ontology-absorb-confluence

# Absorb a wiki page through an agent This is not a native Confluence integration. Atlas never connects to Confluence. A third-party wiki MCP already registered by the user reads the page; Atlas then classifies the returned Markdown with `absorb_document`. The same path works for Notion or an internal wiki that can return Markdown or HTML text. ## Preconditions - A user-registered wiki MCP is available in this session. An existing registration suffices; do not ask the user to register it again. Atlas does not bundle or configure it and never receives account credentials. - If no wiki MCP exists, ask for pasted prose and use `/ontology-extract`. - Skip personal drafts, transient notes, and pages already represented by a vault document. ## 1. Read the page, never write the wiki Call the wiki MCP's page-read tool and record the exact title and URL. Do not call its write tools. `absorb_document` accepts a local file path, not raw text. Save the returned body to a temporary file such as: ```text .ontology-atlas/wiki-import/<page-slug>.md ``` Pass an absolute path when MCP cwd may differ from the shell cwd. Files outside the repo remain blocked unless a reviewed dry-run is followed by explicit `allowOutsideRepo: true`. ## 2. Dry-run classification ```text absorb_document({ filePath: ".ontology-atlas/wiki-import/<page-slug>.md" }) ``` Omitting `confirm` writes nothing. Inspect every row in `sections[]`: | Action | Meaning | |---|---| | `absorb` | policy, convention...