Ma Scout

MA Scout Skill

You are helping a medical researcher discover meta-analysis topics. Two modes are available depending on the starting point.

This skill handles the pre-protocol phase — from idea to ranked topic list. For actual MA execution (PROSPERO, screening, analysis), hand off to /meta-analysis.

Mode Selection

Determine the mode from user input:

If ambiguous, ask the user whether to search by professor (supervisor-first) or by topic (question-first).

Communication Rules

• Communicate with the user in their preferred language (typically Korean).
• Research questions, PICO/PIRD, and README content in English.
• Medical terminology always in English.

Inputs

Mode A: Professor-first

• Professor name (native-language + English)
• Profile URL (ScholarWorks, SKKU Faculty, Google Scholar, ORCID)
• PubMed author link (preferably with cauthor_id for disambiguation)
• Known specialty (e.g., "thoracic imaging", "abdominal imaging")
• Affiliation history (e.g., "Hospital A → Hospital B → retired")
• Minimum required: name + at least one profile URL or PubMed link

Mode B: Topic-first

• Clinical question or keyword (e.g., "AI for lung-nodule malignancy prediction", "dual-energy CT body composition")
• Radiology subspecialty scope (e.g., thoracic, abdominal, neuro)
• MA type preference (DTA, prognostic, intervention — optional)
• Desired role: solo first author / co-first / supervisor-matched
• Minimum required: clinical question or keyword

Workflow

Mode A (Professor-first): Phase 0 → 1 → 2 → 3 → 4 → 5 Mode B (Topic-first): T-Phase 0 → T-1 → T-2 → T-3 → T-4 → T-5 Phase 2 (MA Gap Analysis) and Phase 4 (README template) are shared between both modes.

═══════════════════════════════════════════

MODE A: PROFESSOR-FIRST WORKFLOW

═══════════════════════════════════════════

Phase 0: Disambiguation & Context Confirmation

Goal: Resolve author identity before any search, and confirm user's relationship context.

CRITICAL — Do this BEFORE any PubMed search:

1. Resolve full English name first:

   • If cauthor_id is provided → fetch that specific PMID page to get full name + affiliation
   • NEVER start with initials-only search (e.g., "Ha HK") — common Korean initials cause massive contamination
   • First search must be "[Full Name]"[Author] (e.g., "Ha Hyun Kwon"[Author])

2. Confirm affiliation chain with user:

   • Ask the user whether {detected affiliation} matches the professor's history, and request the user's relationship to the professor so topic proposals can be tuned accordingly.
   • This prevents wrong-institution assumptions
   • Skip only if user already provided explicit affiliation history

3. Profile URL fallback chain (Scopus requires auth, so plan alternatives):

   • 1st: PubMed full name search (always works)
   • 2nd: Google Scholar profile (WebSearch "[Full Name]" radiology scholar)
   • 3rd: ResearchGate profile (WebSearch "[Full Name]" researchgate radiology)
   • 4th: ScholarWorks / SKKU / university faculty page (if URL provided)
   • Last: Scopus/ScienceDirect (often fails due to auth — do NOT rely on it)

Phase 1: Profile Exploration (E-utilities API)

Goal: Identify the professor's 5-6 distinct research pillars using PubMed E-utilities API.

CRITICAL — Use E-utilities API, NOT WebFetch for PubMed:

• Scripts: ~/.claude/skills/search-lit/references/pubmed_eutils.sh + parse_pubmed.py
• Rate limit: 350ms between calls (100ms with NCBI_API_KEY)
• These are faster, more reliable, and return structured data (JSON/XML)

Step 1 — Total publication count + PMID list:

bash ~/.claude/skills/search-lit/references/pubmed_eutils.sh search \
  '"[Full Name]"[Author]' 200 \
  | python3 ~/.claude/skills/search-lit/references/parse_pubmed.py esearch

Step 2 — Fetch metadata for MeSH-based clustering (parallel):

# Get PMIDs from Step 1, then fetch summaries
bash ~/.claude/skills/search-lit/references/pubmed_eutils.sh fetch_json \
  "PMID1,PMID2,..." \
  | python3 ~/.claude/skills/search-lit/references/parse_pubmed.py esummary

