Imported from dongzhang84/snowboat-blog (.claude/skills/baoyu-diagram/SKILL.md). Install upstream with npx skills add dongzhang84/snowboat-blog --skill baoyu-diagram. Copyright stays with the author.

Diagram Generator

Write real SVG code directly, following a consistent design system, the output is self-contained .svg files (embedded styles, auto dark-mode), editable by humans, scales to any size without quality loss, and embeds cleanly into articles, WeChat posts, slide decks, Notion, and markdown.

When given source material (topic descriptions, documents, technical specs, pasted content), the skill analyzes what diagrams would best convey the material, recommends diagram type(s) and whether the content should be split into multiple focused diagrams, confirms the plan once, then generates all diagrams.

This is not an image-generation skill — it does not call any LLM image model. Claude writes the SVG node-by-node, doing the layout math by hand so every diagram honors the rules in references/.

Usage

# Topic string — skill analyzes and proposes a plan
/baoyu-diagram "how JWT authentication works"

# File path — skill reads, analyzes, and proposes a plan
/baoyu-diagram path/to/content.md

# Pasted content — prompts for input if no argument given
/baoyu-diagram

# Force a specific diagram type (skips type recommendation)
/baoyu-diagram "transformer attention"    --type illustrative
/baoyu-diagram "Kubernetes architecture"  --type structural
/baoyu-diagram "CI/CD pipeline"           --type flowchart
/baoyu-diagram "OAuth 2.0 flow"           --type sequence
/baoyu-diagram "Shape hierarchy"          --type class

# Language and output path
/baoyu-diagram "微服务架构"              --lang zh
/baoyu-diagram "build pipeline" --out docs/build-pipeline.svg

Options

Diagram types

Pick the type by what the reader needs, not by the noun in the prompt.

The primary test: is the reader trying to document this, or understand it? Documentation wants precision — flowchart, sequence, or structural. Understanding wants the right mental model — illustrative.

Routing heuristic: "how does X work" is the default ambiguous case. Prefer illustrative unless the user specifically asks for steps or components. A diagram that makes the reader feel "oh, that's what it's doing" is illustrative — even if the subject is software.

Multi-actor test for sequence: if the prompt names ≥2 distinct actors/participants/services (User + Server, Client + Auth + Resource, Browser + CDN + Origin), prefer sequence even when the verb is "flow" or "process". Single-actor "X flow" (build pipeline, request lifecycle, GC) stays flowchart. When you pick sequence for a multi-actor reason, announce it: "Picked sequence because the prompt names N actors (…). Rerun with --type flowchart to force the step-list version."

Worked examples of verb-based routing: same subject, different diagram depending on what was asked. Use these as a sanity check after picking a type.

Most common routing failure: picking a flowchart because it feels safer when an illustrative diagram would give the reader more insight. Illustrative is the more ambitious choice, and almost always the right one when the reader needs understanding rather than documentation.

Cycles, ERDs, and gantt charts are out of scope for v1. For cycles, draw the stages linearly with a small ↻ returns to start return glyph (see flowchart.md). For ERDs, suggest a dedicated tool (mermaid, plantuml) — do not attempt to fake them in pure SVG.

Workflow

Step 1: Capture input

Read the user's prompt, content file, or pasted content. Note any flags (--type, --lang, --out).

If --out is given, the skill will generate exactly one diagram at that path — the analysis in Step 2 produces a single-diagram plan focused on the most important aspect of the material.

Step 2: Analyze material and produce plan

Analyze the source material and make three decisions:

Decision A: Type routing

For the input material, determine which diagram type(s) are appropriate using the routing table in "Diagram types."

Decision B: Content splitting

Assess whether the material should produce one diagram or multiple sub-diagrams.

Single diagram when:

• Material is focused on one concept, one mechanism, one process
• Named elements count is manageable (under ~6 for flowchart, under ~4 actors for sequence, under ~3 containers for structural — but architecture diagrams may have 10–20 elements in a single diagram; see Step 5a item 6, "Architecture enrichment")
• One "After seeing this diagram, the reader understands ___" sentence covers the whole material

Multiple sub-diagrams when:

• Material covers 2+ independent mechanisms or processes
• Named element count exceeds comfortable limits for one diagram type
• Material has natural subsections that each deserve visual treatment
• Different parts of the material map to different diagram types

For each sub-diagram, determine: focus area, recommended type, named elements, and the "reader understands ___" sentence.

