Menu
Hybrid vector + full-text search across the document store with hierarchical context.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
Generate a Word document from ingested source documents using the agnostic Document tree (CommonMark/Pandoc-aligned)
Generate a single Document (Word/PowerPoint) from all source documents in a collection, with attributions back to each source
Generate a PDF from ingested source documents using the agnostic Document tree
Generate a PowerPoint deck from ingested source documents using the agnostic Document tree (parametrised by slide_count, style, audience, tone)
Generate an Excel workbook (.xlsx) from ingested source documents — tabular with charts, formulas, KPI cells
Find PII (emails, phones, SSN, credit cards, IBANs, names, addresses, DOB) in an ingested document; return redaction plan
| name | document-search |
| description | Hybrid vector + full-text search across the document store with hierarchical context. |
| inputs | [{"name":"query","type":"string","required":true,"description":"Natural-language question or keywords to search for."},{"name":"top_k","type":"integer","required":false,"description":"Maximum number of chunks to return. Defaults to 8."},{"name":"collection_id","type":"string","required":false,"description":"Optional collection scope."}] |
| outputs | [{"name":"results","type":"array","description":"Ranked list of {chunk_id, document_id, document_title, chapter_title, section_title, score, snippet}."}] |
| tools | ["vector-search","db-query"] |
Pick this skill when the user asks a semantic question that may be answered by indexed
content. Prefer it over document-extract for retrieval; prefer it over document-summarize
when the user wants citations rather than a synthesis.
Document.summary and Document.topics to shortlist candidate documents.Chunk.embedding column via vector-search.Chapter.title and Section.title via db-query.collection_id scoping if provided.Emits no ExecutionHistory rows (read-only). Each call may still produce a structured log line for observability.