deep-research

name	deep-research
description	Citation-disciplined literature pipeline backed by paper MCP servers (semantic-scholar, arxiv, paper-search). Use for /research, lit review, fact-check, systematic review.

Deep Research

This skill orchestrates a multi-phase academic research workflow. It is citation-disciplined: every claim in the final output must resolve to a real source reachable through one of the configured paper MCP servers. Hallucinated DOIs are a failure mode, not a typo -- they get caught and rejected here, not after the fact.

When to use which mode

User intent	Mode	Output length
Vague topic, user is exploring	socratic	dialogue only
"Give me a quick summary of X"	quick	500-1,500 words
"Write me a research report on X"	full	3,000-8,000 words
"Find me papers on X" / "annotated bibliography"	lit-review	1,500-4,000 words
"Is this claim true? / Does this paper exist?"	fact-check	300-800 words
"Review this draft / argument"	review	varies
"PRISMA / systematic review of X"	systematic-review	5,000-15,000 words

If the user did not specify, infer from their phrasing. Default to socratic when the question is too broad to research without scoping. See references/mode_selection.md.

How to invoke

User can pass --mode {name} and a topic via slash command (/research --mode lit-review "ion-gated transistors for reservoir computing") or by free text. Read the args; if none, infer from the prompt.

Iron Rules (apply to every mode)

1. No claim without a citation that resolves through Semantic Scholar, OpenAlex, arXiv, or a publisher DOI. If a claim cannot be cited, mark it as [UNVERIFIED] and either drop it or escalate it to the user -- never invent a citation.
2. The devils-advocate agent runs at three checkpoints: after scoping, after synthesis, after composition. It can block progression if claims are weak or sources gray-zone. See agents/devils-advocate.md.
3. The ethics agent can halt delivery when the topic crosses into harm-enablement territory (weapons synthesis, targeted harassment, etc.). See agents/ethics.md.
4. Socratic mode never gives direct answers -- only guiding questions. If the user demands an answer, exit Socratic mode explicitly and re-enter via a different mode.
5. Vault writes go to 00_Inbox/, not 30_Literature/. The user curates and promotes. This matches the user's feedback_obsidian_writes.md rule.
6. Capture every paper you actually read. Whenever an agent fetches and reads a paper's full text (not just the abstract), call the paper-capture skill so it lands in 30_Literature/. Do not skip this -- the whole point of the pack is that papers consumed during research are reusable on the next project.

Phases (every mode runs through these in order)

1. Scoping -- sharpen the research question, choose the mode, write a methodology blueprint.
2. Scope confirmation (mandatory pause; see Scope Confirmation Protocol below). Skip ONLY in quick and fact-check modes.
3. Investigation -- search the MCP servers in priority order (see below), download and read promising papers. Spawn the parallel investigator sub-agents listed under "Spawning sub-agents".
4. Analysis -- synthesize across sources, run risk-of-bias on each, build a claim/evidence table.
5. Composition -- draft the deliverable in the mode-appropriate format.
6. Review -- editorial pass + ethics check + devil's advocate final pass.
7. Citation pre-flight (mandatory; see Citation Pre-flight Protocol below).
8. Hand-off -- write to vault 00_Inbox/research-{slug}-{YYYY-MM-DD}.md via the obsidian MCP.

Skip phases that don't apply (e.g. fact-check skips composition).

Scope Confirmation Protocol (mandatory pause after Phase 1)

In every mode EXCEPT quick and fact-check, after Phase 1 (Scoping) you MUST present the scope back to the user and wait for confirmation. Do not start Phase 3 (Investigation) until the user replies.

Format:

=== SCOPE CONFIRMATION ===

Before I search, here's how I'm reading your request:
- Topic:        {one sentence}
- Year range:   {YYYY-YYYY or "open"}
- Languages:    {English / multilingual / etc.}
- Inclusions:   {peer-reviewed lab, preprints OK, ...}
- Exclusions:   {blog posts, retracted papers, ...}
- Depth:        {quick / lit-review / full / systematic-review / review}
- Time budget:  {e.g. 5-15 min for a quick run}

Reply 'go' to proceed, or correct any of the above. I'll wait.

If the user's original request was genuinely vague (no topic, no scope at all), don't synthesize a scope -- drop into socratic mode instead.

Citation Pre-flight Protocol (mandatory before Phase 8 Hand-off)

After Composition (Phase 5) and the Review pass (Phase 6), and before writing the final note to the vault:

1. Walk every in-text citation in the draft. Extract the DOI / arXiv id / Semantic Scholar paperId for each.
2. For each citation, call mcp__semantic-scholar__get_semantic_scholar_paper_details (or mcp__paper-mcp__paper_get_metadata as fallback).
3. Compare the returned title/authors against what the draft claims:
   • Confirmed: title + first-author both match -> mark [verified] internally.
   • Mismatch: title doesn't match -- the draft cited the wrong DOI. Either fix the DOI (re-search Semantic Scholar) or replace the in-text reference with [UNVERIFIED -- DOI mismatch].
   • 404 / not-found: re-resolve via title-author-year search; if that also fails, replace with [UNVERIFIED -- could not re-confirm].
4. Surface the count of unverified to the user in the final delivery line: Captured N citations; M re-verified, K flagged unverified.
5. If K > 0, list each unverified citation in a ## Unverified Citations section at the bottom of the draft, with what was attempted.

