feat(sharepoint): add SharePoint context source support

[Aryan1718]

Summary

This PR adds a new sharepoint context-source driver to ktx. It supports ingesting Markdown and Word documents from a
Microsoft Graph drive folder, covering both SharePoint document libraries and OneDrive folders through the same
drive_id + folder_id model.

The implementation includes connection parsing, credential resolution, Graph-based folder traversal, staged ingest
output, CLI setup and verification flows, runtime skill wiring, documentation, and test coverage.

What was added

New sharepoint context-source driver

Added a new adapter under packages/cli/src/context/ingest/adapters/sharepoint/ with behavior aligned to the existing
ingest stack:

• Supports driver: sharepoint

• Ingests .md and .docx files only

• Traverses a Microsoft Graph drive folder, with optional recursion

• Stages output as:

  • manifest.json
  • per-document metadata.json
  • per-document page.md

• Uses stable compact staged directory names derived from document title and item ID

Microsoft Graph authentication and connection resolution

Added SharePoint connection resolution using Azure AD client credentials:

• Resolves tenant_id_ref, client_id_ref, and client_secret_ref
• Enforces env: references only for these credential fields
• Exchanges credentials against the tenant-specific Microsoft identity token endpoint
• Produces a lean pull config with resolved credentials plus driveId and folderId

Graph-based file discovery and download

Implemented a lightweight Graph client using native fetch:

• Token acquisition
• Folder children pagination
• Recursive traversal of nested folders
• File download for supported document types
• Optional site-to-drive lookup during setup and verification when a site ID is provided

Document conversion and normalization

Added content handling for both supported formats:

• .md files are preserved as authored Markdown with line-ending normalization
• .docx files are converted to Markdown using mammoth
• Converted Markdown is normalized for cleaner headings, lists, and pipe-table output
• A leading H1 is added when converted Word content does not begin with one

CLI and runtime integration

Integrated the new source across user-facing CLI flows:

• ktx connection can verify SharePoint credentials and folder access
• ktx setup and source setup now support SharePoint-specific prompts and flags
• Added runtime skill wiring through sharepoint_synthesize
• Registered the driver in schemas, adapter resolution, public ingest mapping, and local ingest entrypoints

Documentation

Updated docs for configuration and usage:

• docs-site/content/docs/integrations/context-sources.mdx
• docs-site/content/docs/configuration/ktx-yaml.mdx

Also documented required Microsoft Graph application permissions with admin consent:

• Files.Read.All
• Sites.Read.All

Testing

Added targeted coverage for the new SharePoint surface, including:

• schema validation for sharepoint connection config
• env-based credential resolution and error handling
• Graph token acquisition, pagination, traversal, and download behavior
• optional site-drive discovery
• Markdown passthrough and .docx conversion behavior
• staged output layout, manifest contents, and metadata generation
• adapter detection, chunking, and scope description
• setup flow, connection verification, and driver registration

Verification

Verified with:

• targeted Vitest suites for SharePoint ingest, setup, connection, and schema coverage
• pnpm --filter @kaelio/ktx run type-check
• pnpm run dead-code

Result:

• SharePoint-specific implementation and tests passed
• Type-check passed
• Dead-code checks passed

Close #171

[vercel]

@Aryan1718 is attempting to deploy a commit to the Kaelio Team on Vercel.

A member of the Team first needs to authorize it.