Step 3 — Topic-specific counts (launch 4-5 searches in parallel via Bash):

# Run these in parallel Bash calls
bash ~/.claude/skills/search-lit/references/pubmed_eutils.sh search \
  '"[Full Name]"[Author] AND "keyword1"' 5
bash ~/.claude/skills/search-lit/references/pubmed_eutils.sh search \
  '"[Full Name]"[Author] AND "keyword2"' 5
# ... repeat for each suspected pillar keyword

Step 4 — MeSH term extraction for automatic pillar clustering:

# Fetch full XML for top-cited papers to extract MeSH headings
bash ~/.claude/skills/search-lit/references/pubmed_eutils.sh fetch \
  "PMID1,PMID2,...,PMID20" \
  | python3 -c "
import sys, xml.etree.ElementTree as ET
from collections import Counter
root = ET.fromstring(sys.stdin.read())
mesh_counts = Counter()
for article in root.findall('.//PubmedArticle'):
    for mh in article.findall('.//MeshHeading/DescriptorName'):
        mesh_counts[mh.text] += 1
for term, count in mesh_counts.most_common(30):
    print(f'{count:3d}  {term}')
"

→ Top MeSH terms reveal natural research pillars (e.g., "Colonography, Computed Tomographic" = CTC pillar).

Step 5 — Google Scholar profile (parallel with PubMed calls):

• WebSearch: "[Full Name]" radiology scholar google for h-index, citation data

Output: Pillar Summary Table

Phase 2: MA Gap Analysis (Multi-Source)

Goal: For each pillar, determine if a viable MA topic exists using PubMed + Consensus + Scholar Gateway + bioRxiv.

For each pillar (run in parallel using meta-analyst agents):

2a. PubMed E-utilities — Existing MAs + Primary studies

# Existing MAs (structured count)
bash ~/.claude/skills/search-lit/references/pubmed_eutils.sh search \
  '[pillar keywords] AND ("meta-analysis"[pt] OR "systematic review"[pt])' 50

# Primary studies with extractable outcomes
bash ~/.claude/skills/search-lit/references/pubmed_eutils.sh search \
  '[pillar keywords] AND ("sensitivity" OR "specificity" OR "accuracy" OR "prognosis" OR "outcome")' 50

2b. Consensus MCP — Semantic MA gap detection

Use mcp__claude_ai_Consensus__search to find existing SRs/MAs that PubMed keyword search might miss:

query: "systematic review OR meta-analysis [pillar topic] [imaging modality]"

Consensus returns citation-ranked results — check if any highly-cited MA already covers the proposed scope. Limit: max 3 Consensus calls per Phase 2 batch (rate limit).

2c. Scholar Gateway — Semantic similarity search

Use mcp__claude_ai_Scholar_Gateway__semanticSearch for:

• Finding MAs with different terminology (e.g., "pooled analysis" instead of "meta-analysis")
• Detecting scope-overlapping MAs that use different keywords
• Identifying methodological review papers that partially cover the topic

2d. bioRxiv/medRxiv — In-press competition detection

Use mcp__claude_ai_bioRxiv__search_preprints to catch:

• MAs posted as preprints but not yet indexed in PubMed
• Ongoing SR/MA protocols shared as preprints
• Very recent primary studies that could change feasibility

query: "[pillar keywords] meta-analysis OR systematic review"
server: "medrxiv"  (for clinical topics)

2e. Assessment matrix

2f. PROSPERO competition check (MANDATORY)

• Search PROSPERO via WebSearch: site:crd.york.ac.uk/prospero [topic keywords]
• Also try WebFetch: https://www.crd.york.ac.uk/prospero/#searchadvanced
• Look for registered-but-unpublished protocols that could block entry
• If PROSPERO match found → flag as 🚫 competition risk in ranking

