hx-docling-advisor

Guides Docling document processing within HX-16 native architecture and smoke-test authority.

Updated Sep 9, 2026
One-click install
npx skills add https://github.com/HX-Infratstructure/HX-Eco-System --skill hx-docling-advisor-hx-infratstructure
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: hx-docling-advisor
Source: https://github.com/HX-Infratstructure/HX-Eco-System/tree/main/skills/docling/hx-docling-advisor
Command: npx skills add https://github.com/HX-Infratstructure/HX-Eco-System --skill hx-docling-advisor-hx-infratstructure

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Applying upstream Docling guidance inside the HX Eco-System risks importing containerized, managed-service, GPU-first, or remote-service defaults that conflict with the owner-approved HX-16 native Linux architecture. This Skill reconciles current official Docling and Docling MCP guidance with HX authority so recommendations never override placement, deployment, or validation rules. ## Core Features & Use Cases - Authority-ordered advisory: Applies a strict authority chain from owner instructions and HX control documents down to upstream Docling documentation, returning labeled recommendations (ACCEPT, ADAPT, REFERENCE_ONLY, REJECT_FOR_HX, OWNER_DECISION_REQUIRED). - Docling and Granite-Docling guidance: Covers CLI and Python SDK conversion, OCR, tables, VLM processing with Granite-Docling 258M under the CPU-first BASE rule, structured extraction, and chunking/RAG preparation boundaries. - Docling MCP companion boundary: Classifies upstream MCP modes, transports, and optional toolsets against HX-16 constraints and the D1/D2 smoke-test proof chain. - Use Case: When planning the HX-16 Docling installation, ask whether containerized docling-serve is acceptable; the Skill rejects it under the native Linux rule and points to the runbook and smoke-test authority instead. ## Quick Start Ask the advisor how to validate Granite-Docling 258M on CPU for the HX-16 D1 smoke test without downloading models at execution time.

Frequently Asked Questions about hx-docling-advisor

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I validate Granite-Docling 258M on CPU with Docling?▼

Run the D1 smoke test, which converts a self-generated local PDF through the normal Docling pipeline and then through the granite_docling VLM path with --device cpu. The Granite model must be staged locally beforehand, and the installed Docling version must be verified to honor the CPU device flag.

Can I use containerized docling-serve in the HX Eco-System?▼

No, containerized Docling Serve is classified REJECT_FOR_HX under the current owner decision requiring native Linux with systemd. Any change to that rule requires an explicit owner decision before execution.

Does Docling support remote VLM or managed services for document processing?▼

Upstream Docling supports remote VLM, remote URL processing, and managed services, but these are OWNER_DECISION_REQUIRED or REJECT_FOR_HX in this environment. Granite-Docling 258M stays on HX-16 and BASE validation uses no remote services.

What is the difference between the D1 and D2 Docling smoke tests?▼

D1 proves deterministic local Docling conversion plus the Granite-Docling CPU VLM path with no live integration. D2 runs only after an accepted D1 PASS and validates the Docling MCP companion through real client negotiation and one safe known-answer tool call.

Why does the Docling CPU device flag get ignored on GPU hosts?▼

Docling v2.101.0 had a bug where --pipeline vlm --device cpu could be ignored on GPU-visible hosts, fixed upstream in June 2026 before v2.126.0. The Skill requires verifying the installed version and observed device during validation rather than trusting documentation alone.

When should Docling RAG and chunking integrations be used in HX?▼

RAG loaders, chunking, and framework integrations are REFERENCE_ONLY during BASE validation. They become relevant only in later integration phases and must never create permanent Qdrant or LightRAG ingestion architecture during the Docling base build.