agentic-engineering-research

Invoke the research skill with one explicit decision and your questions. It produces a bounded evidence pack containing cited sources, provenance, confidence levels, and documented conflicts, stopping before any specification or implementation work.

An evidence pack is a bounded research deliverable containing decision-ready conclusions with provenance for every claim, confidence ratings, and flagged conflicts between sources. It gives reviewers durable, citable grounding for technical decisions.

Yes. When NotebookLM is appropriate, the skill resolves the bundled core skill map and selects the notebooklm-project-memory capability, maintaining one curated research notebook per project with cited evidence.

Do not use it for specification, implementation, or provider dispatch, since the skill deliberately stops before those phases. It also treats missing provenance or missing permission as a blocker rather than proceeding with ungrounded claims.

Conflicts between sources are documented explicitly in the evidence pack alongside confidence ratings. This lets reviewers see where evidence disagrees instead of receiving a single unverified conclusion.