From research-workspace
Verifies a NotebookLM brief against the uploaded source bundle, flagging missed sources, unsupported claims, contradictions, and overgeneralizations. Supports both research-hub-managed and manual fallback modes.
How this skill is triggered — by the user, by Claude, or both
Slash command
/research-workspace:notebooklm-brief-verifierThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
NotebookLM is great at producing readable briefs, but it can:
NotebookLM is great at producing readable briefs, but it can:
This skill verifies a downloaded brief against the actual source bundle that research-hub uploaded, so the user can trust (or distrust) the brief before sharing or citing.
Trigger phrases:
Not for:
research-hub notebooklm generate.literature-triage-matrix.academic-writing-skills.In priority order:
research-hub-managed mode (default). When the brief was generated
via research-hub notebooklm generate + download:
.research_hub/artifacts/<cluster>/brief-*.txt.research_hub/bundles/<cluster>/manifest.json
— list of which source files were uploaded.raw/<cluster>/*.md — for
spot-checking specific claims.pdfs/<cluster>/ — last-resort spot-check
only; cap at 3 per session.Manual fallback mode (new in v0.68.x). When the user generated the brief themselves on notebooklm.google.com — direct upload, web UI, copy-paste — research-hub never saw the bundle. Accept either CLI flags or a paste-into-chat:
--brief <path-to-brief.{md,txt,pdf}> — the downloaded brief
file (any path, not just .research_hub/artifacts/).--sources <path-to-source-list.{yml,md,json}> — a plain list
of the source titles + DOIs / URLs the user uploaded to NLM.Conversational variant: paste the brief and the source list directly into the chat. The skill should ask explicitly for the source list if missing — do NOT assume coverage without ground truth.
The verification logic (source coverage scan, claim attribution, contradiction scan, overgeneralization scan, spot-check, follow-up prompts) is identical in both modes. Only the input-loading layer differs.
If the user names a brief file directly, prefer that path over guessing.
S_bundle.S_bundle item, search the brief
text for the citation key, DOI, or first-author name. Call any
bundle item with zero hits a "missed source".In-conversation report (no file written by default). The report has 7 sections: source coverage, unsupported claims, cross-source contradictions, potential overgeneralizations, spot-checked claims, recommended follow-up NotebookLM prompts, and verdict (reliable for / use with caution for / do not cite without spot-check).
Full template + worked example: references/report-template.md.
If the brief is well-attributed and bundle coverage is complete, the report is short — that's a feature, not a bug.
.research_hub/artifacts/<cluster>/brief-verify-<ts>.md
optionally if the user says "save this report"..research/ or .paper/ — this is verification, not
workspace setup.references/report-template.md — full 7-section verification report templatenpx claudepluginhub wenyuchiou/ai-research-skills --plugin research-workspaceScores note quality (depth, sourcing, linking, voice, atomicity) and verifies claims against cited sources. Detects URL fabrication, overclaim, stale references, and duplicates. Produces a fix plan for vault notes.
Audits citations and source claims in academic manuscripts. Verifies whether cited papers support attributed claims and checks quantitative claims.
Performs deep critical analysis of a specific paper, dissecting experimental setup, extracting numbers, and evaluating claims against a hypothesis. Useful for arXiv IDs, URLs, or paper titles.