Research Lit

Research Literature Review

Research topic: $ARGUMENTS

Constants

• PAPER_LIBRARY — Local directory containing user's paper collection (PDFs). Check these paths in order:
  1. papers/ in the current project directory
  2. literature/ in the current project directory
  3. Custom path specified by user in CLAUDE.md under ## Paper Library
• MAX_LOCAL_PAPERS = 20 — Maximum number of local PDFs to scan (read first 3 pages each). If more are found, prioritize by filename relevance to the topic.
• SOURCES = all — Which literature sources to search. Options: zotero, obsidian, local, web, semantic-scholar, deepxiv, exa, gemini, openalex, all. Full source table and selection rules: see ## Data Sources below.
• ARXIV_DOWNLOAD = false — When true, download top 3-5 most relevant arXiv PDFs to PAPER_LIBRARY after search. When false (default), only fetch metadata (title, abstract, authors) via arXiv API — no files are downloaded.
• ARXIV_MAX_DOWNLOAD = 5 — Maximum number of PDFs to download when ARXIV_DOWNLOAD = true.

💡 Overrides:

• /research-lit "topic" — paper library: ~/my_papers/ — custom local PDF path
• /research-lit "topic" — sources: zotero, local — only search Zotero + local PDFs
• /research-lit "topic" — sources: web — only search the web (skip all local)
• /research-lit "topic" — sources: web, semantic-scholar — also search Semantic Scholar for published venue papers (IEEE, ACM, etc.)
• /research-lit "topic" — sources: all, deepxiv — use default sources plus DeepXiv
• /research-lit "topic" — arxiv download: true — download top relevant arXiv PDFs
• /research-lit "topic" — arxiv download: true, max download: 10 — download up to 10 PDFs

Data Sources

This skill checks multiple sources in priority order. All are optional — if a source is not configured or not requested, skip it silently.

Source Selection

Parse $ARGUMENTS for a — sources: directive:

• If — sources: is specified: Only search the listed sources (comma-separated). Valid values: zotero, obsidian, local, web, semantic-scholar, deepxiv, exa, gemini, openalex, all.
• If not specified: Default to all — search every available source in priority order (semantic-scholar, deepxiv, exa, gemini, and openalex are excluded from all; they must be explicitly listed).

Examples:

/research-lit "diffusion models"                                    → all (default, no S2)
/research-lit "diffusion models" — sources: all                     → all (default, no S2)
/research-lit "diffusion models" — sources: zotero                  → Zotero only
/research-lit "diffusion models" — sources: zotero, web             → Zotero + web
/research-lit "diffusion models" — sources: local                   → local PDFs only
/research-lit "topic" — sources: obsidian, local, web               → skip Zotero
/research-lit "topic" — sources: web, semantic-scholar              → web + S2 API (IEEE/ACM venue papers)
/research-lit "topic" — sources: deepxiv                            → DeepXiv only
/research-lit "topic" — sources: all, deepxiv                       → default sources + DeepXiv
/research-lit "topic" — sources: all, semantic-scholar              → all + S2 API
/research-lit "topic" — sources: exa                               → Exa only (broad web + content extraction)
/research-lit "topic" — sources: all, exa                          → default sources + Exa web search
/research-lit "topic" — sources: gemini                            → Gemini only (AI-powered broad discovery)
/research-lit "topic" — sources: all, gemini                       → default sources + Gemini discovery
/research-lit "topic" — sources: gemini, semantic-scholar           → Gemini + S2 (broad discovery + venue metadata)
/research-lit "topic" — sources: openalex                          → OpenAlex only (open citation graph + institutions)
/research-lit "topic" — sources: semantic-scholar, openalex         → S2 + OpenAlex (complementary metadata)

Source Table

Graceful degradation: If no MCP servers are configured, the skill works exactly as before (local PDFs + web search). Zotero and Obsidian are pure additions.

Workflow

Step 0a: Search Zotero Library (if available)

Skip this step entirely if Zotero MCP is not configured.

Try calling a Zotero MCP tool (e.g., search). If it succeeds:

1. Search by topic: Use the Zotero search tool to find papers matching the research topic
2. Read collections: Check if the user has a relevant collection/folder for this topic
3. Extract annotations: For highly relevant papers, pull PDF highlights and notes — these represent what the user found important
4. Export BibTeX: Get citation data for relevant papers (useful for /paper-write later)
5. Compile results: For each relevant Zotero entry, extract:
   • Title, authors, year, venue
   • User's annotations/highlights (if any)
   • Tags the user assigned
   • Which collection it belongs to

📚 Zotero annotations are gold — they show what the user personally highlighted as important, which is far more valuable than generic summaries.

Step 0b: Search Obsidian Vault (if available)

Skip this step entirely if Obsidian MCP is not configured.