What to diagram:

• Core mechanisms the reader needs to understand (→ illustrative)
• Multi-step processes described in prose (→ flowchart)
• Multi-actor interactions (→ sequence)
• Architectural descriptions with containment or hierarchy (→ structural)
• Type hierarchies or data models (→ class)
• Comparisons between two approaches or systems (→ structural subsystem)

What NOT to diagram:

• Simple lists — a bullet list is already visual enough
• Concepts already shown in an existing image or figure
• Purely emotional or narrative passages with no underlying mechanism
• Content that is a single sentence or trivially simple
• Decorative filler — every diagram must earn its place with a concrete reader need

Decision C: Overview diagram

When the plan includes multiple sub-diagrams, assess whether an additional overview diagram that shows the big picture is worthwhile.

Plan output

Save the plan as outline.md (for multiple diagrams) or hold in memory (for single diagram).

Single-diagram plan format:

## Diagram Plan
**Material**: [source description]
**Diagrams**: 1
**Type**: [type] (rationale)
**Named elements**: [list]
**Reader need**: "After seeing this diagram, the reader understands ___"
**Slug**: [slug]

Multi-diagram plan format:

---
material: [source description]
slug: [material-slug]
diagram_count: N
language: en
---

Per-diagram entry:

## Diagram 1: [focus area]
**Type**: [type] (rationale)
**Named elements**: [list]
**Reader need**: "After seeing this diagram, the reader understands ___"
**Slug**: [2-4 kebab-case words]
**Filename**: 01-{type}-{slug}/diagram.svg

## Diagram 2: [focus area]
...

## Overview diagram (if applicable)
**Type**: [structural/illustrative]
**Purpose**: Shows how diagrams 1-N relate as parts of a larger system
**Named elements**: [high-level elements]
**Slug**: overview-[slug]
**Filename**: overview-{type}-{slug}/diagram.svg

Requirements:

• Each diagram justified by a concrete reader need (the "After seeing this..." sentence)
• Type chosen per the routing table, not arbitrarily
• If input was pasted content, also save it as source-{slug}.md in the output directory

Step 3: Confirm plan (one-time)

Maximum 1 AskUserQuestion call for the entire workflow. This is the only confirmation step — no further questions during generation.

Language question: only include if material language differs from user's language and --lang is not given.

Example full plan confirmation:

I analyzed the material and recommend N diagrams [+ an overview]:

1. [Focus area] — [type] — "Reader understands ___"
2. [Focus area] — [type] — "Reader understands ___"
3. [Focus area] — [type] — "Reader understands ___"
[Overview: [type] — "Shows how 1-N relate as a system"]

Adjust the plan? (add/remove diagrams, change types, skip/add overview)

After confirmation (or after skipping confirmation for obvious plans), the plan is locked. Proceed to generation.

Save the finalized plan:

• Multiple diagrams: diagram/{material-slug}/outline.md
• Single diagram: plan is saved as plan.md beside the SVG in Step 5g

Step 4: Load shared references

Always read:

• references/design-system.md — philosophy, typography, color palette, hard rules
• references/svg-template.md — the <style> + <defs> boilerplate to copy verbatim
• references/layout-math.md — text-width estimation, viewBox sizing, arrow routing
• references/pitfalls.md — the pre-save checklist

Per-type reference files are loaded inside the generation loop (Step 5b) since each diagram may have a different type.

Step 5: Per-diagram generation loop

For each diagram in the confirmed plan (1 to N, overview diagram generated last):

5a: Capture intent

Read the current diagram's plan entry. Extract or refine these five things from the source material:

1. Named elements — list every distinct actor, component, service, state, or phase explicitly named. Count them. If the count is 6+ for simple flowcharts, plan multiple diagrams rather than cramming everything into one (see flowchart.md → "Planning before you write SVG"). Exception: structural architecture diagrams (microservices, cloud topologies, system designs) routinely need 10–20 named elements in a single diagram — services, databases, gateways, message buses, clients. 10 is the minimum where splitting is unnecessary; enriched diagrams (item 6 below) typically reach ≥12. Do not split an architecture diagram just because the element count is high; instead, see "Architecture enrichment" (item 6 below) and structural.md → "Full architecture layout".