2g. Realistic k estimation

• Raw PubMed hit count is NOT the real k — most studies lack 2x2 data or HR
• Apply conservative discount: k_realistic ≈ raw_count × 0.15–0.30 for DTA topics
• Flag if k_realistic < 8 (DTA) or < 6 (prognostic) as ⚠️ feasibility risk
• Report both raw and realistic estimates, e.g., estimated k: ~130 (raw) → ~20–40 (extractable DTA data)

2h. Niche subtopic discovery (if pillar appears saturated)

• AI/radiomics angle on a classical topic
• Specific modality comparison (e.g., CEUS vs MRI)
• Treatment response (vs diagnosis which is often saturated)
• Specific subpopulation or disease subtype
• Use Consensus to check if the niche angle has already been covered

Phase 3: Topic Ranking

Goal: Rank all viable topics by composite score.

Score each candidate on 5 criteria (★1-5):

Output: Ranked Topic Table

Phase 4: Folder & README Scaffolding

Goal: Create project folders and README for each viable topic.

1. Folder location: {working_dir}/ma-scout/{initials}_{professor_name}/

2. Naming convention: {NN}_{topic_slug}/ (within professor folder)

   • Professor folder: {initials}_{name} (e.g., KDK_Kim, LKS_Lee)
   • NN: sequential number within professor (01, 02, ...)
   • topic_slug: English, underscore-separated
   • Check existing folders with ls before creating

3. README.md template (PROSPERO-ready): Load the bilingual template block from ${CLAUDE_SKILL_DIR}/references/project_readme_template.md and copy it into {topic_folder}/README.md. The reference covers both supervised (Mode A) and solo-mode (Mode B, no supervisor) variants and contains the PICO/PIRD frame, preliminary search, target journal table, and backward-planned timeline.

Phase 5: Output Summary

Goal: Persist findings for the user.

1. Save the ranked topic table and README files to the working directory.
2. Summarize: total topics scanned, viable topics found, recommended next steps.
3. Suggest the user save results to their project management system (e.g., /manage-project).

Niche Topic Discovery Heuristics

When all major pillars are saturated (>5 prior MAs), try these angles:

1. "First MA" rule: Professor's most unique/niche subtopic where MA = 0
2. AI/radiomics overlay: Classical imaging topic + AI approach = new MA angle
3. Treatment response: Diagnosis MAs saturated → treatment monitoring MA often open
4. Modality comparison: Head-to-head (e.g., CEUS vs MRI) often underserved
5. Guideline gap: Professor authored guidelines → MA supporting/updating those guidelines
6. Geographic/population niche: Regional population-specific MA (e.g., parasitic diseases, TB)
7. Temporal update: Last MA >5 years old + significant new primary studies since

Quality Gates

Before finalizing a topic as viable:

• Author identity confirmed — full name resolved via E-utilities efetch, no initials-only contamination
• Affiliation confirmed with user (or from reliable source)
• Confirmed MA = 0 or last MA >5 years (via PubMed E-utilities, not assumption)
• Cross-validated via Consensus/Scholar Gateway — no hidden MAs with different terminology
• bioRxiv/medRxiv checked — no preprint MA in progress
• Confirmed k_realistic ≥ 8 (DTA) or ≥ 6 (prognostic) after discount
• PROSPERO searched — no registered competing protocol found
• No 2024-2026 competing MA in press (check PubMed + preprints)
• Professor's publication record demonstrates clear authority in this area
• Research question is specific enough for PROSPERO registration
• README contains: complete PICO/PIRD, PubMed search strategy, Embase draft, target journal with IF, timeline

Handoff

After MA Scout completes:

• To /meta-analysis: When a topic is approved and ready for PROSPERO protocol (README has PICO + search strategy ready)
• To manage-project: When project folder needs full scaffolding
• To search-lit: When deeper preliminary search is needed before committing
• To /analyze-stats: When feasibility requires power/sample-size calculation for the estimated k

Parallel Execution Strategy

For efficiency, launch multiple agents and API calls in parallel:

Phase 0 (Identity):

1. E-utilities esearch: "[Full Name]"[Author] → total count + PMIDs (FIRST)
2. E-utilities efetch: top 20 PMIDs → MeSH terms → automatic pillar clustering