Try calling an Obsidian MCP tool (e.g., search). If it succeeds:

1. Search vault: Search for notes related to the research topic
2. Check tags: Look for notes tagged with relevant topics (e.g., #diffusion-models, #paper-review)
3. Read research notes: For relevant notes, extract the user's own summaries and insights
4. Follow links: If notes link to other relevant notes (wikilinks), follow them for additional context
5. Compile results: For each relevant note:
   • Note title and path
   • User's summary/insights
   • Links to other notes (research graph)
   • Any frontmatter metadata (paper URL, status, rating)

📝 Obsidian notes represent the user's processed understanding — more valuable than raw paper content for understanding their perspective.

Step 0c: Scan Local Paper Library

Before searching online, check if the user already has relevant papers locally:

1. Locate library: Check PAPER_LIBRARY paths for PDF files

   Glob: papers/**/*.pdf, literature/**/*.pdf
   

2. De-duplicate against Zotero: If Step 0a found papers, skip any local PDFs already covered by Zotero results (match by filename or title).

3. Filter by relevance: Match filenames and first-page content against the research topic. Skip clearly unrelated papers.

4. Summarize relevant papers: For each relevant local PDF (up to MAX_LOCAL_PAPERS):

   • Read first 3 pages (title, abstract, intro)
   • Extract: title, authors, year, core contribution, relevance to topic
   • Flag papers that are directly related vs tangentially related

5. Build local knowledge base: Compile summaries into a "papers you already have" section. This becomes the starting point — external search fills the gaps.

📚 If no local papers are found, skip to Step 1. If the user has a comprehensive local collection, the external search can be more targeted (focus on what's missing).

Step 1: Search (external)

• Use WebSearch to find recent papers on the topic
• Check arXiv, Semantic Scholar, Google Scholar
• Focus on papers from last 2 years unless studying foundational work
• De-duplicate: Skip papers already found in Zotero, Obsidian, or local library

arXiv API search (runs when — sources: is unset, contains web or all; no download by default — arXiv API is part of the Priority-4 Web tier, see Source Table above):

Policy D2 tracking discipline (orchestrator-managed): the executor (you, the LLM) maintains an in-context list of contributing sources. For helper-backed bash sources (arxiv, semantic-scholar, deepxiv, exa, openalex), a source contributes iff its bash block ran its helper successfully (helper resolved AND invocation exited 0; note: the helper exiting 0 with an empty result list still counts as "ran" — downstream relevance ranking is what decides whether the user actually sees content). For non-helper sources (zotero / obsidian / local PDF / WebSearch / Gemini), the contribution rule is stated in the Step-1 finalization block below — these are tracked separately because they don't emit D2 contribution: log lines from bash. Sources that were not requested via — sources: do not count. At the end of Step 1 (before "Optional PDF download"), if zero sources contributed, surface a D2 empty-aggregate error and stop. (See integration-contract.md §2 Policy D2 — the in-context tracking replaces a shared bash accumulator because SKILL bash blocks are executed as separate shells; state does not survive.)

Resolve $ARXIV_FETCHER via the canonical chain (Policy D2 — this source contributes to the multi-source aggregate; warn-and-continue on failure, never abort the whole aggregate):

# Canonical strict-safe resolver (see shared-references/integration-contract.md §2).
cd "$(git rev-parse --show-toplevel 2>/dev/null || pwd)" || exit 1
if [ -z "${ARIS_REPO:-}" ] && [ -f .aris/installed-skills.txt ]; then
    ARIS_REPO=$(awk -F'\t' '$1=="repo_root"{print $2; exit}' .aris/installed-skills.txt 2>/dev/null) || true
fi
ARXIV_FETCHER=".aris/tools/arxiv_fetch.py"
[ -f "$ARXIV_FETCHER" ] || ARXIV_FETCHER="tools/arxiv_fetch.py"
[ -f "$ARXIV_FETCHER" ] || { [ -n "${ARIS_REPO:-}" ] && ARXIV_FETCHER="$ARIS_REPO/tools/arxiv_fetch.py"; }
[ -f "$ARXIV_FETCHER" ] || ARXIV_FETCHER=""

if [ -n "$ARXIV_FETCHER" ]; then
  # Search arXiv API for structured results (title, abstract, authors, categories).
  # Wrap with if/then/else so set -e doesn't abort the SKILL.
  if python3 "$ARXIV_FETCHER" search "QUERY" --max 10; then
    echo "D2 contribution: arxiv (helper invocation exit 0)" >&2
  else
    echo "WARN: arxiv_fetch.py invocation failed; D2 aggregate continues with WebSearch results." >&2
  fi
else
  echo "WARN: arxiv_fetch.py not resolved; falling back to WebSearch for arXiv hits." >&2
fi

Record-keeping: track the D2 contribution: … lines emitted by each source's bash block. They form the contributing-source list the orchestrator uses for the Step-1 finalization gate below. WebSearch (Priority 4) is treated as having contributed iff WebSearch was requested (no — sources: filter, or the list contains web or all) AND was actually invoked; the orchestrator records that separately. (The finalization block below restates this rule canonically — both lines must stay in sync.)

