microsim-generator

Version: 1.0

Overview

This meta-skill routes MicroSim creation requests to the appropriate specialized generator based on visualization requirements. It consolidates 17 individual MicroSim generator skills into a single entry point with on-demand loading of specific implementation guides.

Six Python batch utilities in src/microsim-utils/ automate the repetitive parts of MicroSim generation (parsing specs, scaffolding files, inserting iframes, fixing iframe heights, validating quality, updating navigation), saving ~430K tokens per batch run. The agent's creative work is focused on writing the .js file.

Default Sequential Execution

When the microsim generator skill us used on all of the #### Diagram elements of a chapter, always run the microsim generator tasks sequentially unless the user specifically uses the phrase "execute in parallel".

When to Use This Skill

Use this skill when users request:

• Interactive educational visualizations
• Data visualizations (charts, graphs, plots)
• Timelines or chronological displays
• Geographic/map visualizations
• Network diagrams or concept maps
• Flowcharts or workflow diagrams
• Mathematical function plots
• Set diagrams (Venn)
• Priority matrices or bubble charts
• Custom simulations or animations
• Comparison tables with ratings
• Matrix comparisons with expandable cell details
• Fact-verified infographic posters where every statistic is traced to a cited source
• Batch generation of MicroSims for a chapter

---

Step 0: Environment Setup

Before any generation work, establish paths and determine scope.

0.1 Set Paths

# Python utilities live in the ibook-skills repo
UTILS="$HOME/Documents/ws/ibook-skills/src/microsim-utils"

# Detect project root (directory containing mkdocs.yml)
PROJECT=$(python3 -c "
import os, sys
d = os.path.abspath('.')
while d != os.path.dirname(d):
    if os.path.isfile(os.path.join(d, 'mkdocs.yml')): print(d); sys.exit()
    d = os.path.dirname(d)
print('ERROR: mkdocs.yml not found', file=sys.stderr); sys.exit(1)
")

If the scripts are not found at $UTILS, check for them in the project's own src/microsim-utils/ directory.

0.2 Determine Scope

| User Request | Route | |-------------|-------| | "Generate MicroSims for chapter 11" | Chapter batch → Step 1 | | "Create a MicroSim for inscribed angles" | Single sim → Step 1B | | "Build a timeline of Unix history" | Single sim → Step 1B |

---

Step 1: Extract Specs from Chapter (Batch Route)

MANDATORY for chapter-level generation. Do NOT manually parse chapter markdown.

python3 $UTILS/extract-sim-specs.py \
    --project-dir $PROJECT \
    --chapter <chapter-dir-name> \
    --output /tmp/ch-specs.json \
    --status-file /tmp/sim-status.json \
    --verbose

What this does:

• Parses all #### Diagram: and #### Drawing: headers from the chapter's index.md
• Extracts <details> block content, iframe paths, sim IDs, Bloom levels, and library hints
• Produces a JSON array of spec objects (one per sim)
• Generates a sim-status.json with lifecycle states: specified → scaffolded → implemented → validated → deployed

Read the output:

cat /tmp/ch-specs.json | python3 -m json.tool | head -60

Check the status file to see which sims need work:

python3 -c "
import json
with open('/tmp/sim-status.json') as f:
    for e in json.load(f):
        if e['status'] != 'deployed':
            print(f\"  {e['status']:12s}  {e['sim_id']}\")
"

After extracting, proceed to Step 2.

---

Step 1B: Single-Sim Shortcut

For a single MicroSim request (not a full chapter batch):

1. Write a 1-entry spec JSON manually:

cat > /tmp/ch-specs.json << 'EOF'
[{
    "sim_id": "inscribed-angle-theorem",
    "title": "Central vs Inscribed Angles Interactive",
    "summary": "Demonstrates the inscribed angle theorem",
    "heading_type": "Diagram",
    "chapter": "",
    "element_type": "microsim",
    "bloom_level": "Analyze",
    "library": "p5.js",
    "iframe_src": "",
    "iframe_height": "",
    "spec_text": "",
    "status": ""
}]
EOF

2. Run scaffold for just that sim:

python3 $UTILS/generate-sim-scaffold.py \
    --spec-file /tmp/ch-specs.json \
    --sim-id inscribed-angle-theorem \
    --project-dir $PROJECT \
    --verbose

