@arnilo/prism-work
v0.12.1
Published
Prism work connectors, Office document capabilities, and bounded document-reader subpaths.
Maintainers
Readme
@arnilo/prism-work
Prism work connectors, Office document capabilities, and bounded document-reader adapters. Import only a subpath; this package has no activating root barrel.
| Subpath | Contents |
| --- | --- |
| @arnilo/prism-work/connectors | Microsoft 365 and Google Workspace adapters, work tools, drafts, approvals, idempotency, and limits. |
| @arnilo/prism-work/connectors/microsoft365 | Microsoft 365 CLI adapter. |
| @arnilo/prism-work/connectors/google-workspace | Google Workspace CLI adapter. |
| @arnilo/prism-work/connectors/drafts | Draft stores and approval helpers. |
| @arnilo/prism-work/documents | Prism Document Model, OOXML generate/parse/patch/preview, and bounded fidelity reporting. |
| @arnilo/prism-work/sheets | CSV/XLSX ingest, exact-decimal safety, and bounded schema inference. |
| @arnilo/prism-work/diagrams | draw.io embed model, canonical XML, origin validation, and bounded postMessage protocol. |
| @arnilo/prism-work/document-extraction | Optional local Markdown conversion (@firecrawl/anydoc). Scanned PDFs fail closed unless the host runs the packaged Docling worker. No hosted OCR. |
| @arnilo/prism-work/document-reader | Bounded PDF/DOCX literal-text extraction; optional parser peers fail closed. |
| @arnilo/prism-work/sandbox | Host-pinned work sandbox image digest and createWorkComposition. |
| @arnilo/prism-work/skills | loadWorkSkills() — vendored Hermes MIT docx/xlsx/powerpoint/pdf. |
| @arnilo/prism-work/tools | createOfficeTools filesystem/artifact office tools. |
Install
bun add @arnilo/prism @arnilo/prism-workimport { createWorkTools } from "@arnilo/prism-work/connectors";
import { generateDocument } from "@arnilo/prism-work/documents";Dependencies and trust boundaries
- Office APIs are in-memory only: no filesystem, network, or
process.envaccess. mammothandpdf-parseare optional peers of/document-reader; it refuses creation if selected defaults are unavailable.@firecrawl/[email protected]is an optional peer of/document-extraction. Conversion stays local; the subpath never passes hosted OCR options or reads an API key. Docling OCR is a host-provisioneddocling/ocr.pyworker, not an npm dependency.createDocumentIngestadapts an extractor to the wiki hook and a RAG parser without importing the memory package.- Caps fail closed before output; embedded document content is not fetched or executed.
