mnote/skills/mnote-mindmap/SKILL.md

MNote mindmap editing

Use this skill only when the user asks to read, summarize, update, or create an MNote mindmap resource.

Typical triggers:

• The user mentions 思维导图, 脑图, mindmap, KMind, or .mindmap.json.
• The current Page AI target is a mindmap resource tab.
• The current page contains a mindmap embed and the task is about that embed.
• The user asks to turn a PDF, Office document, Markdown page, or pasted material into a mindmap.

Rules:

• First call mnote.context.snapshot or mnote.context.resolve_target when target metadata is needed.
• Prefer the current mindmap resource tab over page embeds when both are present.
• For existing mindmaps, call mnote.mindmap.fetch before proposing or applying changes.
• For new mindmaps from PDF or document summaries, convert the source material into a concise hierarchical outline, then call mnote.mindmap.create_from_outline.
• Do not write a naked simple-mind-map tree as the final file. MNote mindmap files use an envelope with data as the root tree and view as runtime view state.
• Keep the root node uid as root for new maps unless MNote returns a different root.
• Let MNote generate child node uid values; do not invent duplicate ids.
• Before writing, confirm the run has read_write permission and the target resource is inside allowedResourceIds.
• After writing or creating a mindmap, read it back with mnote.mindmap.fetch and report the changed .mindmap.json path.
• If authorization is missing, ask the user to grant access instead of guessing a path.

Current apply_ops contract:

• Supported operations are updateText / updateNode, insertChild / addChild, and deleteNode.
• updateText / updateNode require nodeId (or node_id / id) and text (or title).
• insertChild / addChild require parentId (or parent_id / nodeId) and a child node payload.
• deleteNode requires nodeId (or node_id / id) and must never target the root node.
• Unknown operations are rejected; do not assume extra edit verbs exist.
• Prefer small, reviewable edits. Keep node text concise and split long prose into child nodes.
• Long text should be summarized into short phrases or child nodes rather than embedded as a single paragraph.
• Respect dryRun and revision checks when the caller provides them.

Mindmap outline format:

• Use one central root topic as the title; it should name the subject, not explain it in a paragraph.
• Use first-level children for the major categories or sections radiating from the root.
• Use concise keywords or short phrases for node text. Avoid long paragraphs, full prose summaries, and multiple sentences inside one node.
• Put one idea in each node. Split causes, decisions, examples, and evidence into child nodes instead of joining them with punctuation.
• Keep each level scannable: prefer 3-8 sibling branches unless the source structure requires more.
• Preserve hierarchy with children; do not simulate structure by embedding Markdown lists or newline-separated text in a node.
• Keep citations, page numbers, and source metadata in sourceRefs or nearby metadata when possible, rather than cramming them into visible node text.

PDF or document to mindmap workflow:

1. Identify the source resource and target page.
2. Extract or receive a title plus outline from the source material.
3. Keep the outline small enough to be useful: preserve chapters, decisions, key concepts, and relationships; remove repeated prose.
4. Call mnote.mindmap.create_from_outline with title, outline, sourceRefs, and the target resource path when provided.
5. Verify the created resource by fetching it and checking the root title and first-level branches.