An AI copilot for instructional design.

idstack helps you do each job of an instructional designer, in ADDIE, backward design or SAM. Each recommendation cites its study and shows how strong the evidence is.

Add to Chrome Free or in Claude Code:
git clone https://github.com/savvides/idstack.git && cd idstack && ./setup
No terminal? Use the Claude desktop app →  See the studies →  Find your ADDIE or backward design phase

Open source MIT v3.11.1.0 · Claude Code & Chrome

Claude Code · sample session

you/idstack:course-import

idstackHow do you want to import your course? LMS or SCORM package file · PDF or Word file · Paste course documents · Canvas API

youCanvas API. Here is my Canvas URL and course ID.

idstackPaste your access token into .idstack/canvas.env, not into the chat. Then tell me when it is done.

idstackConnected. Found 12 modules, 8 assignments, 4 discussions, 6 outcomes. Import quality triage: 4 flags (4 warning). The most important: all 8 assessments are auto-graded quizzes with no rubrics. Read the draft in .idstack/exports/bio-101/course-import.md.

In the report:

warning T1

All 8 assessments are auto-graded quizzes with no rubrics

JustificationElaborated feedback produces larger learning gains than correctness-only feedback. [Assessment-8] [T1]

+ 3 more flags in the report, for example: only 4 of 12 modules state learning objectives.

AI help for each instructional design job

You are the designer, and idstack is the copilot. It does the first draft, the review and the search for the research. Each skill does the work of one phase of your framework. You make each decision.

For one designer

A second opinion in one minute
Paste an objective, a quiz question, a rubric or a slide. You get findings, each with its study, and a corrected version.
A first draft to start from
Objectives with Bloom's levels, rubrics, question banks, storyboards and module pages. You approve, change or remove each row.
An answer to "why?"
When a subject matter expert disagrees, you show the study and how strong its evidence is.
A full course audit in one afternoon
Import a course from your LMS. You get a quality score, an accessibility score and a list of what to correct first.

For a design team

One standard of review
Each course is examined against the same standards: Quality Matters, WCAG and constructive alignment.
Reports that you can send
Each skill writes a plain Markdown report for faculty, clients and curriculum committees. Each finding has its evidence tier.
Design decisions stay with the course
The needs, the objectives and the alignment are saved in the course folder. A different designer continues where the first one stopped.
New designers learn the reasons
idstack explains each step and each term to a new designer, and it is brief with an expert.

Free and open source. idstack keeps the course files in a folder on your computer.

Three ways to install idstack

Chrome extension

In-browser (Canvas / Docs)

Audit Canvas modules and assignments, Google Docs syllabi and web course pages in the Chrome side panel. You do not need a command line.

Install from Chrome Web Store Free

Claude desktop app

No terminal

In the Code tab of the Claude desktop app, select a folder for your course. Then paste this line in the message box:

/plugin marketplace add savvides/idstack

Click +, then Add plugins, and install idstack. Then start a new session and select the course folder again, because Claude Code loads a plugin only when a session starts. Type /idstack:start. Each skill that writes a report gives you a link to open it in the app.

Claude Code plugin

All skills
git clone https://github.com/savvides/idstack.git && cd idstack && ./setup

./setup registers idstack with Claude Code and installs the skills. Re-run it after a git pull to update.

You need a bash shell (Terminal on macOS or Linux, WSL or Git Bash on Windows). ./setup does not run in PowerShell or cmd. You also need Python 3.9 or later, because the skills save their results with it. scripts/idstack-doctor checks for it.

To check the install, run scripts/idstack-doctor. To see the health of a course, run scripts/idstack-status.

New to Claude Code? Install the Claude Code command-line tool first, then paste the command above in a terminal. In the Claude desktop app, use the steps in the Claude desktop app card. When Claude Code starts, run /idstack:start.

Not using Claude Code? The idstack skills run in Claude Code. Use the Chrome extension, or fork the repo and adapt the skills for a different agent, for example Codex or Antigravity. idstack does not support or test those forks.