If $ARXIV_FETCHER is empty (D2 graceful degradation), fall back to WebSearch for arXiv (same as before).

The arXiv API returns structured metadata (title, abstract, full author list, categories, dates) — richer than WebSearch snippets. Merge these results with WebSearch findings and de-duplicate.

Semantic Scholar API search (only when semantic-scholar is in sources):

When the user explicitly requests — sources: semantic-scholar (or — sources: web, semantic-scholar), search for published venue papers beyond arXiv:

# Re-resolve $ARIS_REPO (SKILL bash blocks may run in separate shells).
cd "$(git rev-parse --show-toplevel 2>/dev/null || pwd)" || exit 1
if [ -z "${ARIS_REPO:-}" ] && [ -f .aris/installed-skills.txt ]; then
    ARIS_REPO=$(awk -F'\t' '$1=="repo_root"{print $2; exit}' .aris/installed-skills.txt 2>/dev/null) || true
fi
# Resolve $S2_FETCHER (Policy D2 — warn-and-skip on missing).
S2_FETCHER=".aris/tools/semantic_scholar_fetch.py"
[ -f "$S2_FETCHER" ] || S2_FETCHER="tools/semantic_scholar_fetch.py"
[ -f "$S2_FETCHER" ] || { [ -n "${ARIS_REPO:-}" ] && S2_FETCHER="$ARIS_REPO/tools/semantic_scholar_fetch.py"; }
[ -f "$S2_FETCHER" ] || S2_FETCHER=""

if [ -n "$S2_FETCHER" ]; then
  # Search for published CS/Engineering papers with quality filters.
  # Wrap with if/then/else so set -e doesn't abort the SKILL.
  if python3 "$S2_FETCHER" search "QUERY" --max 10 \
      --fields-of-study "Computer Science,Engineering" \
      --publication-types "JournalArticle,Conference"; then
    echo "D2 contribution: semantic_scholar (helper invocation exit 0)" >&2
  else
    echo "WARN: semantic_scholar_fetch.py invocation failed; D2 aggregate continues with remaining sources." >&2
  fi
fi

If $S2_FETCHER is empty (canonical chain exhausted), skip silently — D2 multi-source aggregate continues with the remaining resolved sources.

Why use Semantic Scholar? Many IEEE/ACM journal papers are NOT on arXiv. S2 fills the gap for published venue-only papers with citation counts and venue metadata.

De-duplication between arXiv and S2: Match by arXiv ID (S2 returns externalIds.ArXiv):

• If a paper appears in both: check S2's venue/publicationVenue — if it has been published in a journal/conference (e.g. IEEE TWC, JSAC), use S2's metadata (venue, citationCount, DOI) as the authoritative version, since the published version supersedes the preprint. Keep the arXiv PDF link for download.
• If the S2 match has no venue (still just a preprint indexed by S2): keep the arXiv version as-is.
• S2 results without externalIds.ArXiv are venue-only papers not on arXiv — these are the unique value of this source.

DeepXiv search (only when deepxiv is in sources):

When the user explicitly requests — sources: deepxiv (or includes deepxiv in a combined source list), use the DeepXiv adapter for progressive retrieval:

# Re-resolve $ARIS_REPO (SKILL bash blocks may run in separate shells).
cd "$(git rev-parse --show-toplevel 2>/dev/null || pwd)" || exit 1
if [ -z "${ARIS_REPO:-}" ] && [ -f .aris/installed-skills.txt ]; then
    ARIS_REPO=$(awk -F'\t' '$1=="repo_root"{print $2; exit}' .aris/installed-skills.txt 2>/dev/null) || true
fi
# Resolve $DEEPXIV_FETCHER (Policy D2 — warn-and-skip on missing).
DEEPXIV_FETCHER=".aris/tools/deepxiv_fetch.py"
[ -f "$DEEPXIV_FETCHER" ] || DEEPXIV_FETCHER="tools/deepxiv_fetch.py"
[ -f "$DEEPXIV_FETCHER" ] || { [ -n "${ARIS_REPO:-}" ] && DEEPXIV_FETCHER="$ARIS_REPO/tools/deepxiv_fetch.py"; }
[ -f "$DEEPXIV_FETCHER" ] || DEEPXIV_FETCHER=""

if [ -n "$DEEPXIV_FETCHER" ] && command -v deepxiv >/dev/null 2>&1; then
  # Wrap each adapter call so set -e doesn't abort the SKILL.
  if python3 "$DEEPXIV_FETCHER" search "QUERY" --max 10; then
    echo "D2 contribution: deepxiv (helper invocation exit 0)" >&2