2. Relationship type — for each interaction between elements, classify it:

   • Sequential steps / order of operations → flowchart signal
   • Containment ("X is inside Y", zones, hierarchies) → structural signal
   • Multi-actor message exchange (A sends to B, B replies to C) → sequence signal
   • Mechanism ("how does X produce Y") → illustrative signal More than one type present? Pick the dominant one, or flag for the plan.

3. What the reader needs — complete this sentence before routing: "After seeing this diagram, the reader understands ___." If you can't finish it, the topic is underspecified — ask.

4. Label preview — for each element name, count the characters. Latin titles >30 chars (CJK >16) will overflow a 180-wide box and need shortening. Draft the abbreviated form now, before layout math, so Step 5d uses real labels.

5. Language — CJK vs. Latin. Affects text-width multipliers in Step 5d (15 px/char vs. 8 px/char for titles). Mixed content (CJK labels with some Latin terms) counts as CJK.

6. Architecture enrichment — when the type is structural and the topic is an architecture or infrastructure diagram ("microservices architecture", "Kubernetes cluster", "cloud topology", "system design"), actively expand the named elements beyond what the user literally wrote. The user's prompt is a seed, not a complete spec. A bare "microservices architecture" should produce a diagram with ≥12 named elements, not 5 generic boxes. Apply these enrichment rules:

   • Multiple client types: if the system serves end users, show ≥2 clients (Web app + Mobile app, or Browser + CLI). Each with a tech subtitle (React SPA, iOS/Android).
   • Gateway details: technology name (Kong, Nginx, Envoy), responsibilities (rate limiting, auth/routing), port (:443). Not just "API Gateway".
   • Per-service specifics: each microservice gets a technology and port subtitle (Go :8081, Java :8082, Python :8083, Node.js :8084). The reader should learn the tech stack from the diagram.
   • Database per service: each service connects to its own data store. Show the databases as a separate column or tier (PostgreSQL, MongoDB, Elasticsearch, Redis) with role subtitles (Users DB, Orders DB, Cache/Queue).
   • Message bus / event bus: if services communicate asynchronously, show messaging infrastructure (Kafka, RabbitMQ, Event Bus) as small labeled connector pills between the services that use them.
   • Auth service: when JWT/OAuth is mentioned or implied, separate it from business services as a distinct component with protocol subtitle (OAuth 2.0 / JWT).
   • Color categories: architecture diagrams with ≥3 component types trigger the structural architecture exception (see design-system.md rule 9). Assign one ramp per category: services=teal, databases=purple, gateways=coral, message buses=amber. Mandatory legend.
   • Summary panel (optional): a bottom section with 2–3 columns summarizing key architecture principles (Client Applications, Microservices, Infrastructure). Add when the diagram has ≥10 named elements.
   • Title + subtitle: architecture diagrams always get a .title at the top with the architecture name and a .ts subtitle describing the approach (e.g., "Domain-driven design", "Event-driven microservices").

   The enrichment principle: a reader should learn something specific from the diagram. "User service / Go :8081" teaches more than "User service / Accounts & profiles". Technology choices, ports, and protocols are the details that make an architecture diagram useful rather than decorative.

   Skip enrichment for non-architecture structural diagrams (biological containment, CPU caches, file systems) — those benefit from simplicity, not tech details.

5b: Load type reference

The type was determined in the plan. Load the matching reference file.

Read the one that matches the type:

• references/flowchart.md
• references/sequence.md
• references/structural.md
• references/illustrative.md
• references/class.md

Read on demand when the plan calls for a small pictorial element (status circle on a decision branch, checkbox inside a list, queue slot inside a box, doc/terminal/script icon inside a subject, annotation circle on a connector, paired pub/sub arrows, dashed future-state node) or when drawing a phase-band diagram (compact tool card icons, operator icons):

• references/glyphs.md — the shared glyph library, tool card icon set, operator icons, and dark-mode rules

Read on demand for diagram type extensions:

• references/flowchart-poster.md — when ≥3 poster-mode triggers fire in Step 5d (topic has a short name, named phases, parallel candidates, a loop termination mechanic, overflow annotations, or a footer quote)
• references/flowchart-phase-bands.md — when the prompt describes a multi-phase sequential operation where each phase contains parallel tools or steps and outcomes propagate between phases
• references/structural-network.md — when drawing network topology: zone containers, wired/wireless device connectivity, security zones
• references/structural-matrix.md — when drawing a comparison matrix: feature table, ✓/✗ cells, side-by-side grid

5c: Check patterns library

If the topic matches a known AI-system pattern, there is a pre-cooked starter plan in references/patterns/. Scan references/patterns/README.md for a pattern name that matches. If one matches, load that pattern file and use its mermaid reference + baoyu SVG plan as the starting point for Step 5d.

