ingestlisted

# ingest — weave a whole source into the wiki (the only LLM write step; top-level orchestration) You are the maintainer of the knowledge base. Weave the user's source into the wiki **concept/topic-first**; lessons are an **optional, downgraded** secondary layer (only for continuous teaching/example/exercise stretches that don't sink into concepts) — **named by theme, never `第X章`, never a chapter recap, never "本章/本书/作者" meta-narrative.** The reader should be immersed in the knowledge and never sense the original document. Work under the work-order transaction protocol the whole way. This file is the **top-level orchestration**; load per-phase detail from sibling `references/*` on demand. Project truth: `CLAUDE.md`. Engineering format: `docs/skill-runtime/skill-standard.md`. > **Thin skill + thick CLI:** the execution layer is the deterministic zero-LLM CLI (`scripts/pipeline.py`); > this skill carries no business code, only orchestrates it. `<src>` = this source's source_id; run commands > from the project root with the study-kb interpreter (on Windows: pwsh + `$env:PYTHONUTF8=1`). ## 1. Triggers / Non-triggers - **Triggers:** "add this book/PDF to the KB", "ingest \<source\>", "index this document", "weave this file into the wiki". - **Non-triggers (never fire):** "summarize this", "explain this", "translate this", "answer a trivia question", "what is this PDF about" (a question, not an ingest request). ## 2. Inputs - The user gives: file path `<path>`, domain `<domain>