Phase 1 (Profile — all parallel): 3. Bash × 4-5: E-utilities esearch with topic-specific filters (parallel Bash calls) 4. WebSearch: Google Scholar profile 5. WebFetch: any provided profile URLs (skip Scopus)

Phase 2 (MA Gap — multi-source parallel): 6. Up to 4 meta-analyst agents in parallel, each covering 1-2 pillars 7. Each agent runs ALL of:

• E-utilities esearch: existing MA count + primary study count
• Consensus MCP: semantic MA search (max 3 calls total across all agents)
• Scholar Gateway: scope-overlap check
• bioRxiv/medRxiv: preprint MA detection
• PROSPERO: competition check (WebSearch)

8. Each agent reports: raw k, realistic k (15-30% discount), all sources checked

Phase 3 (Ranking): Sequential, uses Phase 2 outputs.

Phase 4 (Scaffolding): Sequential, creates folders + PROSPERO-ready READMEs.

Total (Mode A): 5-8 parallel agents per professor, ~8-12 minutes per professor.

Mode B Parallel Strategy

T-Phase 0: Sequential (user interaction for scope clarification).

T-Phase 1 (Landscape — all angles in parallel):

1. Per angle: Bash (PubMed MA count) + Bash (primary k) + Consensus + bioRxiv + PROSPERO
2. 3-5 angles × 5 sources = 15-25 parallel calls

T-Phase 2 (Deep-dive): Same as Mode A Phase 2, only for viable angles (typically 1-2).

T-Phase 4 (Co-author — if needed): 3. Bash: PubMed author frequency search 4. WebSearch: Google Scholar profiles for top candidates

Total (Mode B): ~5-8 minutes per topic scan (faster than Mode A — no profile exploration).

Known Pitfalls (from 3 professor analyses)

• Common Korean/Asian initials (e.g., "Lee KS", "Kim DK") return 300+ papers with massive contamination. Always use full name first.
• Scopus/ScienceDirect → 403 or redirect to login. Never rely on Scopus as primary data source.
• Raw PubMed counts overestimate by 3-7x. ~130 hits often means 20-40 with extractable DTA data.
• Professor may have moved institutions. Don't assume affiliation without verification.
• Consensus rate limit: Max 3 batch calls. If rate-limited, wait 30s and retry once.
• E-utilities rate limit: 350ms between calls (100ms with NCBI_API_KEY). Scripts handle this automatically.
• bioRxiv MCP: Use server: "medrxiv" for clinical topics, server: "biorxiv" for preclinical.

═══════════════════════════════════════════

MODE B: TOPIC-FIRST WORKFLOW

═══════════════════════════════════════════

T-Phase 0: Topic Clarification & Scope

Goal: Refine the user's clinical question into a searchable, PROSPERO-registrable scope.

1. Parse the input — extract:

   • Disease/condition (e.g., "lung nodule", "hepatocellular carcinoma")
   • Imaging modality or intervention (e.g., "dual-energy CT", "AI CAD")
   • Outcome type: DTA (Se/Sp), prognostic (HR/OR), intervention (RR/MD), dosimetry
   • Population specifics (e.g., "screening setting", "cirrhotic patients")

2. Expand to neighboring angles — propose 3-5 variations:

   user input: "AI for lung nodule malignancy prediction"
   → variant 1: AI vs radiologist for lung nodule malignancy prediction (DTA)
   → variant 2: Radiomics for lung nodule malignancy (DTA)
   → variant 3: Deep learning for incidental pulmonary nodule management (prognostic)
   → variant 4: AI-assisted Lung-RADS upgrade accuracy (DTA)
   → variant 5: Low-dose CT AI for lung cancer screening (DTA)
   

3. User selects 1-3 angles to investigate further.

T-Phase 1: Landscape Scan (Multi-Source)

Goal: For each selected angle, rapidly assess the MA landscape.

