菜单
Search and query the ForgeRAG engineering knowledge graph. Ask questions about materials, welding, standards, formulas, reference tables, and specifications — get answers with page citations from ASM handbooks, ASME codes, NFPA, IEEE, and other engineering references.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | knowledge-graph |
| description | Search and query the ForgeRAG engineering knowledge graph. Ask questions about materials, welding, standards, formulas, reference tables, and specifications — get answers with page citations from ASM handbooks, ASME codes, NFPA, IEEE, and other engineering references. |
| version | 1.2.0 |
| author | system |
| tools | ["ask_engineering_question","find_relevant_chunks","search_engineering_docs","smart_search","get_forgerag_status","query_knowledge_graph","explore_entity","list_knowledge_collections"] |
| dependencies | [] |
Search engineering handbooks, standards, and specifications. The knowledge graph contains materials, processes, standards, equipment, formulas, reference tables, and their relationships extracted from ingested PDF documents — with paragraph-level structural chunking, LLM-generated chunk summaries, RRF hybrid retrieval (BGE-M3 dense + BM25 + cross-encoder reranker), fuzzy entity matching, OCR typo tolerance, community weighting, and a circuit breaker for reliability.
| Tool | Use when… |
|---|---|
ask_engineering_question | The user wants a synthesized answer with citations. Uses RRF hybrid + VLM reading of page images. Primary tool. |
find_relevant_chunks | You need precise evidence to quote (a specific paragraph, table, or equation). Returns raw chunk text + summary without the VLM synthesis step. Faster than ask_engineering_question. |
search_engineering_docs (mode="keyword") | Looking up a specific code, alloy designation, clause number, or standard (C12000, QW-451.1, NFPA 70, SEMI S2). Supports fuzzy: true for OCR typo tolerance. |
search_engineering_docs (mode="visual") | Finding pages by visual similarity — charts, tables, diagrams, schematics. Uses Nemotron ColEmbed. |
smart_search | General-purpose entry point when unsure which mode is best. Auto-detects: codes/designations -> keyword, questions -> answer, else -> hybrid/RRF. |
get_forgerag_status | Check ForgeRAG capabilities, live stats (documents, pages, entities, communities), and service health before searching. |
query_knowledge_graph | How entities relate to each other — what standards govern a material, what materials are compatible with a process, which standards cross-reference. |
explore_entity | Everything connected to one specific entity (N-hop neighborhood). |
list_knowledge_collections | Discovery — what engineering databases are available. |
ask_engineering_question (narrative answer)find_relevant_chunks (precise quotation)find_relevant_chunks with chunk_type="table"find_relevant_chunks with chunk_type="equation", then ask_engineering_question if more context neededsearch_engineering_docs with mode="visual" (page images)smart_search (auto-routes to the best strategy)search_engineering_docs with mode="keyword" and fuzzy: true (OCR typo tolerance)get_forgerag_status (live stats and capabilities)ask_engineering_question is the primary synthesis tool — RRF hybrid retrieval + VLM reading pages + graph context → answer with [Page N] citationsfind_relevant_chunks returns paragraph/table-precise results with LLM summaries — use it for precise quoting or when you want to inspect the raw evidence before synthesizingexplore_entity to discover cross-references and relationshipscode (e.g., "SEMI S2") and a title (e.g., "Environmental, Health, and Safety Guideline for Semiconductor Manufacturing Equipment") — queries resolve either formtopic_tags classify content topically (tap-drill-chart, fastener-torque, conductor-ampacity, etc.) — helpful for filteringControls smart home devices and reads sensor data via Home Assistant
Lets a Choom queue its OWN future followup. The Choom writes a short prompt to itself; the bridge fires it as a one-shot heartbeat at the scheduled time. Use to self-trigger proactive check-ins, reminders to revisit an unfinished thread, or to wake up after an expected interval.
Scrapes webpages for text and images using a headless browser (JS-rendered) or static HTML. Downloads images and files from the web.
Start a live group conversation with one or more sister Chooms — a real back-and-forth chat (not a task hand-off). Use to check in, think together, plan, or connect. The user can see and join.
Sandboxed file operations in project workspace — create projects, write, read, list, create folders, delete files, and rename projects. Use for saving reports, code, notes, or managing project files.
Generates images using Stable Diffusion with checkpoint switching, LoRA support, and self-portrait mode. Use when the user requests pictures, artwork, selfies, or portraits.