deep-research

Provide a clear research topic or question, such as "research the impact of AI on higher education." The pipeline runs six phases from scoping through revision, producing an APA 7.0 report with verified sources. Use quick mode for a 500-1500 word brief instead.

Three-way-scan gives a fast WHY/HOW/WHAT comparison of a paper shortlist. Lit-review produces an annotated bibliography plus thematic synthesis. Systematic-review runs a full PRISMA 2020 pipeline with risk of bias assessment, meta-analysis, and GRADE tables.

It triangulates each citation across Semantic Scholar, OpenAlex, Crossref, and arXiv APIs, flags preprints published after the 2024 LLM inflection point, checks retraction status, and screens for predatory journals. Degraded API lookups are recorded as omissions rather than false negatives.

Use Socratic mode when you have a vague interest but no answerable research question. A mentor agent guides you through five questioning layers to converge on a focused question. It never generates candidate questions unless you explicitly request them.

Yes. The research question brief, methodology blueprint, annotated bibliography, and synthesis report transfer to the academic-paper skill, whose intake agent skips redundant steps. A preregistration sidecar artifact can also be carried byte-for-byte into the writing phase.

Devil's advocate checkpoints at three stages can block progression on critical-severity issues like fatal logical flaws. Ethics review stops once to confirm critical integrity concerns such as fabrication or missing AI disclosure, and revision loops are capped at two iterations.