Instructional Design Models & Frameworks
ADDIE, ARCS, ASSURE, and UDL each fit different contexts. Reviews across the literature find no universal best; using two or three frameworks together tends to outperform strict adherence to one.
Evidence-based instructional design, transparently cited.
108 peer-reviewed studies across 11 research domains. Every recommendation tagged with its evidence tier — T1 meta-analyses and RCTs through T5 expert opinion. Runs in Claude Code.
git clone https://github.com/savvides/idstack.git && cd idstack && ./setup
Evidence
Every recommendation cites its source.
Stronger evidence wins when tiers conflict. All 108 references are auditable in evidence/references.md — open source, no proprietary studies, no cherry-picking.
T1 meta-analyses & RCTs
T2 quasi-experimental w/ controls
T3 systematic reviews
T4 observational
T5 expert opinion & theory
Constructive Alignment & Learning Objectives
Objectives, activities, and assessments need to line up; gaps in that sequence predict learning-outcome failures. The revised Bloom's taxonomy supplies the verb vocabulary so each level can be assessed against the level it claims to teach.
Needs Analysis
Three layers — organizational, task, learner — analyzed before any content gets written. When this step is skipped, courses tend to teach what the designer found interesting rather than what the audience came for.
Cognitive Load Theory & Instructional Sequencing
Working memory limits how much new information learners can hold and manipulate at once. Element interactivity, intrinsic vs extraneous load, and the expertise-reversal effect explain why scaffolding that helps a novice can slow an expert down.
Formative Assessment & Feedback
Feedback that explains what to do next produces larger learning gains than feedback that only marks right and wrong. Nicol's seven principles describe what "next" should look like in writing, peer review, and rubric design.
Multimedia Learning Principles
Mayer's principles (coherence, signaling, segmenting, modality, redundancy) predict when combinations of words and visuals help learning and when they interfere with it. Violations produce measurable working-memory load.
Learner Analysis & Differentiation
Expertise level changes which instructional moves help and which backfire. Scaffolding that supports a novice can interfere with an expert who has already automated the underlying skill.
Evaluation Models
Evaluation only measures what it was designed to measure, so the design has to happen alongside the course. Kirkpatrick's four levels (reaction, learning, behavior, results) map what's possible; most courses report only the first.
Rapid Prototyping & Design-Based Research
Short iteration cycles with real learners tend to produce better courses than long planning phases, particularly under time pressure. Design-based research formalizes that iteration so it stays methodologically defensible.
Online Course Quality Frameworks
Quality Matters' eight standards and the Community of Inquiry framework (teaching, social, cognitive presence) are the two rubrics with the most published validation studies behind them. idstack's quality audit runs against both.
Accessibility & Universal Design for Learning
WCAG 2.1 AA is what most jurisdictions require by law. UDL 3.0 (multiple means of representation, action, engagement) sets out how to design for the learner variation real cohorts already contain, rather than retrofitting it later.
Recommendations carry a domain code and tier — [Alignment-14] [T1] — so you can trace any claim back to the source paper. The references file is just Markdown: grep it, fork it, audit it.
Pipeline
Eleven skills. Pipeline optional.
Eight run end-to-end as the standard sequence, /idstack:pipeline orchestrates them, and two utilities (course-import and learn) sit outside the chain. Invoke any skill on its own — none of them require an existing manifest.
Sequence — runs left to right under /idstack:pipeline
- needs-analysis Three-level needs assessment; design-fit gate.
- learning-objectives Bloom's classification; bidirectional alignment.
- assessment-design Rubrics, checkpoints, Nicol's 7 feedback principles.
- course-builder Cognitive-load-aware content generation.
- course-quality-review QM 8 standards + Community of Inquiry.
- accessibility-review WCAG 2.1 AA + UDL 3.0 audit.
- red-team Adversarial audit across 5 dimensions.
- course-export Common Cartridge, SCORM 1.2, or Canvas API.
course-import
Branches in mid-pipeline when starting from an existing course.
learn
Cross-project memory utility — runs outside the pipeline.
Run the full sequence with /idstack:pipeline. The orchestrator auto-skips completed steps and resumes where you left off.
Output
Each skill writes both a report and a manifest.
Every finding-producing skill writes both: a branded HTML report you open in a browser, and a JSON manifest the next skill in the pipeline reads. All per-course artifacts land under one self-describing folder, .idstack/exports/<course-slug>/.