    # Then deepen only for the most relevant papers (sub-calls don't change D2 aggregate count):
    python3 "$DEEPXIV_FETCHER" paper-brief ARXIV_ID \
      || echo "WARN: deepxiv_fetch.py paper-brief failed; skipping deepen step." >&2
    python3 "$DEEPXIV_FETCHER" paper-head ARXIV_ID \
      || echo "WARN: deepxiv_fetch.py paper-head failed; skipping deepen step." >&2
    python3 "$DEEPXIV_FETCHER" paper-section ARXIV_ID "Experiments" \
      || echo "WARN: deepxiv_fetch.py paper-section failed; skipping deepen step." >&2
  else
    echo "WARN: deepxiv_fetch.py search invocation failed; D2 aggregate continues with remaining sources." >&2
  fi
fi

If $DEEPXIV_FETCHER is empty or the deepxiv CLI is unavailable, skip this source gracefully and continue with the remaining requested sources (Policy D2 graceful degradation).

Why use DeepXiv? It is useful when a broad search should be followed by staged reading rather than immediate full-paper loading. This reduces unnecessary context while still surfacing structure, TLDRs, and the most relevant sections.

De-duplication against arXiv and S2:

• Match by arXiv ID first, DOI second, normalized title third
• If DeepXiv and arXiv refer to the same preprint, keep one canonical paper row and record deepxiv as an additional source
• If DeepXiv overlaps with S2 on a published paper, prefer S2 venue/citation metadata in the final table, but keep DeepXiv-derived section notes when they add value

Exa search (only when exa is in sources):

When the user explicitly requests — sources: exa (or includes exa in a combined source list), use the Exa tool for broad AI-powered web search with content extraction:

# Re-resolve $ARIS_REPO (SKILL bash blocks may run in separate shells).
cd "$(git rev-parse --show-toplevel 2>/dev/null || pwd)" || exit 1
if [ -z "${ARIS_REPO:-}" ] && [ -f .aris/installed-skills.txt ]; then
    ARIS_REPO=$(awk -F'\t' '$1=="repo_root"{print $2; exit}' .aris/installed-skills.txt 2>/dev/null) || true
fi
# Resolve $EXA_FETCHER (Policy D2 — warn-and-skip on missing).
EXA_FETCHER=".aris/tools/exa_search.py"
[ -f "$EXA_FETCHER" ] || EXA_FETCHER="tools/exa_search.py"
[ -f "$EXA_FETCHER" ] || { [ -n "${ARIS_REPO:-}" ] && EXA_FETCHER="$ARIS_REPO/tools/exa_search.py"; }
[ -f "$EXA_FETCHER" ] || EXA_FETCHER=""

if [ -n "$EXA_FETCHER" ]; then
  # Search for research papers with highlights.
  # Wrap with if/then/else so set -e doesn't abort the SKILL.
  exa_contributed=false
  if python3 "$EXA_FETCHER" search "QUERY" --max 10 --category "research paper" --content highlights; then
    exa_contributed=true
  else
    echo "WARN: exa_search.py research-paper invocation failed; D2 aggregate continues." >&2
  fi
  # Search for broader web content (blogs, docs, news)
  if python3 "$EXA_FETCHER" search "QUERY" --max 10 --content highlights; then
    exa_contributed=true
  else
    echo "WARN: exa_search.py broad-web invocation failed; D2 aggregate continues." >&2
  fi
  [ "$exa_contributed" = "true" ] && echo "D2 contribution: exa (at least one invocation exit 0)" >&2
fi

If $EXA_FETCHER is empty or the exa-py SDK is unavailable, skip this source gracefully and continue with the remaining requested sources (Policy D2 graceful degradation).

Why use Exa? Exa provides AI-powered search across the broader web (blogs, documentation, news, company pages) with built-in content extraction. It fills a gap between academic databases (arXiv, S2) and generic WebSearch by returning richer content with each result.

De-duplication against arXiv, S2, and DeepXiv:

• Match by URL first, then normalized title
• If Exa returns an arXiv paper already found by arXiv/S2, prefer the structured metadata from those sources
• Exa results from non-academic domains (blogs, docs, news) are unique value not covered by other sources

Gemini search (only when gemini is in sources):

When the user explicitly requests — sources: gemini (or includes gemini in a combined source list), use Gemini for AI-powered broad literature discovery.

Priority 1 — Gemini MCP (preferred): Call mcp__gemini-cli__ask-gemini with the search prompt:

mcp__gemini-cli__ask-gemini({
  prompt: 'You are a research literature scout. Search comprehensively for papers on: "QUERY"

IMPORTANT CONSTRAINTS:
1. Search from MULTIPLE angles — decompose the topic into sub-problems, aliases, neighboring tasks, and common benchmark/settings variants.
2. Prefer papers that are genuinely relevant, not merely keyword-adjacent.
3. Include top venues, journals, surveys, recent preprints, and papers with code when available.
4. Focus on papers from 2022 onward unless older foundational work is necessary.

