Deep research

Template-native research workflow. This skill routes to existing repository systems instead of running an autonomous external agent suite.

Natural invoke

• "Research this topic and build the source corpus"
• "Do a literature review with citation keys"
• "Fact-check these manuscript claims"
• "Guide me from vague research interest to answerable question"
• "Plan a systematic review with inclusion and exclusion criteria"

Inputs to confirm

• Research goal - question, audience, and expected deliverable.
• Mode - full, quick, lit-review, fact-check, guided, or systematic-review.
• Corpus source - arXiv, Crossref, local JSON corpus, provided papers, or project artifacts.
• Project target - if results feed a manuscript, use docs/_generated/active_projects.md.

Workflow

1. Question framing - turn vague topics into scoped research questions, explicit in/out boundaries, and search terms. Guided mode asks before converging.
2. Source discovery - use infrastructure.search.literature and project-local corpora; record query, backend, date, DOI/arXiv IDs, and failures.
3. Source verification - resolve every candidate reference against Crossref / OpenAlex / arXiv with the deterministic reference-existence gate before it enters the corpus; reject fabricated/mismatch records, surface unverifiable/unchecked honestly, and never invent unavailable bibliographic fields. This is the tier-0 anti-leakage step: flag the gap, do not hallucinate the citation.
4. Synthesis - hand off to literature-synthesis for per-paper notes, thematic clusters, contradictions, and gap analysis.
5. Claim bridge - when research supports manuscript prose, create citekey-linked notes that manuscript-claim-verification can audit later.

Deliverables

• Research brief: question, scope, search strategy, corpus table, and evidence gaps.
• Literature synthesis with bracket-key citations and source-tier notes.
• Fact-check report: claim, source, verdict, uncertainty, action.
• For systematic review mode: protocol draft, inclusion/exclusion criteria, screening counts, and limitations.

Verification commands

bash
uv run python -m infrastructure.search.literature search "QUERY" --max-results 20
uv run python -m infrastructure.reference.citation validate projects/<project>/manuscript/references.bib
uv run python -m infrastructure.reference.verification verify projects/<project>/manuscript/references.bib --live --json
uv run python -m infrastructure.validation.cli evidence projects/<project> --fail-on-issues

References

• MODE_REGISTRY.md
• literature-synthesis
• manuscript-claim-verification