Finding obj-3: Three ILOs without assessment warning T1
What we saw. Of 8 stated objectives, 3 have no matching assessment item.
What the evidence says. Constructive alignment requires every objective to map to an activity and an assessment; orphaned objectives predict learning-outcome gaps. [Alignment-14] T1
Consider. Run /idstack:assessment-design to close the gap, or remove the orphaned objectives.
{
"learning_objectives": {
"findings": [
{
"id": "obj-3",
"severity": "warning",
"tier": "T1",
"domain": "Alignment-14",
"observation": "3 of 8 ILOs lack assessment",
"consider": "/idstack:assessment-design"
}
],
"report_path": ".idstack/exports/<course-slug>/learning-objectives.html",
"updated": "2026-05-13T14:32:18Z"
}
}The HTML report is what an instructional designer reads (or hands to a stakeholder — the folder is self-contained, CSS bundled, zip-friendly); the JSON is what /idstack:assessment-design reads next. Both files always reflect the most recent run; the timeline at .idstack/timeline.jsonl carries the run history. /idstack:pipeline additionally writes an index.html course dashboard at the folder root, linking to every per-skill report.
Install
Install in 30 seconds.
git clone https://github.com/savvides/idstack.git && cd idstack && ./setup
./setup registers idstack with Claude Code and installs all 11 skills. Re-run it after a git pull to update.
Requires a bash shell — Terminal on macOS or Linux, WSL or Git Bash on Windows. ./setup is a bash script, so PowerShell and cmd can't run it. python3 is recommended but not required; without it, quality-score trends and search filtering are unavailable.
If something looks wrong afterwards, bin/idstack-doctor diagnoses the plugin registration, and bin/idstack-status reports course health from inside a project directory.
New to Claude Code? Download it first (free), then paste the command above. When Claude Code starts, run /idstack:pipeline.
What's new
v3.4.0.0 — idstack is a Claude Code plugin.
Chrome Side Panel co-pilot & Course Dossier export.
For instructional designers working directly in the browser, idstack now includes a native Chrome Side Panel extension. Audit Canvas assignments, syllabi, Google Docs, and web course materials with 1-click Bloom's classification, constructive alignment reviews, and empirical citations (T1–T5) without touching the terminal.
Also in this release: an automated Canvas course crawler for whole-course audits, an active Course Dossier drawer that compiles multi-page findings into synthesized institutional Markdown (.md) reports, atomic file writes for project learnings logs (Issue #60), and local-over-global search precedence (Issue #61).
One host, supported properly.
idstack shipped an OpenAI Codex CLI build from v2.5.0.0 through v3.3.0.4. It is removed. Every skill, the evidence base, the project manifest, and the HTML reports are unchanged — idstack now targets Claude Code and nothing else, which is the only place it was ever tested end to end.
To update: cd into your idstack clone, then git pull && ./setup. If you had the Codex build installed, ./setup can no longer clean it up — the release notes carry the two paths to remove.
Patched in v3.4.0.1 (August 7): three evidence cards on this page overstated their domain's evidence tier. They are derived from evidence/references.md now, and the test suite fails if they ever disagree again. The same patch added the outbound Canvas API upload in /idstack:course-export and the git fetch in bin/idstack-update-check to the privacy policy, and named WSL or Git Bash in the Windows install instructions.
Welcome-back messages work on stock macOS now.
On any Python older than 3.12 — including the 3.9 that macOS ships — the session-memory code inside every skill hit a silent syntax error. No welcome-back summary, no quality-score trend, no suggested next step. Fixed, and now exercised on Python 3.9 in CI so it stays fixed.
Also in this release: /idstack:pipeline invokes its child skills correctly in Claude Code, skills notice previous runs again ("update the results or start fresh?"), imported courses get a proper next-step suggestion, and a course name with an apostrophe no longer blanks the status dashboard. To update: cd into your idstack clone, then git pull && ./setup.
Patched in v3.3.0.1 (August 5): skills no longer point you at commands that don't exist. Welcome-back and next-step messages now name the runnable /idstack:<skill> form.
Patched in v3.3.0.2 (August 6): searching your learnings for a term with an apostrophe — Bloom's, learner's — returned a wrong-typed result instead of the right one. Fixed.
Next: additional specialist skills (content sequencing, media selection, Kirkpatrick evaluation), and push-to-LMS so improvements land back in Canvas without copy-paste. See the roadmap.