For EACH paper found, provide ALL of the following:
- Title: [exact title]
- Authors: [full author list]
- Year: [publication year]
- Venue: [exact conference/journal name + year, or "arXiv preprint"]
- arXiv ID: [format 2401.12345, or "N/A"]
- DOI: [if available, or "N/A"]
- Code URL: [GitHub/GitLab link if available, or "No code"]
- Summary: [one-sentence core contribution]

Find at least 15 papers.',
  model: 'auto-gemini-3'
})

Priority 2 — Gemini CLI fallback (if MCP unavailable): Use gemini -p "...same prompt..." 2>/dev/null via Bash (timeout: 120s).

If both MCP and CLI are unavailable, skip this source gracefully and continue with the remaining requested sources.

Why use Gemini? Gemini provides AI-driven discovery that goes beyond keyword matching — it decomposes topics, explores naming variants, and surfaces papers that traditional API-based searches (arXiv, S2) may miss. It fills a different retrieval niche from structured database queries.

De-duplication against arXiv, S2, DeepXiv, and Exa:

• Match by arXiv ID first, DOI second, normalized title third
• If Gemini returns a paper already found by S2, prefer S2's citation count and venue metadata
• If Gemini returns a paper already found by arXiv, prefer arXiv's structured metadata
• Gemini's unique value is discovering papers that other keyword-based indexes did not surface
• Do not use Gemini-reported citation counts — they may be inaccurate. Use S2 for authoritative citation data.

OpenAlex search (only when openalex is in sources):

When the user explicitly requests — sources: openalex (or includes openalex in a combined source list), use OpenAlex API for comprehensive academic metadata:

# Re-resolve $ARIS_REPO (SKILL bash blocks may run in separate shells).
cd "$(git rev-parse --show-toplevel 2>/dev/null || pwd)" || exit 1
if [ -z "${ARIS_REPO:-}" ] && [ -f .aris/installed-skills.txt ]; then
    ARIS_REPO=$(awk -F'\t' '$1=="repo_root"{print $2; exit}' .aris/installed-skills.txt 2>/dev/null) || true
fi
# Resolve $OPENALEX_FETCHER (Policy D2 — warn-and-skip on missing).
OPENALEX_FETCHER=".aris/tools/openalex_fetch.py"
[ -f "$OPENALEX_FETCHER" ] || OPENALEX_FETCHER="tools/openalex_fetch.py"
[ -f "$OPENALEX_FETCHER" ] || { [ -n "${ARIS_REPO:-}" ] && OPENALEX_FETCHER="$ARIS_REPO/tools/openalex_fetch.py"; }
[ -f "$OPENALEX_FETCHER" ] || OPENALEX_FETCHER=""

# Preflight: skip OpenAlex silently if the helper is unresolved OR the
# `requests` Python package is missing. Both checks must pass before
# the script is invoked, so users without `requests` installed never see
# a stack trace from a default `/research-lit` run.
if [ -z "$OPENALEX_FETCHER" ] || ! python3 -c "import requests" >/dev/null 2>&1; then
  echo "OpenAlex source not available (openalex_fetch.py unresolved or 'requests' module missing); skipping." >&2
else
  # Search for papers with comprehensive metadata.
  # Wrap with if/then/else so set -e doesn't abort the SKILL.
  if python3 "$OPENALEX_FETCHER" search "QUERY" --max 10 \
      --year "2022-" \
      --type article \
      --sort relevance; then
    echo "D2 contribution: openalex (helper invocation exit 0)" >&2
  else
    echo "WARN: openalex_fetch.py invocation failed; D2 aggregate continues with remaining sources." >&2
  fi
fi

If openalex_fetch.py is not found or requests module is missing, skip this source gracefully and continue with the remaining requested sources.

Why use OpenAlex? Fully open citation graph (no API key required), institutional affiliations, funding data (NSF, NIH), comprehensive topic/keyword metadata, and coverage across all disciplines (not just CS).