3. Jump to Step 3 (Instructional Design Checkpoint), then Step 4 (Implement .js).

---

Step 2: Scaffold Sim Directories

MANDATORY for batch generation. Do NOT manually create main.html, index.md, or metadata.json.

python3 $UTILS/generate-sim-scaffold.py \
    --spec-file /tmp/ch-specs.json \
    --project-dir $PROJECT \
    --verbose

What this does:

• Creates docs/sims/<sim-id>/ directory for each spec
• Generates main.html with correct CDN links, <main> tag, and schema meta tag
• Generates index.md with frontmatter, iframe embed, fullscreen link, lesson plan skeleton
• Generates metadata.json with Dublin Core fields and educational metadata
• Skips directories that already exist (unless --force is used)

Using --force for partially-built sims:

When sim directories exist (with main.html + .js) but lack index.md or metadata.json, the scaffold script skips the entire directory by default. Use --force to regenerate all scaffold files. This is safe because the creative work lives in the .js file, which the scaffold never overwrites.

python3 $UTILS/generate-sim-scaffold.py \
    --spec-file /tmp/ch-specs.json \
    --project-dir $PROJECT \
    --force \
    --verbose

After scaffolding, the agent ONLY writes .js files from here on.

---

Step 3: Instructional Design Checkpoint (MANDATORY)

Before writing any .js file, you MUST complete this checkpoint.

3.1 Identify Learning Objective Details

Extract from the specification:

• Bloom Level: Remember, Understand, Apply, Analyze, Evaluate, or Create
• Bloom Verb: The action verb (explain, demonstrate, calculate, etc.)
• Learning Objective: The full statement of what learners will be able to do

3.2 Match Interaction Pattern to Bloom Level

| Bloom Level | Appropriate Patterns | Inappropriate Patterns | |-------------|---------------------|------------------------| | Remember (L1) | Flashcards, matching, labeling | Complex simulations | | Understand (L2) | Step-through worked examples, concrete data visibility | Continuous animation, particle effects | | Apply (L3) | Parameter sliders, calculators, practice problems | Passive viewing only | | Analyze (L4) | Network explorers, comparison tools, pattern finders | Pre-computed results | | Evaluate (L5) | Sorting/ranking activities, rubric tools | No feedback mechanisms | | Create (L6) | Builders, editors, canvas tools | Rigid templates |

3.3 Answer These Questions

Before proceeding, answer these questions:

1. What specific data must the learner SEE?

   • Not "animated particles" but "the tokenized array ['physics', 'ball']"

2. Does the learner need to PREDICT before observing?

   • If YES → Use step-through with Next/Previous buttons
   • If YES → Do NOT use continuous animation

3. What does animation add that static arrows don't?

   • If you can't answer this clearly → Don't use animation

4. Is continuous animation appropriate for this Bloom level?

   • For Understand (L2) with verb "explain" → Almost always NO
   • For Apply (L3) with real-time feedback → Often YES

3.4 Modify Specification If Needed

If the specification requests animation/effects for an UNDERSTAND level objective:

• Flag this as a potential instructional design issue
• Recommend step-through pattern instead
• Ask user: "The specification requests animation, but for an 'explain' objective, a step-through approach with concrete data visibility typically supports learning better. Should I proceed with step-through instead?"

3.5 Document Your Decision

Add to your response:

Instructional Design Check:
- Bloom Level: [level]
- Bloom Verb: [verb]
- Recommended Pattern: [pattern]
- Specification Alignment: [aligned/modified]
- Rationale: [why this pattern supports the learning objective]

---

Step 4: Implement .js Files

This is where the agent's creative work happens. For each sim that needs implementation:

4.1 Analyze Request and Match Generator

Scan the spec for trigger keywords and match to the appropriate generator guide.