Run all angles in parallel. For each angle:

1a. PubMed — Existing MA count

bash ~/.claude/skills/search-lit/references/pubmed_eutils.sh search \
  '[topic keywords] AND ("meta-analysis"[pt] OR "systematic review"[pt])' 50

1b. PubMed — Primary study pool

bash ~/.claude/skills/search-lit/references/pubmed_eutils.sh search \
  '[topic keywords] AND ("sensitivity" OR "specificity" OR "hazard" OR "outcome")' 100

1c. Consensus MCP — Semantic MA discovery

query: "systematic review [topic] [modality]"

Check for MAs using different terminology.

1d. bioRxiv/medRxiv — Preprint competition

query: "[topic] meta-analysis"
server: "medrxiv"

1e. PROSPERO — Registered protocols

WebSearch: site:crd.york.ac.uk/prospero [topic keywords]

Output: Landscape Summary Table

T-Phase 2: Feasibility Deep-Dive

Goal: For viable angles (MA ≤ 2, no PROSPERO conflict), run full gap analysis.

This phase uses the same Phase 2 (MA Gap Analysis) as Mode A — steps 2a through 2h. The only difference: no "Professor fit" to evaluate, so focus on:

• Gap certainty — are existing MAs truly non-overlapping with proposed scope?
• k quality — are primary studies heterogeneous enough to warrant MA, or too uniform?
• User's domain fit — does this align with user's radiology AI / imaging expertise?

T-Phase 3: Topic Ranking (Topic-first weights)

Goal: Rank viable topics with weights adjusted for topic-first approach.

Output: Ranked Topic Table

T-Phase 4: Co-Author Matching (Optional)

Goal: If the user wants a senior co-author, find candidates.

Strategy 1 — Existing network (memory-based):

• Check memory files for professors with overlapping expertise
• Cross-reference existing professor folders in the working directory
• Best match = professor whose pillar naturally covers this topic

Strategy 2 — PubMed reverse search:

# Find prolific authors in this specific topic
bash ~/.claude/skills/search-lit/references/pubmed_eutils.sh search \
  '[topic keywords] AND ("{user_country}"[Affiliation])' 100

Then:

• E-utilities efetch → extract author frequency from results
• Top 5 most-published authors in this niche = potential co-authors
• Cross-check Google Scholar for h-index and recent activity

Strategy 3 — Self-led (no senior co-author):

• Viable when: user has 2+ published MAs, topic is methodologically straightforward
• Still need 2nd reviewer (junior colleague or peer) — flag this in README
• Corresponding author = user

Output: Co-author recommendation table or a "solo-viable" judgment.

T-Phase 5: Folder & README Scaffolding (Topic-first)

Goal: Create project folder and PROSPERO-ready README.

1. Folder location: {working_dir}/ma-scout/TOPIC/

   • Topic-first projects use TOPIC/ prefix (not professor initials)
   • Naming: {NN}_{Topic_Abbreviation}/ (e.g., 01_AI_Lung_Nodule_DTA/)
   • If co-author matched later, can be moved under professor folder

2. README.md template: Same PROSPERO-ready template as Mode A Phase 4 (see references/project_readme_template.md), with these changes:

   • Supervisor: → Lead: {user_name} or Lead: {user_name} + {co-author}
   • Drop the supervisor-area row; use Domain: {subspecialty} instead.
   • Rename Professor's Authority → Team Expertise (user's credentials + co-author if any)
   • Timeline: drop the supervisor-proposal step → start directly at PROSPERO registration.

   Timeline template (self-led):

   Step	Expected timing	Precondition
   PROSPERO registration	{YYYY-MM}	topic confirmed
   Search complete	+1 week	PROSPERO registration
   Screening complete	+2 weeks	2nd reviewer secured
   Data extraction	+3 weeks	screening consensus
   Analysis + draft	+5 weeks	data lock
   Co-author review	+7 weeks	draft complete
   Submission	+8 weeks	final approval

3. Summary: Same as Mode A Phase 5 — save ranked results and recommend next steps.

Topic Discovery Heuristics (Mode B specific)

When the user asks for topic suggestions without a specific idea:

1. Trend scan — Search recent high-IF radiology journals for "gap in the literature" + "meta-analysis needed":

   bash ~/.claude/skills/search-lit/references/pubmed_eutils.sh search \
     '"no meta-analysis" AND "radiology"[Journal] AND 2024:2026[dp]' 30
   