De-duplication against arXiv, S2, DeepXiv, Exa, and Gemini:

• Match by DOI first (OpenAlex has DOI for most works), then arXiv ID, then normalized title
• If OpenAlex and S2 both have the same paper:
  • Prefer S2 for citation counts (more up-to-date)
  • Prefer S2 for venue metadata (more accurate for CS/AI papers)
  • Use OpenAlex for institutional affiliations and funding data (unique value)
  • Merge both into a richer record
• If OpenAlex and arXiv overlap, prefer arXiv's PDF link and metadata, but keep OpenAlex's citation/institution data
• OpenAlex's unique value: institutional affiliations, funding sources, comprehensive topic classification, and cross-discipline coverage

D2 aggregate finalization (per integration-contract §2 Policy D2):

The orchestrator (you, the LLM) maintains an in-context list of contributing sources by reading the D2 contribution: <name> log lines emitted by each source's bash block above, plus:

• zotero if Step 0a returned non-empty Zotero hits.
• obsidian if Step 0b returned non-empty Obsidian hits.
• local if Step 0c found at least one relevant local PDF.
• web if WebSearch (Priority 4) was requested (either no — sources: filter, or the list contains web or all) AND was actually invoked. Note: — sources: all covers the default-on tier (zotero, obsidian, local, web) — it does NOT include the opt-in fetchers (semantic-scholar, deepxiv, exa, gemini, openalex). To enable those, add them explicitly (e.g. — sources: all, semantic-scholar, openalex), matching the existing convention at L42-43 / L51-63 of this SKILL.
• gemini if Gemini MCP / CLI returned at least one paper.

If the resulting contributing-source list has zero entries, surface:

ERROR: D2 aggregate empty — every requested source either was unresolved, not invoked, failed, or (for MCP / local PDF / Gemini sources) returned no usable result. (Note: WebSearch contributes when requested and invoked, even if the result set is empty.) The multi-source aggregate cannot proceed. Suggest the user retry with a wider — sources: list (e.g. web, local) or check helper resolution and SDK installation.

Then stop before Step 1.5. Otherwise log the contributing-source list to the user (e.g. "Sources contributed: arxiv, semantic_scholar, web") and proceed.

Optional PDF download (only when ARXIV_DOWNLOAD = true):

After all sources are searched and papers are ranked by relevance:

# Re-resolve $ARXIV_FETCHER (SKILL bash blocks may run in separate shells).
cd "$(git rev-parse --show-toplevel 2>/dev/null || pwd)" || exit 1
if [ -z "${ARIS_REPO:-}" ] && [ -f .aris/installed-skills.txt ]; then
    ARIS_REPO=$(awk -F'\t' '$1=="repo_root"{print $2; exit}' .aris/installed-skills.txt 2>/dev/null) || true
fi
ARXIV_FETCHER=".aris/tools/arxiv_fetch.py"
[ -f "$ARXIV_FETCHER" ] || ARXIV_FETCHER="tools/arxiv_fetch.py"
[ -f "$ARXIV_FETCHER" ] || { [ -n "${ARIS_REPO:-}" ] && ARXIV_FETCHER="$ARIS_REPO/tools/arxiv_fetch.py"; }
[ -f "$ARXIV_FETCHER" ] || ARXIV_FETCHER=""

# Download top N most relevant arXiv papers; skip silently if helper unresolved.
[ -n "$ARXIV_FETCHER" ] && python3 "$ARXIV_FETCHER" download ARXIV_ID --dir papers/

• Only download papers ranked in the top ARXIV_MAX_DOWNLOAD by relevance
• Skip papers already in the local library
• 1-second delay between downloads (rate limiting)
• Verify each PDF > 10 KB

Step 1.5: Verify Candidate Papers (anti-hallucination, mandatory)

Before analysis, run pre-search verification on all candidate papers collected from Steps 0a-1 to filter out LLM-fabricated arXiv IDs / DOIs / titles. Helper: verify_papers.py (canonical name; resolved per shared-references/integration-contract.md §2, Policy D1 — primary helper with degraded-output fallback). If the helper is unresolved on this machine, the SKILL emits a fallback verified_papers.json tagging every candidate [UNVERIFIED] so downstream analysis proceeds with audit-visible degraded output rather than silently dropping candidates.

# 1. Resolve $VERIFY_PAPERS via the canonical strict-safe chain (§2).
cd "$(git rev-parse --show-toplevel 2>/dev/null || pwd)" || exit 1
if [ -z "${ARIS_REPO:-}" ] && [ -f .aris/installed-skills.txt ]; then
    ARIS_REPO=$(awk -F'\t' '$1=="repo_root"{print $2; exit}' .aris/installed-skills.txt 2>/dev/null) || true
fi
VERIFY_PAPERS=".aris/tools/verify_papers.py"
[ -f "$VERIFY_PAPERS" ] || VERIFY_PAPERS="tools/verify_papers.py"
[ -f "$VERIFY_PAPERS" ] || { [ -n "${ARIS_REPO:-}" ] && VERIFY_PAPERS="$ARIS_REPO/tools/verify_papers.py"; }
[ -f "$VERIFY_PAPERS" ] || VERIFY_PAPERS=""