Quick Reference Routing Table

| Trigger Keywords | Guide File | Library | |------------------|------------|---------| | timeline, dates, chronological, events, history, schedule, milestones | references/timeline-guide.md | vis-timeline | | map, geographic, coordinates, latitude, longitude, locations, markers | references/map-guide.md | Leaflet.js | | function, f(x), equation, plot, calculus, sine, cosine, polynomial | references/plotly-guide.md | Plotly.js | | network, nodes, edges, graph, dependencies, concept map, knowledge graph | references/vis-network-guide.md | vis-network | | flowchart, workflow, process, state machine, UML, sequence diagram | references/mermaid-guide.md | Mermaid.js | | venn, sets, overlap, intersection, union, categories | references/venn-guide.md | Custom | | chart, bar, line, pie, doughnut, radar, statistics, data | references/chartjs-guide.md | Chart.js | | bubble, priority, matrix, quadrant, impact vs effort, risk vs value | references/bubble-guide.md | Chart.js | | causal, feedback, loop, systems thinking, reinforcing, balancing, CLD, systems archetype | references/causal-loop-guide.md | vis-network | | comparison, table, ratings, stars, side-by-side, features | references/comparison-table-guide.md | Custom | | matrix, framework comparison, clickable cells, detail panel, expandable | references/html-table.md | Custom | | animation, celebration, particles, confetti, effects | references/celebration-guide.md | p5.js | | classify, classifier, categorize, sort scenarios, identify types, recognize patterns | references/concept-classifier-guide.md | p5.js | | diagram overlay, callout labels, anatomy, labeled illustration, infographic overlay, explore/quiz modes | references/infographic-overlay-guide.md | Custom (diagram.js) — interactive sim; labels live in data.json, image carries no text | | python lab, code runner, runnable code block, interactive python exercise, docker | references/docker-python-lab-guide.md | Custom (docker-lab.js) | | verified infographic, statistics poster, fact-checked poster, cited data, sourced claims, evidence-based comparison | references/verified-infographic-guide.md | Custom (text-verify → image) — static PNG; numbers baked into pixels, never the final deliverable on its own | | custom, simulation, physics, interactive, bouncing, movement, p5.js | references/p5-guide.md | p5.js |

Disambiguating the Two infographic Routes

Both routes have "infographic" in the name and they are routinely confused. Pick by asking where the words live:

| | infographic-overlay-guide.md | verified-infographic-guide.md | |---|---|---| | Produces | A MicroSim in docs/sims/{sim-id}/ | A static PNG + audit trail in docs/posters/<slug>/ | | Problem solved | Making an image explorable and measurable | Keeping baked-in numbers from being fabricated | | Text in the image | None — not even a title | Text and numbers are the content | | Where labels live | data.json, rendered at runtime by diagram.js | Inside the pixels, locked before rendering | | Web search / citations | Not used | Mandatory (2+ searches per claim, source_id per element) | | Emits interaction events | Yes (hover, quiz attempts) | No — a flat image emits nothing |

Rule of thumb: numeric claims that could be wrong → verified guide; structures that need naming → overlay guide. They are not alternatives — see the two policies below, which make them a pipeline rather than a fork.

Policy: Interactivity Is the Default, Static Images Are Not

Never generate a static image as a deliverable unless the user specifically requests one. A flat PNG is a Level 1 artifact: it emits no interaction events, so it tells us nothing about whether readers understood it. This library targets Level 2+ (see the root CLAUDE.md), which means the default answer to "make an infographic" is an interactive one. When a request could be served either way, route to the overlay guide, not the poster route.

Policy: Every Verified Poster Gets an Overlay

When a static poster carrying verifiable facts is generated — because the user asked for one — it must always be given an interactive overlay afterward. This is not optional and not a follow-up suggestion to offer; it is the second half of the poster route.

The reason is measurement. The poster's value is its verified claims, but a bare PNG cannot tell us whether anyone engaged with those claims. Wrapping it in a grid-diagram.js overlay turns each poster region into an instrumented zone: hovers, zone opens, and quiz attempts become interaction events that let us predict, in aggregate, whether readers have mastered the concepts the infographic teaches. Per the "2.99" target in the root CLAUDE.md, aggregate those events across all readers to estimate concept understanding — never retain per-student performance history tied to an identifiable reader.

Mechanically: after Phase 8, follow the Grid Overlay Workflow in references/infographic-overlay-guide.md, pointing data.json.image at the rendered poster.png and keeping showLabels: false (the poster already has printed column titles). Carry each claim's source_id into the zone facts[] so the citation stays visible.

Decision Tree

Numeric claims that must be verified against sources?
  → YES: run Phases 1–4 of verified-infographic-guide.md to lock a cited claim set, then:
      · user explicitly asked for a static poster/image/PNG?
          → render it (Phases 5–8), THEN always add a grid overlay on top (see Exit Route)
      · otherwise → carry the verified claims into the matched sim guide below.
        Never ship a static image the user did not ask for.