AI course design skills for ADDIE, backward design and SAM

The pipeline skills run in sequence through /idstack:pipeline. Other skills handle first-run setup, course import and course memory. Companion skills work at any time: feedback on one item, a storyboard, a question bank and an evaluation plan. You can run any skill alone, with no manifest.

Show the phases of

ADDIE evaluates two times. The three review skills are the formative evaluation, before the launch. /idstack:evaluation-plan plans the summative evaluation, after it.

Results first, then the evidence, then the lessons. The pipeline uses this sequence, so /idstack:course-builder writes no lesson before the objectives and the assessments are there.

Each skill can run again, and the course folder keeps your decisions. Storyboard one module as the prototype, review it with /idstack:quick-check, and start the next cycle.

Sequence (left to right) under /idstack:pipeline

  1. AnalyzeStage 1: resultsPreparation needs-analysis Three-level needs assessment and training-decision gate. Skipped after a course import.
  2. DesignStage 1: resultsIterative design learning-objectives Bloom's classification and bidirectional alignment.
  3. DesignStage 2: evidenceIterative design assessment-design Rubrics, checkpoints and feedback principles.
  4. DevelopStage 3: lessonsIterative development course-builder Cognitive-load-aware content generation.
  5. EvaluateAlignment checkIterative development course-quality-review Quality Matters + Community of Inquiry.
  6. EvaluateAlignment checkIterative development accessibility-review WCAG + UDL audit.
  7. EvaluateAlignment checkIterative development red-team Adversarial audit before launch.
  8. ImplementPublishIterative development course-export Common Cartridge, SCORM, or Canvas API.
start Checks that idstack can run in a course folder, then starts the correct skill. The first skill to run.
course-importAnalyzeStage 1: resultsPreparation Extracts existing modules, assignments, and syllabi from an IMSCC or SCORM package, a PDF or Word file, pasted documents, or the Canvas API.
learn Searches, deletes, promotes, and exports the learnings that skills save, in this course and across courses.

Companion skills, at any time

quick-checkEach phaseEach stageEach review Feedback on one objective, quiz question, rubric, slide or page in one minute. You get cited findings and a corrected version. It needs no setup.
storyboardDevelopStage 3: lessonsIterative design Script and screens for a video, an e-learning module or slides, checked against the multimedia learning principles.
question-bankDevelopStage 2: evidenceIterative development Quiz and test questions from your objectives, with three options, an answer key and feedback for each option.
evaluation-planEvaluateStage 2: evidenceAfter release Measures of reaction, learning, behavior and results, with the survey items and a schedule.

The framework guide has a diagram for each framework and the check that tells you when to leave a phase. It also helps you select a framework. Run the complete workflow with /idstack:pipeline. It tracks course state, skips completed skills, and continues from where you stopped when you run it again.

Evidence-based instructional design: the research behind idstack

The 11 research domains below are ranked by study count. Each bar shows the tier mix of that domain's references. You can see where idstack rests on trials and where it rests on reviews or expert opinion. When findings conflict, the stronger tier wins. All the references are in evidence/references.md. Every source is public, and you can check every citation.

T1 meta-analyses & RCTs T2 quasi-experimental with controls T3 systematic reviews T4 observational T5 expert opinion & theory
  1. Formative Assessment & Feedback

    Feedback that explains the error and the next step improves retention. Feedback that only marks an answer right or wrong does less. Nicol's principles give criteria for rubrics, self-assessment and peer review. T1–T5

  2. Cognitive Load Theory & Instructional Sequencing

    Working memory holds only a few new ideas at one time. Cognitive load research shows when a visual cue helps, and when a repeated explanation overloads the learner. T1–T5

  3. Instructional Design Models & Frameworks

    ADDIE, ARCS, ASSURE and UDL fit different contexts. A 2025 review [ID-7] [T3] compares the strengths and limits of ADDIE, ARCS, and ASSURE. It proposes a framework that combines them. T1–T5

  4. Evaluation Models

    Plan the evaluation while you design the course, not after it. Kirkpatrick's four levels measure reaction, learning, behavior and results. Satisfaction alone does not show learning. T1–T5