# 2. Emit candidates as JSON. Verification scratch lives under .aris/
#    (NOT under research-wiki/ — Step 6's wiki ingest predicate is
#    "research-wiki/ exists", and we must not trip it from Step 1.5).
mkdir -p .aris/verify-papers
cat > .aris/verify-papers/candidate_papers.json <<'JSON'
[
  {"id": "p1", "arxiv_id": "2307.03172", "doi": null, "title": "Lost in the Middle"},
  {"id": "p2", "arxiv_id": null, "doi": "10.1145/...", "title": "..."},
  {"id": "p3", "arxiv_id": null, "doi": null, "title": "Some Paper Title"}
]
JSON

# 3. Run 3-layer verification (arXiv batch → CrossRef → Semantic Scholar fuzzy).
#    Policy D1: when the helper is unresolved OR its invocation fails, emit
#    a degraded verified set tagging everything [UNVERIFIED] so the user
#    can audit search quality. If python3 itself is missing, we BLOCK
#    rather than hand-roll JSON in shell.
verify_ok=false
if [ -n "$VERIFY_PAPERS" ]; then
  if python3 "$VERIFY_PAPERS" \
        --input  .aris/verify-papers/candidate_papers.json \
        --output .aris/verify-papers/verified_papers.json; then
    verify_ok=true
  else
    echo "WARN: verify_papers.py invocation failed (resolved at $VERIFY_PAPERS); falling back to [UNVERIFIED] tagging." >&2
  fi
else
  echo "WARN: verify_papers.py not resolved at .aris/tools/, tools/, or \$ARIS_REPO/tools/." >&2
  echo "      Fix: rerun bash tools/install_aris.sh, export ARIS_REPO, or copy the helper to tools/." >&2
fi
if [ "$verify_ok" = "false" ]; then
  if ! command -v python3 >/dev/null 2>&1; then
    echo "ERROR: python3 unavailable; cannot emit fallback verified_papers.json." >&2
    echo "       Status: BLOCKED. Install python3 or restore the helper to proceed." >&2
    exit 1
  fi
  echo "      Emitting unverified candidate set with [UNVERIFIED] tags." >&2
  python3 - <<'PY'
import json
cands = json.load(open('.aris/verify-papers/candidate_papers.json'))
out = {
  'verdict': 'WARN',
  'reason_code': 'verify_papers_unavailable',
  'summary': 'verify_papers.py helper unresolved or invocation failed; all candidates tagged [UNVERIFIED] for audit visibility.',
  'papers': [dict(p, status='unverified', method='none') for p in cands],
}
with open('.aris/verify-papers/verified_papers.json', 'w') as f:
  json.dump(out, f, indent=2)
PY
fi

# 4. Read verdict + per-paper status from .aris/verify-papers/verified_papers.json;
#    surface warnings to the user.

Mandatory output rules (see shared-references/citation-discipline.md § Pre-Search Verification Protocol for the full contract):

• Tag every paper in the analyzed list with its status: ✅ verified (via arxiv|crossref|s2) or ⚠️ UNVERIFIED (reason) or … verify_pending.
• Never silently drop unverified papers — keep them in the output with the [UNVERIFIED] marker so the user can audit the search quality.
• Never fabricate a DOI or arXiv ID from memory. If a field is unknown, leave it null in candidate_papers.json — the helper will fall through to title search.
• If the helper returns WARN with high_hallucination_rate, surface the warning verbatim and recommend re-running with narrower queries.
• For papers tagged verify_pending, do not promote them to verified — show the pending state to the user and retry on the next session.

Optional: set ARIS_VERIFY_EMAIL=you@institution.edu in your shell to lift CrossRef rate limits to the polite pool.

Step 2: Analyze Each Paper

Fan-out (Tier-aware). Per-paper extraction is pure breadth — each paper is independent — so it parallelizes cleanly. Tier 1 (Workflow): spawn one Claude subagent per paper (or per small batch) to extract the fields below. Tier 2 (Agent tool, no Workflow): the same per-paper subagents via the Agent tool. Tier 3: iterate sequentially. This follows the extraction shard schema from shared-references/fan-out-pattern.md — {shard_id: "<paper-or-batch id>", entries: [{dedup_key: "<canonical arXiv-id / DOI / title-hash, already assigned upstream in Step 1.5>", problem, method, results, relevance, source, verification_status}]}.

The "jury" here is not a model — it is the deterministic verify_papers.py gate already run in Step 1.5 (3-layer arXiv / CrossRef / Semantic Scholar cross-check). Because the acceptance gate is a deterministic verifier, not a model verdict, the cross-model-family rule is automatically satisfied (a process is not a model family — see acceptance-gate.md), so this is the near-zero-risk corner of the fan-out design space. The per-paper work is extraction, not adjudication: shards report what each paper says and its verification status verbatim; they do not decide which papers "count" (Step 1.5 already did, mechanically) and they do not drop a paper for any status other than verified. Synthesis (Step 3) is interpretive aggregation — grouping by theme, spotting gaps our work could fill — over an already-admitted set; it is the executor's normal job, NOT an accept/reject verdict on whether a paper counts. The cross-model-family rule governs admission verdicts, and here admission is the deterministic Step-1.5 gate, so the invariant is satisfied without a model jury.