Has dates/timeline/chronological events?
  → YES: timeline-guide.md

Has geographic coordinates/locations?
  → YES: map-guide.md

Mathematical function f(x) or equation?
  → YES: plotly-guide.md

Nodes and edges/network relationships?
  → YES: vis-network-guide.md (or causal-loop-guide.md if systems thinking)

Flowchart/workflow/process diagram?
  → YES: mermaid-guide.md

Sets with overlaps (2-4 categories)?
  → YES: venn-guide.md

Priority matrix/2x2 quadrant/multi-dimensional?
  → YES: bubble-guide.md

Standard chart (bar/line/pie/radar)?
  → YES: chartjs-guide.md

Comparison table with ratings/stars?
  → YES: comparison-table-guide.md

Matrix comparison with clickable cells/detail panels?
  → YES: html-table.md

Celebration/particles/visual feedback?
  → YES: celebration-guide.md

Students classify scenarios into categories (sorting quiz)?
  → YES: concept-classifier-guide.md

Labeled illustration with interactive callout markers or hover zones?
  → YES: infographic-overlay-guide.md

Runnable Python code block executed in Docker?
  → YES: docker-python-lab-guide.md

Custom simulation/animation/physics?
  → YES: p5-guide.md

Exit Route: Fact-Verified Poster (not a MicroSim)

Gate this route first. Only take it when the user has specifically asked for a static poster, image, or PNG. If they asked for an "infographic" without naming a static format, do not render a poster — run Phases 1–4 for the verified claim set and hand it to an interactive guide instead (see "Policy: Interactivity Is the Default" above).

references/verified-infographic-guide.md is the one route in this skill that does not produce a sim directory by itself. It produces a static poster PNG in docs/posters/<slug>/ alongside its verification report and source sidecar — which is then always wrapped in an overlay (see below). When a request matches it:

• Skip Steps 0–9 of this file entirely. No scaffold, no .js, no CANVAS_HEIGHT, no iframe insertion, no quality validator, no docs/sims/ entry.
• Read the guide and follow its 8-phase workflow instead. Phases 1–6 are text-only work by the agent; Phase 7 is the single image-model call; Phase 8 audits the render.
• The hard rule that makes this route worth having: no claim reaches the image until it has a verified source. An unverified claim blocks the pipeline, gets downgraded to qualitative language, or is dropped — it is never rendered on a guess.

MANDATORY Phase 9: wrap the poster in an interactive overlay. A rendered poster is never the final deliverable. As soon as Phase 8 passes, build a grid overlay MicroSim over it by following the Grid Overlay Workflow in references/infographic-overlay-guide.md:

• Point data.json.image at the rendered poster.png
• Keep showLabels: false — the poster already has printed column titles
• Define one zone per poster region, carrying each claim's source_id into the zone facts[]
• Add quiz questions so the overlay measures comprehension, not just exposure

Why this is mandatory: the poster's numbers are verified, but a flat PNG emits no interaction events, so nothing tells us whether readers actually engaged with those claims. The overlay turns each region into an instrumented zone whose hovers, opens, and quiz attempts let us predict — in aggregate across all readers, per the "2.99" target in the root CLAUDE.md — whether the concepts in the infographic have been mastered. Never retain per-student histories tied to an identifiable reader.

Reusing the verification phases for a real MicroSim. Phases 1–4 (claim plan → source discovery → per-claim verification → verification report) are output-format agnostic. When a sim must carry sourced facts — a comparison table of real products, a chart of published measurements, a timeline of dated events — run Phases 1–4 first, then hand the locked claim set to the matched sim guide and carry each source_id into the sim's data file so the citation is visible to the student. Skip Phases 5–8; those are poster-rendering steps.

4.2 Load the Matched Guide

Read the corresponding guide file from the references/ directory and follow its workflow for writing the .js file.

4.3 Write ONLY the .js File

The scaffold (Step 2) already created main.html, index.md, and metadata.json. You only need to write docs/sims/<sim-id>/<sim-id>.js.

Each guide contains:

1. Library-specific requirements
2. Code templates and patterns
3. Best practices for that visualization type

4.4 MANDATORY: Add CANVAS_HEIGHT Comment to .js File

Every .js file MUST include a // CANVAS_HEIGHT: comment on its own line near the top of the file (within the first 10 lines). This is the primary source of truth for iframe height across all library types. The sync-iframe-heights.py utility and manual height-fixing rely on it.