Pre-flight is the integrity gate that catches the rare case where a sub-step of the pipeline (or the model itself) introduced a fabricated or mistyped DOI.

MCP server priority order

The investigator MUST try these in the order below. The order is biased toward broad-source coverage and institutional full-text access -- arXiv is only authoritative for preprints and is missing most journal-published work (Nature, Science, ACS Nano, Adv. Materials, Applied Physics Letters, most of EE/MatSci, biology, chemistry). Do NOT default to arXiv-only searches.

For SEARCH (broad coverage):

1. semantic-scholar -- best for metadata, paper resolution, citation graphs, abstract preview. Spans all journals + preprints.
2. paper-search -- multi-source unified search (arXiv, bioRxiv, medRxiv, PubMed, Google Scholar). Casts the widest net for preprints AND PMC open-access journal papers.
3. paper-mcp -- secondary metadata + DOI resolution path. Useful when a venue isn't well-indexed by Semantic Scholar.
4. arxiv -- only when the topic is explicitly preprint-relevant (recent CS / physics / quant-bio that hasn't been formally published yet, OR seminal preprints from those fields). Don't make arxiv the primary source for an applied-physics or experimental-biology topic -- you'll miss the journal corpus.

For FULL-TEXT DOWNLOAD (after a paper has been chosen):

5. university-paper-access -- institutional download path via Unpaywall + on-campus IP. First choice for full-text PDF of any journal paper. Often returns a legal full-text where the user has institutional access.
6. arxiv (download_paper) -- for arXiv ids only. Free, reliable, but only covers what's posted to arXiv.
7. paper-search (download_*) -- per-source download from bioRxiv/medRxiv/PubMed/PMC.
8. scihub -- last-resort PDF when institutional access fails. Note legal/network limitations.

Always prefer abstract-level reasoning before downloading PDFs. Only download a paper when an agent has decided it actually needs the full text. Then capture it via the paper-capture skill (Iron Rule 6).

Spawning sub-agents

Use the Agent tool with subagent_type=Explore for searches and subagent_type=general-purpose for synthesis/composition. Cap at 3 parallel agents. Each agent prompt lives in agents/ -- load the file with Read and pass its body as the agent prompt, prepended with the specific topic.

For a full run, the recommended spawn pattern is:

• Phase 1: 1x scoping agent (sequential).
• Phase 2: 3x investigator agents in parallel -- one per top-level subtopic. Each may make several MCP calls.
• Phase 3: 1x synthesizer + 1x bias-auditor in parallel, sharing the Phase 2 output.
• Phase 3.5: 1x devils-advocate (checkpoint 1).
• Phase 4: 1x composer (sequential, large context).
• Phase 5: 1x editor + 1x ethics + 1x devils-advocate (checkpoint 2 & 3) in parallel.

For quick, run scoping -> 1x investigator -> 1x composer -> 1x devils-advocate. Skip bias auditor.

For socratic, never spawn agents. Stay in the main loop and ask questions until the user has a researchable question, then propose a mode.

Mode-specific instructions

Read the mode file before starting:

• modes/full.md
• modes/quick.md
• modes/lit-review.md
• modes/fact-check.md
• modes/socratic.md
• modes/systematic-review.md
• modes/review.md

Templates

Use these as the skeleton for the deliverable:

• templates/report_full.md
• templates/report_lit_review.md
• templates/prisma_flow.md
• templates/claim_evidence_table.md

Citation format

APA 7.0 in-text and reference list. For each cited source:

• DOI mandatory if it exists; arXiv id otherwise; URL only if neither.
• Mark provenance: [SS] Semantic Scholar / [XV] arXiv / [UPA] university-access / [SH] Sci-Hub. This trail is what makes the integrity gate falsifiable.

Hand-off note format

The vault note dropped in 00_Inbox/ MUST include:

---
type: research-output
mode: {mode}
topic: {topic}
date: {YYYY-MM-DD}
status: draft           # draft -> reviewed -> promoted
papers_captured: [{citekey1}, {citekey2}, ...]
---

followed by the body and a final ## Provenance section listing every MCP call made and every paper downloaded. Reviewers (you in three months, or a collaborator) should be able to reproduce the search.

Iron Rule reference

Read references/iron_rules.md at the start of every run. If you're not sure whether something violates a rule, stop and ask the user.