If nothing matches, skip and plan from scratch in Step 5d. Do not force a near-miss.

5d: Plan on paper

Before writing any SVG, draft a short layout plan. Do the math once, correctly, so the SVG comes out right on the first pass.

5d-0. Draft the Mermaid sketch first — write a Mermaid code block that captures the structural intent of the diagram: which nodes exist, how they connect, what direction they flow, and any grouping (subgraphs). This is the single source of truth for what to draw; everything after it (coordinates, widths, arrows) answers how.

Rules for the Mermaid sketch:

• Use the Mermaid dialect that best matches the diagram type: flowchart TD/LR for flowcharts, sequenceDiagram for sequence, classDiagram for class, flowchart with subgraphs for structural/illustrative.
• Include every node, every edge, every label, and every subgraph/container. If a node won't appear in the Mermaid, it won't appear in the SVG.
• Edge labels must match the final SVG labels — write them now, not later.
• Keep it concise: the sketch is a structural contract, not a rendering. Mermaid can't express baoyu's visual design (colors, rounded rects, dark mode), so don't try — those come in 5d-ii and 5e.
• For patterns that have a Mermaid reference in references/patterns/, start from that reference and adapt it to the specific topic.

Save the Mermaid block in the plan file. When writing SVG in Step 5e, cross-check every node and edge against this Mermaid sketch — if the sketch has it, the SVG must have it; if the SVG adds something the sketch doesn't have, update the sketch first.

5d-i. Extract structure from the source — don't just transcribe bullets into boxes. Read the source looking for these elements. Not every element will be present, but every present element should land in the diagram:

• Mechanism name — does the topic have a short, nameable identity (Autoreason, AutoResearch, OAuth, JWT auth, Reflexion loop)? If yes, that's a candidate .title.
• Framing question — does the source contain a "why does this exist" sentence? That's a candidate subtitle.
• Phases — do the stages naturally cluster into 2–4 named groups? Each cluster is a candidate .eyebrow section.
• Anchor inputs — is there a constant input (the task prompt, a dataset, a knowledge base) that every stage references? That's a candidate anchor box above the main flow.
• Parallel candidates — at some point, does the process generate N alternatives that are then compared? Watch for the implicit "keep unchanged" candidate.
• Loop scope + termination — which boxes are inside a loop that repeats? What is the specific termination rule? That's a candidate left-rail loop bracket + a dedicated termination box.
• Per-box context that won't fit in a subtitle — those are candidate right-column .anno annotations.
• Quotable hook — does the source end with a test result, a quote, or a memorable framing? That's a candidate footer .caption.
• Role categories — how many distinct kinds of operation does the process have? This determines the color budget. Identity is a category, not a sequence.

Write the answers to these in the plan file. If ≥3 of them land, you're building a poster flowchart — load references/flowchart-poster.md and follow its coordinate budget. Otherwise, it's a simple flowchart and the linear-top-down pattern applies.

5d-ii. Draft the layout:

1. List the nodes / regions / shapes with their full label text (title + optional subtitle).
   • Simple flowchart: ≤5 nodes.
   • Poster flowchart: ≤12 nodes grouped into ≤4 eyebrow-divided phases.
   • Structural: ≤3 inner regions.
   • Illustrative: 1 subject.
   • Sequence: list actors (2–4, max 4) in left-to-right order, each with a short title (≤12 chars) and optional role subtitle; then list messages as ordered (sender, receiver, short label) tuples (6–10 total, 10 is the sweet spot); mark any self-messages; draft a side-note title for the protocol.
2. For every rect, compute the width using the formula in layout-math.md:
   • width = max(title_chars × 8, subtitle_chars × 7) + 24 (Latin)
   • Replace 8 with 15 and 7 with 13 for CJK
   • Round up to the nearest 10
3. Pick colors by category, not sequence. ≤2 accent ramps per diagram. Gray for neutral/start/end. Reserve blue/green/amber/red for semantic meanings.
   • Sequence exception: assign one ramp per actor (default [gray, teal, purple, blue]), up to 4 ramps total — arrows inherit the sender's ramp.
   • Poster-flowchart exception: up to 4 ramps, one per distinct agent/role (drafter=purple, critic=coral, synthesizer=teal, judge=amber). Baseline/anchor/convergence stay gray.
4. Check tier packing: N × box_width + (N-1) × gap ≤ 600. For sequence, use the lane table in layout-math.md (N=4 → centers 100/260/420/580) and verify every message label fits its lane span with label_chars × 7 ≤ |sender_x − receiver_x| − 8. For poster fan-out rows (3 candidates), see the coordinate sketch in flowchart.md.
5. Map arrows and verify none cross an unrelated box. Use L-bends where a straight line would collide. (Sequence messages are always straight horizontal lines — no L-bends. Fan-out candidates converge to a common ymid channel just above the judge box.)
6. Compute viewBox height: H = max_y + 20 where max_y is the bottom of the lowest element. Poster flowcharts routinely reach H=800–950 — don't force them to be compact.

