Chengeng KB Curation L0-L5

Problem

If you have project docs, course decks, asset folders, question-bank exports, and scattered PDFs, the hard part is often not “making better summaries” but turning them into a retrievable, traceable, reusable knowledge system. This skill is scoped to a knowledge-base case, not a single PDF or meeting-notes summary. It helps teams assess where the current corpus sits across L0-L5 and then add directories, source audits, object schemas, retrieval tests, and acceptance checks.

How It Works

• Layer judgment before action: the default output starts with input validation, a one-sentence verdict, knowledge-base type, L0-L5 maturity assessment, standard directory suggestions, database and object schema recommendations, source-priority strategy, retrieval test plan, acceptance checklist, and next steps.
• Source traceability first: each knowledge object should retain at least source_ref, document_id, and library_path, so answers can be traced back to the source document.
• Treat intermediate artifacts carefully: PDFs, images, audio, and video are treated as asset registration before OCR or transcript caches exist; the model should work from evidence_packet, object summaries, and sampled original text rather than raw unregistered files.
• L5 as the completion bar: a version is complete only after unified query, source traceability, source priority, synonym handling, idempotence, and cost logging have been validated.

Boundaries

• Not for summarizing a single PDF, recording, screenshot, or Markdown file.
• Not for producing only a summary, reading note, or meeting minutes.
• It does not promise that all materials are already usable, nor does it treat low-confidence OCR or poor-quality transcripts as reliable knowledge objects.
• It should not output API keys, tokens, or account credentials, and it should not invent database tables, object counts, or retrieval results.