diagram（diagram 图表绘制）

Diagram

Mission

Generate human-readable .todiagram JSON that is valid for ToDiagram's custom schema and optimized for fast visual comprehension.

Mode Router

1. schema-build: user asks to create/fix .todiagram structure or convert notes into valid nodes/edges.
2. system-architecture: user asks for platform/service/component topology.
3. flow-diagram: user asks for workflow, journey, runbook, or decision flow.
4. detail-calibration: user asks to simplify, enrich, or split dense diagrams.

If a request spans multiple modes, run in this order:

1. schema-build
2. system-architecture or flow-diagram
3. detail-calibration
4. final schema validation

Global Output Contract

• Default to file-first output for stability: create a .todiagram file and return its path.
• Only return inline JSON when the user explicitly asks for inline output.
• Use canonical raw .todiagram keys: nodes, edges, direction, id, label, from, to, imageUrl, iconUrl, custom_theme, fields.
• Put node display data inside the fields object. Direct properties on the node are deprecated by the schema.
• Never combine imageUrl with fields on the same node — imageUrl replaces row rendering.
• Include imageUrl and iconUrl only when used; omit the keys entirely otherwise.
• Include node-level nodes only for true containers with real children.
• Restrict direction to RIGHT, LEFT, UP, or DOWN (UP, not TOP).
• Keep IDs globally unique; ensure every edge endpoint resolves to an existing ID.
• Prefer Iconify specs (logos:*, fa7-solid:*, short forms like fa:gear) for icons; use external URL icons only if the user explicitly requests them.
• When the exact Iconify slug is unknown, look it up via curl -fsSL 'https://api.iconify.design/search?query=<term>&limit=10' (optionally &prefix=logos) and pick a slug from the returned icons array. Use the ToDiagram MCP icon-search tool when available.
• Apply per-node theming via custom_theme (NODE, EDGE, TEXT groups) — see schema contract for full key list.

File Output Policy

• If user provides a target path/name, write there.
• If user does not provide one, derive a short kebab-case filename from the diagram's topic or subject (e.g., "system architecture" → system-architecture.todiagram, "user onboarding flow" → user-onboarding-flow.todiagram, "AWS infrastructure" → aws-infrastructure.todiagram). Write to the current working directory.
• For multi-diagram outputs, use stable descriptive names: overview.todiagram, domain-<name>.todiagram, flow-<name>.todiagram.
• After writing, run: python scripts/validate_todiagram_schema.py <file>.todiagram

Core Workflow

1. Identify audience and decision intent.
2. Select mode and detail tier (overview, working, deep-dive).
3. Draft container hierarchy first, then leaf nodes, then edges.
4. Add concise labels and only high-signal metadata.
5. Validate with references/authoring-checklist.md and run: python scripts/validate_todiagram_schema.py <file>.todiagram
6. If overloaded, split using references/split-strategies.md while preserving shared IDs/names.

Mode Playbooks

1) Schema Build

Read references/schema-contract.md first.

Use this scaffold unless the user provides a different envelope:

{
  "direction": "RIGHT",
  "nodes": [],
  "edges": []
}

Rules:

• Add configuration only when key mappings deviate from defaults.
• Normalize payloads to canonical keys (children → nodes, source/target → from/to).
• Move deprecated direct node properties into the fields object.
• If a node has both imageUrl and fields, drop one — they cannot coexist.

2) System Architecture

Read:

• references/grouping-playbook.md
• references/architecture-patterns.md

Rules:

• Prefer 2-3 hierarchy levels.
• Keep most containers within 3-8 children.
• Connect leaf-to-leaf interactions by default.
• Keep overview diagrams metadata-light.

3) Flow Diagrams

Read:

• references/flow-patterns.md
• references/labeling-guide.md

Rules:

• Build happy path first, then branches/failures/retries.
• Keep one dominant direction per diagram.
• Keep edge labels short (1-3 words) and meaningful.

4) Detail Calibration

Read:

• references/complexity-budgets.md
• references/split-strategies.md

Rules:

• Choose a detail tier before generating.
• If above budget, split by domain, lifecycle, environment, or audience.
• Keep cross-diagram naming and shared IDs consistent.

Example Intents

• "Create a high-level SaaS architecture in ToDiagram." -> system-architecture + overview tier.
• "Fix this invalid ToDiagram JSON." -> schema-build.
• "Turn this incident runbook into a readable flow." -> flow-diagram (+ detail-calibration if dense).

Troubleshooting

• Invalid payload shape: normalize keys and flatten wrappers first.
• Diagram too noisy: remove low-signal metadata and split into focused views.
• Diagram too abstract: add 1-2 critical metadata properties on key leaves.
• Unclear relationships: add concise protocol/action/outcome edge labels.
• Schema validation dependency error: install jsonschema via python3 -m pip install --user jsonschema.

References

• Schema contract: references/schema-contract.md
• Authoring checklist: references/authoring-checklist.md
• Architecture patterns: references/architecture-patterns.md
• Grouping playbook: references/grouping-playbook.md
• Flow patterns: references/flow-patterns.md
• Labeling guide: references/labeling-guide.md
• Complexity budgets: references/complexity-budgets.md
• Split strategies: references/split-strategies.md

Scripts

• Canonical + schema validator: scripts/validate_todiagram_schema.py