DiaScope

Turn any codebase, system, or flow into a narrated interactive diagram story — a self-contained .html file your audience can step through with arrow keys.

You author a D2 diagram and a YAML narration sidecar. DiaScope renders the diagram, wires up the step-by-step navigation, and writes one file you can open in a browser or share anywhere.

Docs: https://diascope.biolytics.ai

Or: install the narrate skill and let an agent do it for you.

---

How it looks

Each story is a full-screen page: diagram on the left, narration panel on the right. Each step highlights the relevant nodes, pans/zooms to them, and shows the title + body text. Click a node for detail. Press → to advance.

Interactive features

• Step-by-step narration with active node highlighting
• Automatic pan and zoom to the current step
• Focus mode to isolate active nodes and edges
• Fit plus manual zoom controls
• Keyboard navigation with arrow keys
• Clickable node detail panels
• Edge hover tooltips

See the live example and feature docs at https://diascope.biolytics.ai.

---

Requirements

• Node.js 18+
• d2 on your PATH

---

Install

npm install -g @biolytics.ai/diascope

---

Human-driven workflow

1. Write your D2 diagram — and optionally annotate it with # @step markers:

# @step step-01
Client -> Server: POST /api/data

# @step step-02
Server -> Database: SELECT ...

# @step step-03
Database -> Server: result row
Server -> Client: 200 OK

# @step comments are invisible to the D2 renderer. They tell diascope init which nodes belong to which narrative beat.

2. Scaffold the narration:

diascope init auth-flow.d2
# → writes auth-flow.story.yaml

3. Fill in the narration — edit auth-flow.story.yaml:

meta:
  title: "Auth Flow"
  d2_source: auth-flow.d2

steps:
  - id: step-01
    tag: "01"
    title: "Client sends authenticated request"
    body: |
      The client attaches a short-lived JWT in the Authorization header.
      Tokens expire after 15 minutes — refresh before expiry or the
      request is rejected at the next step.
    nodes:
      - Client
      - Server

  - id: step-02
    tag: "02"
    title: "Server queries the database"
    nodes:
      - Server
      - Database

4. Build:

diascope build auth-flow.d2 auth-flow.story.yaml -o auth-flow.html
open auth-flow.html

---

Agent-driven workflow

Install the narrate skill into your coding agent. Then just point it at code:

"Narrate the payment flow in src/payments/" "Create a diagram story explaining how the sync engine works" "Help me explain the request lifecycle to a new developer"

The agent reads your code, writes the D2 diagram and .story.yaml, runs diascope build, and hands you the .html.

Install for Claude Code

git clone https://github.com/Biolytics-AI/DiaScope.git ~/.claude/diascope
mkdir -p ~/.agents/skills
ln -s ~/.claude/diascope/skills ~/.agents/skills/diascope

Restart Claude Code. Full instructions: .claude/INSTALL.md

Install for Codex

Tell Codex:

Fetch and follow instructions from https://raw.githubusercontent.com/Biolytics-AI/DiaScope/main/.codex/INSTALL.md

Full instructions: .codex/INSTALL.md

---

Story format reference

The .story.yaml file is the source of truth for all narration. The .d2 file is never modified by DiaScope tooling.

Full reference: docs/story-format.md

Linking steps to nodes

nodes: values must match D2 node names exactly (case-sensitive):

D2 source	nodes: value
Client	Client
System.Client	System.Client
"My Service"	My Service

diascope build warns you if any referenced node ID isn't found in the rendered SVG.

Detail panels and edge tooltips

detail_panels:
  Server: |
    <p>Handles auth and routing. Rate limited: 1000 req/s.</p>

edge_tooltips:
  "POST /api/data": "Authenticated with Bearer token"

---

CLI reference

diascope build <diagram.d2> <story.yaml> [options]

  -o, --out <file>          Output HTML (default: <story>.html)
  --viewer-bundle <path>    Custom path/URL for viewer JS bundle

diascope init <diagram.d2> [options]

  -o, --out <file>          Output story file (default: <diagram>.story.yaml)

---

Examples

Working examples live in examples/. The current vLLM example includes:

• examples/vLLM/deployment.d2
• examples/vLLM/deployment.story.yaml
• examples/vLLM/README.md

You can also browse the same example in the docs at https://diascope.biolytics.ai/examples/vllm-deployment/.

---

JS library

Embed in a framework or custom shell:

import { DiaScopeViewer } from "@biolytics.ai/diascope";
import { parseStoryFile, storyToViewerOptions } from "@biolytics.ai/diascope/story";

const story = parseStoryFile(yamlString);
const viewer = new DiaScopeViewer({
  ...storyToViewerOptions(story),
  autoBindControls: true,
});
viewer.init();

The viewer expects svg-pan-zoom on window.svgPanZoom (or via options.svgPanZoom). See templates/story.html for a complete working shell.

---

Release process

DiaScope releases are published from GitHub, not from local machines.

1. Open a PR into main.
2. Merge only after build-and-test and build-docs pass.
3. Create and push a version tag like v0.1.0.
4. GitHub Actions publishes @biolytics.ai/diascope to npm and creates the matching GitHub Release.

Local npm publish should only be used for dry-runs and debugging.

---

LLM patching guide

When asking an LLM to update a .story.yaml:

• Add steps by appending to steps[] with a new unique id
• Node IDs come from the .d2 file — never invent them
• body supports inline HTML; use | block scalar for multi-line
• Preserve existing id values — they match # @step annotations in the .d2
• Run diascope build after edits to validate and regenerate