  5. Multimedia Learning Principles

    Mayer's principles (coherence, signaling, segmenting, modality, redundancy) tell how to combine words and pictures. Text or decoration that does not teach adds load and reduces learning. T1–T5

  6. Online Course Quality Frameworks

    Quality Matters (QM) standards and the Community of Inquiry (CoI) model are the most studied frameworks for online course quality. CoI measures teaching, social and cognitive presence. idstack audits against both. T1–T5

  7. Accessibility & Universal Design for Learning

    WCAG 2.1 AA is the legal baseline for digital accessibility. Universal Design for Learning (UDL 3.0) plans three pathways (engagement, representation, and action) from the start, so that more learners can succeed. T1–T5

  8. Constructive Alignment & Learning Objectives

    When objectives, activities and assessments do not agree, students learn less. The revised Bloom's taxonomy classifies each objective, so that the assessment can measure the same level. T1–T5

  9. Learner Analysis & Differentiation

    Prior knowledge is the strongest predictor of what a learner gains from instruction. Novices need support that slows experienced learners. This is the expertise-reversal effect. T1–T5

  10. Needs Analysis

    A needs analysis examines three levels (organization, task and learner) before anyone writes content. It keeps the course on the skill gap, and it shows when training is not the answer. T1–T3

  11. Rapid Prototyping & Design-Based Research

    A pilot with real learners finds usability and content problems early. Design-based research gives the revisions a method. T2–T5

Every recommendation has a research domain code and an evidence tier, for example [Alignment-14] [T1]. The code points to the paper in evidence/references.md.

Each pipeline skill writes a report and a manifest.

Each skill that produces findings writes two files: a Markdown report for faculty and stakeholders, and a JSON manifest that later skills read.

Markdown · .idstack/exports/<course-slug>/learning-objectives.md

Finding align-1 warning T5

What we saw: 3 of 8 stated learning objectives have no associated assessment items.

Why it matters: In constructive alignment, each objective has an assessment that measures it, so learners and teachers can see what the course asks for. [Alignment-1] T5

Recommendation: idstack recommends that you run /idstack:assessment-design to add aligned assessments, or change the objective list.

JSON · .idstack/project.json
{
  "learning_objectives": {
    "report_path": ".idstack/exports/<course-slug>/learning-objectives.md",
    "alignment_matrix": {
      "gaps": [
        "ILO-3 has no matching assessment."
      ]
    }
  },
  "updated": "2026-05-13T14:32:18Z"
}

The report is plain text that you can send to faculty and curriculum committees. The manifest keeps the state of the course for the skills. /idstack:pipeline writes a dashboard with readiness scores, the issues that recur across skills, and a link to each report.

AI instructional design FAQ

What is an AI instructional design tool?

It is software that uses AI to help you do the work of an instructional designer. idstack has skills that run in Claude Code, and a Chrome extension that audits Canvas, Google Docs and web pages. Each skill does one job, for example a needs analysis, learning objectives, a rubric or a quality review.

Does idstack replace an instructional designer?

No. idstack writes a first draft and finds the research. You make each decision: you approve, change or remove each row.

Which instructional design frameworks does idstack support?

idstack helps you apply ADDIE, backward design or SAM. Each skill does the work of one phase, for example Analyze in ADDIE or Stage 1 in backward design. You can also name the phase, for example "I am in the Design phase", and Claude starts the correct skill.

Where do the recommendations come from?

Each recommendation cites a peer-reviewed study from the idstack reference list. Each citation has an evidence tier, from T1 (meta-analyses and randomized controlled trials) to T5 (expert opinion). The full list is in evidence/references.md, so you can check each citation.

Does idstack work with Canvas, Blackboard and Moodle?

Yes. /idstack:course-import reads a Common Cartridge file from Canvas, Blackboard, Moodle or D2L, and it can connect to Canvas through the Canvas API. /idstack:course-export writes a Common Cartridge file for these four systems or a SCORM package, or it sends the course to Canvas.

Is idstack free, and what do I need to run it?

idstack is free and open source, with the MIT license. The skills need Claude Code and Python 3.9 or later. The Chrome extension needs a free Google AI Studio API key for real audits, and without a key it shows sample findings.