Save this plan (including the Mermaid sketch from 5d-0):

• One diagram: diagram/{slug}/plan.md
• Multiple diagrams: diagram/{material-slug}/NN-{type}-{slug}/plan.md

5e: Write the SVG

Start from the Mermaid sketch in the plan. Walk the sketch node-by-node, edge-by-edge, and translate each element into SVG using the coordinates and widths computed in 5d-ii. The Mermaid sketch is the structural checklist — every node and edge in it must appear in the SVG. If you find yourself adding an element that isn't in the sketch, stop and update the sketch first so the plan stays authoritative.

Emit a single <svg width="100%" viewBox="0 0 680 H"> element. Copy the <style> + <defs> block from svg-template.md verbatim — don't abbreviate or edit the color ramp definitions. Then add visual elements in z-order:

1. Background decorations (rare)
2. Containers (outer <rect> for structural diagrams)
3. Connectors and arrows (drawn first so nodes paint on top)
4. Nodes (rects with text)
5. Labels outside boxes (leader callouts, legends, external I/O labels)

Typography rules:

• Two sizes only: 14px (t, th) and 12px (ts)
• Two weights only: 400 and 500
• Sentence case everywhere — "User login" not "User Login"
• Every <text> element gets a class (t, ts, or th) — never hardcode fill colors on text

5f: Run the pre-save checklist

Mermaid–SVG consistency check (run before the pitfalls checklist): re-read the Mermaid sketch from the plan. For every node in the sketch, confirm the SVG has a corresponding <rect> + <text>. For every edge, confirm a <path> or <line> connects the correct pair. Missing elements are bugs — fix them before continuing.

Walk through every item in references/pitfalls.md. The top failures to catch every time:

1. viewBox height covers every element with a 20px buffer
2. No rect extends past x=640
3. Every labeled rect is wide enough for its text (char-width check)
4. No arrow crosses an unrelated box
5. Every <path> connector has fill="none" (or uses class="arr")
6. Every <text> has a class — no hardcoded fill="black"
7. No text-anchor="end" at low x values (label would clip past x=0)
8. ≤2 accent ramps, colors encode category not sequence
9. No <!-- comments --> in the final output

If any item fails, fix the SVG before saving. Don't rationalize past a failure — the checklist exists because these bugs are silent: the SVG is valid but looks wrong when rendered.

5g: Save and report progress

Save the SVG and plan:

• One diagram: diagram/{slug}/plan.md + diagram.svg
• Multiple diagrams: diagram/{material-slug}/NN-{type}-{slug}/plan.md + diagram.svg

Backup rule: if diagram.svg already exists at the target path, rename the existing one to diagram-backup-YYYYMMDD-HHMMSS.svg before writing the new file — never overwrite prior work silently.

Multiple diagrams progress: after each diagram, report progress: "Generated 2/4: 02-illustrative-jwt-token-structure".

Step 6: Report

One diagram — tell the user in 4-6 lines:

• Diagram type picked (and one-sentence why)
• Node count / complexity
• viewBox dimensions
• Language
• Output file path
• One suggestion for how to preview it (e.g., "Open in Chrome for light/dark check")

Multiple diagrams:

Diagram Generation Complete!

