OneNote (Beta)

View saved version →

---
name: export-onenote-to-markdown
description: Export, back up, archive, download, or migrate personal Microsoft OneNote pages, sections, or notebooks to downloadable Markdown files. Use for OneNote-to-Markdown and OneNote-to-Obsidian requests, including bounded multi-page exports.
---

# Export OneNote to Markdown

1. Establish the exact personal page, section, or notebook to export. Explain before a large scan that the current workflow returns one Markdown file per page; it does not create a ZIP, write directly to an Obsidian vault, embed images or attachments, rewrite OneNote links as Obsidian wikilinks, or export shared group/site pages.
2. Keep the source read-only. During export, do not invoke `create_page`, `update_page`, `append_to_page`, `copy_page_to_section`, or `delete_page`. `list_notebooks` is a read-only action even though Microsoft Graph assigns it the connector's `Notes.Create` scope. Treat any request to delete originals as a separate consequential workflow requiring confirmation after export verification.
3. Inventory one batch before exporting. Resolve the notebook with `list_notebooks`; recursively walk `list_section_groups` and `list_sections`; enumerate selected sections with `list_pages`. Collect no more than 20 pages in one turn, requesting only the remaining batch size. Retain the current section and `next_skip` cursor when the cap is reached instead of continuing discovery, along with every observed page ID, title, parent path, and source URL.
4. Call `export_page_as_markdown` once for each page in the current inventory batch and preserve every returned file reference. Do not replace the deterministic export with model-generated Markdown from `get_page_text` or raw `fetch_page` HTML.
5. Treat `resources_not_included`, `unsafe_links_removed`, `complex_table_structure_simplified`, and `unsupported_content_omitted` warnings as partial-fidelity results. The complex-table warning means nested tables or merged cells were flattened because GitHub-Flavored Markdown cannot represent their structure exactly. If the user needs a personal-page image or attachment separately, inspect that page with `fetch_page`, then call `fetch_page_resource` with both that page ID and the exact resource ID observed in its current HTML. State that the returned resource is separate and is not linked from the Markdown file.
6. Reconcile the inventory against exported, partial, and failed page counts. Report every failure and warning, return all observed file references, and never say the migration is complete when a page failed or a pagination window remains.
7. For a large notebook, finish the current bounded batch and offer to continue with the remaining observed page IDs. Do not claim durable checkpoint or resume support based only on chat history.

Use a result summary shaped like:

```text
Exported: <count>
Partial: <count>
Failed: <count>
Remaining: <count or unknown>
Files: <returned Markdown file references>
Warnings: <page-specific fidelity limitations>
```

Referenced files: 1

View saved version →

---
name: onenote
description: Use when asked to find, read, summarize, create, organize, update, export, back up, archive, or migrate Microsoft OneNote notebooks, sections, pages, tables, checklists, and shared group or site notes; also for copying a known page, attaching files to a personal page, or deleting a personal page.
---

# OneNote

- For requests to export, back up, archive, download, or migrate OneNote content to Markdown or Obsidian, use [Export OneNote to Markdown](../export-onenote-to-markdown/SKILL.md). The live `export_page_as_markdown` action exports one personal page at a time as a downloadable Markdown file; do not promise a ZIP, direct Obsidian-vault write, embedded resources, or shared-page export.
- Start with the narrowest discovery path. Use `list_notebooks`, then `list_sections` and `list_section_groups` for a named personal notebook; use `list_pages` for a known section or a bounded recent-page window. Match returned IDs, titles, parent metadata, and web URLs rather than guessing opaque IDs from a browser URL.
- Use `search_pages_by_title` only for case-insensitive title matching. For a body-content request, use `search_pages_by_text` with the narrowest known section and date window, then continue with `next_skip` only when broader coverage is needed. It scans at most ten owner-verified personal pages per call, searches sanitized visible text rather than raw HTML, and is not a global native index; say when pagination or truncation leaves the search incomplete.
- Prefer `get_page_text` for visible paragraph-like text, checklist tags, and addressable blocks; use `get_page_tables` for table structure, `find_page_text` for one exact case-sensitive occurrence in one known page, and `fetch_page` when the original bounded HTML is needed. Use `fetch_page_resource` only with both the page ID and exact resource ID observed in that page's current content.
- Raw `fetch_page` HTML can contain non-visible provider markup; do not surface it or choose it as an update target, and rely on visible structured extractors for user-visible content.
- Personal-only discovery, read, export, hierarchy, and update actions operate only on notebooks owned by the signed-in Microsoft account; they intentionally omit directly shared personal notebooks. Treat the mixed `create_page` and `copy_page_to_section` actions as shared-capable. For a Microsoft 365 group or SharePoint site notebook, require a known `resource_type` and `resource_id`, then use `list_shared_notebooks`, `list_shared_sections`, `list_shared_section_groups`, `list_shared_pages`, and `fetch_shared_page`. Do not pass shared-page IDs to personal-only page actions or imply a shared structured extractor that is not present in the live tool surface.
- Before a targeted edit, read a fresh snapshot with element IDs or structured update targets. Use `update_page` or `update_shared_page` for exact insert, replace, prepend, or append operations, and re-read after any mutation because targets can become stale.
- Use `append_to_page` or `append_to_shared_page` for simple additive notes. Use `add_page_resources` on a personal page for staged images, PDF printouts, or attachments, and keep the supplied filenames and MIME types accurate.
- Use `create_notebook`, `create_section_group`, and `create_section` only in the destination the user selected; `create_page` can target a personal section or a known shared section. Use simple OneNote-compatible HTML for new content, and disambiguate similarly named notebooks, sections, or pages before writing.
- Treat page copying as asynchronous: call `copy_page_to_section`, retain its returned operation reference, and poll `get_operation` until it reaches a terminal state. Never report the copy as complete from the initial accepted response.
- Use `next_skip` for bounded pagination; never invoke or reconstruct an opaque `next_link` URL.
- Read-only requests do not authorize edits. Confirm the exact target and intended change before a consequential write, and require explicit confirmation immediately before `delete_page` because deletion is permanent.

Referenced files: 3