For every paper in .aris/verify-papers/verified_papers.json (verified, unverified, verify_pending, and error alike — see Retention rule above), extract:

• Problem: What gap does it address?

• Method: Core technical contribution (1-2 sentences)

• Results: Key numbers/claims

• Relevance: How does it relate to our work?

• Source: Where we found it (Zotero/Obsidian/local/web) — helps user know what they already have vs what's new

• Verification status (one of):

  • ✅ verified (via arxiv|crossref|s2)
  • ⚠️ UNVERIFIED (verification unavailable: helper unresolved or invocation failed)
  • ⚠️ UNVERIFIED (searched: not found in any source)
  • … VERIFY_PENDING (transient API failure — retry next session)
  • ❌ ERROR (malformed input: no arxiv, no DOI, no title)

  Show the status in the analyzed table — never silently drop a paper because its status is anything other than verified.

Step 3: Synthesize

• Group papers by approach/theme
• Identify consensus vs disagreements in the field
• Find gaps that our work could fill
• If Obsidian notes exist, incorporate the user's own insights into the synthesis

Step 4: Output

Present as a structured literature table:

| Paper | Venue | Method | Key Result | Relevance to Us | Source |
|-------|-------|--------|------------|-----------------|--------|

Plus a narrative summary of the landscape (3-5 paragraphs).

If Zotero BibTeX was exported, include a references.bib snippet for direct use in paper writing.

Step 5: Save (if requested)

• Save paper PDFs to literature/ or papers/
• Update related work notes in project memory
• If Obsidian is available, optionally create a literature review note in the vault

Step 6: Update Research Wiki

Required when research-wiki/ exists. Skip entirely (no action, no error) if the directory is absent. Per shared-references/integration-contract.md, this step follows the canonical ingest contract — business logic lives in tools/research_wiki.py, not in this prose.

When research-wiki/ exists, resolve $WIKI_SCRIPT per the canonical chain documented in shared-references/wiki-helper-resolution.md (Variant B — warn-and-skip):

cd "$(git rev-parse --show-toplevel 2>/dev/null || pwd)" || exit 1
ARIS_REPO="${ARIS_REPO:-$(awk -F'\t' '$1=="repo_root"{print $2; exit}' .aris/installed-skills.txt 2>/dev/null)}"
WIKI_SCRIPT=".aris/tools/research_wiki.py"
[ -f "$WIKI_SCRIPT" ] || WIKI_SCRIPT="tools/research_wiki.py"
[ -f "$WIKI_SCRIPT" ] || { [ -n "${ARIS_REPO:-}" ] && WIKI_SCRIPT="$ARIS_REPO/tools/research_wiki.py"; }
[ -f "$WIKI_SCRIPT" ] || {
  echo "WARN: research_wiki.py not found; literature synthesis will be reported but wiki ingest will be skipped. Fix: bash tools/install_aris.sh, export ARIS_REPO, or cp <ARIS-repo>/tools/research_wiki.py tools/." >&2
  WIKI_SCRIPT=""
}

📋 Research Wiki ingest (runs once, at end of research-lit):
   [ ] 1. Predicate: `research-wiki/` exists? If no, skip this step.
   [ ] 2. If $WIKI_SCRIPT empty (helper unreachable), skip the rest of this step
          (the warning above already explains why).
   [ ] 3. For each of the top 8–12 relevant papers (arxiv IDs collected above):
          python3 "$WIKI_SCRIPT" ingest_paper research-wiki/ \
              --arxiv-id <id> [--thesis "<one-line>"] [--tags <t1>,<t2>]
   [ ] 4. For each explicit relationship to an existing wiki entity,
          add an edge:
          python3 "$WIKI_SCRIPT" add_edge research-wiki/ \
              --from "paper:<slug>" --to "<target_node_id>" \
              --type <extends|contradicts|addresses_gap|inspired_by|...> \
              --evidence "<one-sentence quote or reasoning>"
   [ ] 5. Confirm papers/<slug>.md files were created (helper prints
          "Paper ingested: ..."); if any failed with a network error,
          retry or fall back to the --title/--authors/--year manual form.

ingest_paper handles slug generation, arXiv metadata fetch, dedup (skips an existing paper by arXiv id), page rendering, index.md rebuild, query_pack.md rebuild, and log append in a single call — do not manually write papers/<slug>.md. If the helper is unavailable (e.g., offline on a non-ARIS machine, or $WIKI_SCRIPT empty), log the gap and let `/research

…(truncated)