ├── source-expansion.md          # compressed-source expansion: cheat-sheet line → mechanism question → artifact → bound question (2 worked conversions)
│   ├── visual-storytelling.md       # narrative thread + per-screen visual budget + mockup-SVG recipes
│   ├── authoring-recommendations.md # when/how/why decision guide
│   ├── mcp-cookbook.md              # exact tool calls + full build_from_spec shape
│   ├── course-patterns.md           # proven course structures
│   ├── instructional-design.md      # objectives, microlearning, anti-template-fatigue
│   ├── screen-types.md              # decision guide for all screen types
│   ├── assessment.md                # question/feedback/scoring design
│   ├── interactivity-and-gamification.md
│   ├── media.md                     # cross-MCP media + built-in Turkish TTS + local helper
│   ├── video-generation.md          # programmatic motion-graphic / data-viz video
│   ├── artifact-to-scorm.md         # escape hatch: arbitrary interactive HTML → tracked SCORM (embed_html / wrap_artifact + postMessage bridge)
│   └── themes.md
├── templates/                       # copy-and-adapt blueprints
└── examples/                        # flagship multi-pack example (evidence-bound, blind-test passed) + deprecated v1 pilot

Requirements

The edumints SCORM MCP server, reachable by your
AI client (self-host it, or point at your own deployment).
An MCP client that supports Agent Skills (e.g., Claude).

Install

claude.ai (Skills): zip the authoring-scorm-courses/ folder and upload it under
Settings → Capabilities → Skills → Create skill.

cd authoring-scorm-courses && zip -r ../authoring-scorm-courses.zip . && cd ..

Claude Code / local: copy the authoring-scorm-courses/ folder into your skills directory
(e.g. ~/.claude/skills/authoring-scorm-courses/).

Then connect the scorm-mcp server and ask the model to build a course — it will follow this skill.

License

MIT — see LICENSE. Developed by edumints.com. Product names referenced are
trademarks of their respective owners (nominative use only).

edumints-scorm-skill