No-.js sims: if a sim is rendered entirely by main.html and ships no <id>.js (some projects author Mermaid, vis-network, or custom-HTML sims this way), there is no .js to hold the comment. Store the height in the sim's metadata.json instead, as "canvasHeight": <integer> — the consistent structured fallback that downstream tooling reads next after the .js comment. See the microsim-utils skill's references/canvas-height-strategy.md for the full resolution order. Everything else in this step (how to calculate the number, the +2 iframe rule) is identical.

Format: // CANVAS_HEIGHT: <integer>

The value is the total pixel height the iframe needs to display the entire MicroSim without clipping — canvas, controls, legends, info panels, titles, and any padding.

How to Calculate CANVAS_HEIGHT by Library Type

| Library | Formula | Example | |---------|---------|---------| | p5.js | drawHeight + controlHeight + graphHeight (the canvasHeight variable) | // CANVAS_HEIGHT: 695 | | vis-network | container height + title (~35px) + info panel (~60px) + legend (~30px) + controls (~40px) | // CANVAS_HEIGHT: 685 | | Chart.js | chart container height + title (~35px) + controls (~40px) + legend (~30px) | // CANVAS_HEIGHT: 505 | | Plotly.js | plot div height + title (~35px) + controls (~40px) | // CANVAS_HEIGHT: 475 | | vis-timeline | timeline container height + title (~35px) + controls (~40px) | // CANVAS_HEIGHT: 475 | | Leaflet.js | map container height + title (~35px) + controls (~40px) + legend (~40px) | // CANVAS_HEIGHT: 515 | | Mermaid.js | diagram container height + title (~35px) + controls (~40px) | // CANVAS_HEIGHT: 475 | | Custom HTML (comparison table, html-table, venn) | total rendered height of all DOM elements | // CANVAS_HEIGHT: 600 | | Interactive Infographic Overlay | image container height + toolbar (~40px) + info panel (~60px) | // CANVAS_HEIGHT: 620 |

Examples for Each Library Type

p5.js:

// Predator-Prey Population Dynamics Simulator
// CANVAS_HEIGHT: 695
let drawHeight = 400;
let graphHeight = 180;
let controlHeight = 115;
let canvasHeight = drawHeight + graphHeight + controlHeight;

vis-network:

// Climate Feedback Loops - vis-network
// CANVAS_HEIGHT: 685
document.addEventListener('DOMContentLoaded', function() {

Chart.js:

// Air Quality Trends Dashboard - Chart.js
// CANVAS_HEIGHT: 505
document.addEventListener('DOMContentLoaded', function () {

Leaflet.js:

// Watershed Map - Leaflet
// CANVAS_HEIGHT: 515
document.addEventListener('DOMContentLoaded', function() {

Interactive Infographic Overlay:

// Ecosystem Components Overlay
// CANVAS_HEIGHT: 620
document.addEventListener('DOMContentLoaded', function() {

Rules

1. The comment MUST appear within the first 10 lines of the .js file so parsing tools can find it quickly.
2. The value MUST be an integer (no px suffix, no expressions).
3. The iframe height = CANVAS_HEIGHT + 2 (2px accounts for the iframe border). The sync-iframe-heights.py utility adds this automatically.
4. For p5.js sims, the CANVAS_HEIGHT value must equal drawHeight + controlHeight (plus graphHeight if present). Keep the existing named variables — the comment is a redundant-but-authoritative declaration for tooling.
5. When updating an existing sim, always update the CANVAS_HEIGHT comment if you change any height-related values.
6. If unsure about the height, render the sim in a browser, measure the total content height, and use that value. Err on the side of being slightly too tall (extra whitespace is better than clipped controls).

4.5 Handling Ambiguous Requests

If the request could match multiple generators:

1. Read references/routing-criteria.md for detailed scoring methodology
2. Score top 3 candidates using the 0-100 scale
3. Present options to user with reasoning:

   Based on your request, I recommend:
   1. [Generator A] (Score: 85) - Best for [reason]
   2. [Generator B] (Score: 70) - Alternative if you need [feature]
   3. [Generator C] 
   

<!-- Content truncated for initial SEO render. Open the source file tab for the full file. -->