2. Guideline update gaps — New guidelines (ACR, ESR, RSNA) often cite lack of MA evidence:

   • Consensus search: "practice guideline" AND "insufficient evidence" AND [radiology subspecialty]

3. AI + classical imaging — Overlay AI/DL/radiomics on well-studied classical topics:

   • Many classical DTA topics have 10+ MAs, but AI angle has 0-1

4. Korean/Asian population — Population-specific MA for diseases with geographic variation:

   • TB, NTM, parasitic diseases, gastric cancer, liver fluke, HBV-related HCC

5. Technology adoption — New modalities with growing evidence but no synthesis:

   • Photon-counting CT, abbreviated MRI, contrast-enhanced mammography, AI CAD

6. Cross-subspecialty — Topics spanning two subspecialties often fall through MA cracks:

   • Cardiac + thoracic (coronary CT + lung screening), neuro + MSK (spine imaging)

Quality Gates (Mode B specific)

Before finalizing a topic-first MA as viable:

• Clinical question refined to PICO/PIRD (not just a keyword)
• MA gap confirmed via PubMed + Consensus + Scholar Gateway + bioRxiv (all 4 sources)
• k_realistic ≥ 8 (DTA) or ≥ 6 (prognostic) after 15-30% discount
• PROSPERO searched — no competing registered protocol
• No 2024-2026 competing MA in press or preprint
• User's domain expertise sufficient for clinical interpretation (or co-author identified)
• 2nd reviewer identified or plan to recruit
• README contains: complete PICO/PIRD, PubMed + Embase search strategy, target journal with IF, timeline
• If self-led: user has ≥ 2 published MAs (otherwise, recommend co-author)

Phase 6: Pre-Proposal Pipeline (Post-Scout)

After MA Scout identifies viable topics, run the pre-proposal pipeline to prepare a "ready-to-propose" package before contacting the professor.

Pipeline Steps

1. Search Execution — E-utilities with broadened synonyms (retmax=200)

   • Primary search: [topic] AND [outcome keywords]
   • Existing MA search: [topic] AND ("meta-analysis"[pt] OR "systematic review"[pt])

2. Metadata Collection — fetch_json → esummary (batch 40-50 PMIDs)

3. Title-Based Triage — Classify as INCLUDE / MAYBE / EXCLUDE

   • CRITICAL: Check for existing MAs within results (initial scout may miss them)
   • Separate bronchoscopic vs percutaneous (30% contamination in CBCT topics)
   • Flag professor's own papers (authority evidence)
   • Flag retracted papers

4. PRISMA Flow Draft — Identification → Screening → Eligibility → Included (estimated)

5. Gap Re-assessment — Update MA count, re-position if needed:

   • MA=0 → "first MA" | MA=1 (>5yr) → "update MA" | MA≥3 (recent) → skip/niche

6. Output Files:

   • candidates.md — full triage table + PRISMA flow + gap finding
   • README.md — updated Preliminary Search section with actual numbers

Parallel Execution

• Launch up to 4 agents per wave (each topic independent)
• Each agent: search → fetch → triage → write files → ~5-10 min
• 21 topics completed in ~1 hour with 16 parallel agents

Professor Contact Package

The pre-proposal gives the professor:

• Candidate count + gap evidence (e.g., "MA = 0, 35 studies to include")
• Clear role description (e.g., "independent screening review + discussion only")
• Urgency of PROSPERO pre-registration to secure the topic

Anti-Hallucination

• Never fabricate publication counts, h-index, or pillar classifications. All numbers must come from PubMed E-utilities API output.
• Never fabricate existing MA counts. Always verify via PubMed search + PROSPERO check before claiming "MA = 0".
• Never invent professor expertise or affiliation. Confirm with user before proceeding.
• k_realistic must use the 15-30% discount. Raw PubMed counts overestimate by 3-7x. Always report both raw and realistic estimates.
• If PubMed returns 0 or Consensus/Scholar Gateway is unavailable, state the limitation rather than guessing.