菜单
Execute the two-phase NotebookLM-to-Gemini corpus summary chain using source evidence sidecars and Gemini Pro model routing.
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。
Upload pending PDFs to the single NotebookLM corpus notebook and archive successful uploads.
Run the local validation checks for scripts, JSON, Gemini skills, and generated markdown.
Verify the generated corpus summary against the prepared NotebookLM notebook, source-evidence.md, optional targets.md, .workflow_state.json, and prompt2.txt.
| name | sequential-chaining |
| description | Execute the two-phase NotebookLM-to-Gemini corpus summary chain using source evidence sidecars and Gemini Pro model routing. |
Phase 1: Query the prepared corpus NotebookLM notebook with prompt1.txt, the processed-source manifest, and targets.md when present:
nlm notebook query <notebook-id> "$(cat prompt1.txt)
CORPUS CONTEXT:
Processed source files:
- input/processed/<source>.pdf
TARGETS_MD_HARD_SCOPE:
<targets.md content or none>"
Save the NotebookLM output to output/source-evidence.md. The evidence should include a citation inventory with NotebookLM markers, page references, source labels, or other available citation detail.
Phase 2: Refine the NotebookLM evidence with Gemini CLI Pro routing:
gemini --model pro --prompt="$(cat prompt2.txt)" < source-evidence-input.txt
The workflow script also accepts --model <model> or GEMINI_MODEL=<model> when an explicit Gemini model ID is required. The final markdown must remain grounded in the NotebookLM evidence. If targets.md exists, treat it as the complete allowed scope and do not expand into unrelated corpus material. Final summaries must use numbered inline citations such as [1] and end with ## References.