Material: [source description]
Language: [lang]
Diagrams: X generated

Results:
- 01-sequence-jwt-auth-flow — "Reader understands the auth handshake"
- 02-illustrative-jwt-token-structure — "Reader understands token anatomy"
- 03-flowchart-token-refresh — "Reader understands the refresh cycle"
[- overview-structural-jwt-system — "Reader sees how all parts connect"]

Output: diagram/{material-slug}/
Preview: Open any .svg in Chrome for light/dark check

Output structure

One diagram

diagram/{slug}/
├── source-{slug}.md          # optional: saved input material
├── plan.md                   # layout sketch from Step 5d
└── diagram.svg               # final output

Multiple diagrams

diagram/{material-slug}/
├── source-{slug}.md          # saved input material
├── outline.md                # plan from Step 2 with all diagram entries
├── 01-{type}-{slug}/
│   ├── plan.md               # layout sketch for this diagram
│   └── diagram.svg           # final SVG
├── 02-{type}-{slug}/
│   ├── plan.md
│   └── diagram.svg
├── 03-{type}-{slug}/
│   ├── plan.md
│   └── diagram.svg
└── overview-{type}-{slug}/   # optional: overview diagram
    ├── plan.md
    └── diagram.svg

• Slug: 2–4 kebab-case words derived from the topic or concept.
• Backup rule: if diagram.svg already exists at the target path, rename the existing one to diagram-backup-YYYYMMDD-HHMMSS.svg before writing the new file.
• Plan: always save plan.md beside the SVG so the next iteration can re-read it.
• Source: if the user pasted source content, save it as source-{slug}.md in the output directory.
• Numbering: NN prefix (01, 02, ...) matches the plan order.
• Outline: when generating multiple diagrams, always save outline.md from Step 2 so the generation can be resumed or individual diagrams can be regenerated.

Modification

Core principles

• Draw the mechanism, not a diagram about the mechanism (illustrative). Draw the sequence, not the architecture (flowchart). Draw the containment, not the flow (structural). Draw the conversation, not the steps (sequence). Picking the wrong type is the single biggest failure mode — more harmful than any layout bug.
• One design system, always. No --style flag, no alternate themes, no per-topic visual variants. The cohesive look across every diagram is the product — if a reader sees two baoyu diagrams in different articles, they should feel they came from the same hand. Any request to "use a different style" is a request to break this principle; push back and ask what the underlying need is instead. All diagrams in a run share the same design system — no per-diagram style overrides.
• Self-contained output. Every SVG carries its own styles and dark-mode rules. The reader should never need to edit anything after pasting it into their article.
• Math before markup. SVG has no auto-layout. Every coordinate is hand-computed. A diagram that "almost fits" has a bug — fix the math, don't nudge pixels.
• Color encodes meaning, not position. Five steps in a flowchart are not five colors. All five are gray unless one specific step deserves emphasis — in which case it gets the accent color.
• The reader has 3 seconds. If the diagram needs prose explanation to parse, it's failing. Simplify until it can stand alone with only its labels.

References

• references/design-system.md — palette, typography, hard rules
• references/svg-template.md — the <style> + <defs> boilerplate (copy verbatim)
• references/layout-math.md — coordinates, text widths, viewBox math, arrow routing
• references/pitfalls.md — the pre-save checklist
• references/flowchart.md — flowchart-specific rules and worked examples (includes state-machine sub-pattern)
• references/flowchart-poster.md — poster flowchart dialect (load on demand when ≥3 poster triggers fire)
• references/flowchart-phase-bands.md — phase-band flowchart (horizontal dashed phase containers, compact tool card rows, cross-band semantic arrows, operator icons, legend strip)
• references/sequence.md — sequence-diagram rules (actors, lifelines, messages, self-messages)
• references/structural.md — structural-specific rules and worked examples (subsystem, bus, radial star, rich interior, mixed arrows)
• references/structural-network.md — network topology sub-pattern (zone containers, wired/wireless, tiered layout)
• references/structural-matrix.md — comparison matrix sub-pattern (feature table, ✓/✗ cells, zebra rows)
• references/illustrative.md — illustrative-specific rules and worked examples
• references/class.md — UML class diagram rules (3-compartment rects, relationships, stereotypes)
• references/glyphs.md — shared glyph library (status circles, checkboxes, queue slots, icons, annotation circles) and concept-to-shape conventions
• references/patterns/ — pre-planned starters for common AI-system topologies (RAG, agents, memory tiers, verifier loops, …)