Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Um comando direto ignora o prompt de revisão. Verifique a origem antes de executá-lo.
Notion workspace integration for searching, reading, and managing pages, databases, blocks, comments, users, and files. Also supports exporting Aurora RCA postmortems and action items to Notion databases.
Notion is a REMOTE service. Use ONLY the Notion tools listed below -- never shell commands or local file access.
Instructions
1. Discovery -- Always Start Here
Before any operation, search to understand the workspace layout:
notion_search(query='...', types=['page']) -- find pages by title keyword
notion_search(query='...', types=['database']) -- find databases by title keyword
notion_search(query='...', max_results=25) -- search all object types (pages + databases)
types accepts ["page"], ["database"], or omit for both.
max_results range: 1--100, default 10.
Returns: id, object type, title, url, last_edited_time for each result.
To read a specific page or database:
notion_fetch(url_or_id='<page_url_or_uuid>', max_length=5000) -- returns page body as markdown
Accepts full Notion URLs or bare UUIDs (with or without dashes).
Multiple sorts are applied in order (first sort is primary).
5. Users & People
notion_list_users(max_results=100, start_cursor=None) -- list all workspace members (people + bots).
notion_get_user(target_user_id) -- fetch a single user by ID.
notion_get_self() -- get the Aurora bot user (the integration itself).
notion_find_person(name_or_email) -- find a person:
Email (contains @): exact match.
Name: case-insensitive substring search across up to 20 pages of users.
Returns: found (bool), count, users array.
notion_list_teamspaces() -- best-effort listing of teamspaces (uses workspace search internally). Often empty on Free/Plus plans.
6. Comments
Post a comment:notion_create_comment(text, page_id=None, block_id=None, discussion_id=None)
Provide exactly ONE of: page_id (new top-level comment), block_id (comment on a block), discussion_id (reply to existing thread).
text: plain text (converted to rich_text internally). Supports inline formatting only (bold, italic, code, links) -- no block-level markdown.
Upload a file:notion_upload_file(file_path_or_url, filename=None, content_type=None)
Accepts local file path OR public https URL.
Auto-selects single-part (< 20 MB) or multi-part (10 MB chunks, up to 500 MB).
Content-type auto-detected if not provided.
Returns: upload_id (reference this when creating/updating pages), filename, size_bytes, status.
SSRF protection: URLs resolving to private/loopback IPs are blocked.
List uploads:notion_list_file_uploads(max_results=25, start_cursor=None) -- list recent file uploads by this integration.
8. Custom Emojis
notion_list_custom_emojis() -- list custom emojis in the workspace. Returns supported: false on plans without this feature.
9. Data Sources (Plan-Gated)
These endpoints require specific Notion plans. All return {"supported": false} gracefully on workspaces without access.
notion_create_data_source(payload) -- create a data source (full payload passed through).
notion_get_data_source(data_source_id) -- fetch by ID.
notion_update_data_source(data_source_id, updates) -- update attributes.
notion_update_data_source_properties(data_source_id, properties) -- update schema.
notion_query_data_source(data_source_id, filter=None, sorts=None, max_results=25, start_cursor=None) -- query with filter/sort/pagination.
notion_list_data_source_templates(data_source_id) -- list templates.
10. Database Views (Plan-Gated)
These endpoints require specific Notion plans. All return {"supported": false} gracefully on workspaces without access.
notion_create_view(payload) -- create a view on a database.
notion_update_view(view_id, updates) -- update filters, sorts, visible properties.
notion_delete_view(view_id) -- irreversible deletion of a view.
notion_list_database_views(database_id) -- list all views on a database.
notion_query_view(view_id, filter=None, sorts=None, max_results=25, start_cursor=None) -- query through a view.
11. RCA Postmortem Export (Aurora-Specific)
Export a postmortem:notion_export_postmortem(incident_id, database_id, action_items_database_id=None)
incident_id: UUID of the Aurora incident.
database_id: target Notion database for the postmortem page.
action_items_database_id: optional separate database for action items.
Creates a structured page with title "Postmortem -- Incident {id}", timeline, root cause, impact sections.
Auto-maps incident properties to database columns:
Parses unchecked markdown checkboxes from the postmortem body.
Creates one Notion database row per action item.
assignee_hints: optional map {"item_text_prefix": "user@email.com"} to assign owners.
Auto-detects title, people, and date properties in the target database schema.
Assignees resolved by email lookup against workspace users.
RCA Investigation Workflow
IMPORTANT: You MUST search Notion early in every RCA investigation. The workspace contains past postmortems, runbooks, and action items that provide critical historical context. Do NOT skip this step.
IMPORTANT: During RCA, Notion is READ-ONLY. Do NOT create pages, export postmortems, create action items, or write to Notion. The engineer reviews your findings and exports via the UI after approving.
Step 1 (EARLY -- do this within your first 3 tool calls) -- Search Notion for historical context:notion_search(query='<service_name>', types=['page']) to find past postmortems or runbooks related to the affected service.
Also try: notion_search(query='<alert_type_or_keyword>') for broader matches.
Step 3 -- Read relevant past postmortems:notion_fetch(url_or_id='<page_id>', max_length=10000) to get full postmortem content. Past incidents on the same service often reveal recurring root causes.
Important Rules
Notion is a REMOTE service. Use ONLY the Notion tools listed above -- never local commands.
All url_or_id / page_id / database_id parameters accept UUIDs with or without dashes.
notion_fetch truncates at max_length characters. For full content of long pages, paginate with notion_get_block_children.
Write operations (create, update, append, move, trash, delete) modify the user's live Notion workspace. Confirm destructive actions before executing.
notion_update_database_properties with a null value REMOVES a column and its data. Always confirm.
notion_delete_block is permanent and irreversible from the API.
notion_delete_view is irreversible.
Rich text content is capped at ~2000 characters per segment by Notion.
Pagination: all list/query endpoints return has_more (bool) and next_cursor (string). Pass start_cursor=next_cursor to fetch the next page.
Data sources, views, and custom emojis are plan-gated features. Tools return {"supported": false} on workspaces without access -- do not retry, inform the user.
Rate limiting: the Notion API enforces 3 requests/second. The client handles retries automatically (429 with Retry-After, 5xx with backoff). No manual throttling needed.
Auth errors: if a tool returns code: "reauth_required", ask the user to reconnect Notion at the connection page, then retry.
If a tool returns code: "not_connected", ask the user to connect Notion first.
