---

name: knowledge-base description: "Use this sub-skill for EverOS knowledge document upload, topic taxonomy, knowledge search, provider gates, and knowledge storage troubleshooting." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

EverOS Knowledge Base

Use this sub-skill for EverOS knowledge documents and topics: uploading source files, replacing documents, patching metadata, listing documents/categories, reading topics, searching knowledge, and diagnosing knowledge provider gates.

Read/run map

• Read knowledge API for endpoint shapes, multipart fields, search request fields, and the provider gate contract.
• Read storage and taxonomy for category/document/topic layout, .taxonomy.md, _original/, and Markdown source-of-truth behavior.
• Read troubleshooting for missing embedding/rerank, parser, unsupported format, oversized upload, and downgrade scenarios.
• Run knowledge_api_probe.py against a running server. It defaults to safe read probes; writes require --allow-write and a file.

Core routes

All routes are under /api/v2/knowledge for new integrations:

Route	Capability
POST /documents	Upload and extract a knowledge document.
PUT /documents/{doc_id}	Replace a document atomically.
PATCH /documents/{doc_id}	Update title/category metadata.
DELETE /documents/{doc_id}	Delete a document.
GET /documents	Paginated document list.
GET /documents/{doc_id}	Document detail with topics.
GET /topics/{topic_id}	Full topic detail.
GET /categories	Taxonomy categories.
POST /search	Keyword/vector/hybrid knowledge retrieval.

Non-negotiable gate facts

Knowledge writes (POST, PUT) and knowledge search require both embedding and rerank capability. Reads, deletes, and metadata patch remain reachable if a deployment downgrades from Tier 3 to Tier 1/2. This lets users inspect, rename, recategorize, and clean up existing documents even when providers are no longer configured.

Plain UTF-8 Markdown/text/RST files can bypass the parser path. PDF, HTML, Office, image, and other non-text formats may require parser and multimodal LLM support; Office formats also need LibreOffice.