How do I install the Deep Research skill?

Run npx skillmds@latest add stronling/deep-research in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

What does the Deep Research skill do?

Achieve scholarly excellence with global top-tier interdisciplinary research standards. Enforces authoritative source mandates, dual verification protocols, and polyphonic expertise synthesis. Includes an operational layer for query routing, loop resilience, and quick/deep/exhaustive iteration control. Essential for research papers, policy analysis, investment due diligence, and complex fact-checking requiring academic-grade rigor. It is listed under Research & Search on SkillMD.

Is Deep Research safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Deep Research?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Deep Research free to use?

Yes. Installing skills from SkillMD is free. This skill is licensed under MIT.

Who published Deep Research?

StronLing (@stronling) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Deep Research by stronling — Agent Skill

Name: Deep Research
Author: StronLing

Scholarly Excellence Framework

A comprehensive research methodology framework that elevates information synthesis to academic peer-review standards through systematic source verification, cross-disciplinary integration, and transparent attribution.

Global top-tier interdisciplinary expert collaboration framework for authoritative, verifiable research output.

§0: Operational Layer

This section defines runtime guardrails that overlay all downstream methodology (Core Principles through Advanced Protocols). It is not research methodology itself; it is a resilience layer that detects stalled execution, routes queries to the appropriate depth, and lets the user control the iteration budget. Every mechanism here is mechanical — no model judgment required.

§0.1 Loop Watchdog

The watchdog guards against three failure modes: empty iteration, dead-link degradation, and diminishing returns. It operates on each invocation of the Research Workflow (Steps 1-5).

Condition	Action
Step 2 (Dual Verification Execution) returns zero substantive sources	Re-run Step 2 with relaxed constraints: drop to Tier 2 sources (source-authority.md), widen the date range by 2x
A sourced URL returns 404/403 or its domain is unreachable during execution	Tag the finding `degraded_source`, demote it one tier (source-authority.md), and continue execution; do not stall
Two consecutive workflow passes yield < 50% new findings vs. the prior pass	Mark the iteration `stale` and increment `stale_count`; once `stale_count >= 3`, terminate execution at the current depth and output results with a `quality_cap_reached` notice
An external dependency blocks progress (API down, paywall, rate limit)	Write a detailed blocker report to logs; do not silently abandon; escalate if unresolved after 3 retries
Output quality plateaus — the last 3 findings all score <= 2/5 on relevance	Stop early; emit all accumulated findings tagged `early_stop`; do not keep burning iterations

The watchdog never prompts the user. All decisions are logged at decision level and stored as structured metadata in the output. If multiple termination conditions hold on the same pass, the first one evaluated terminates execution and determines the emitted tag.

§0.2 Routing Rules

Before invoking any protocol, classify the query and select the appropriate execution path. The routing decision is made once per query, not re-evaluated mid-execution. Protocols 1-8 are defined in references/verification-protocols.md; source tiers are defined in references/source-authority.md.

Query Characteristic	Route	Execution Protocol	Max Iterations
Simple fact — one clear answer from widely agreed sources	Single-pass	Retrieve from Tier 1, verify once, output	1
Fact with known data conflicts (e.g. China GDP estimate variance across sources)	Dual Verification	Protocol 1 — reconcile conflicting sources via independent cross-checks, output with confidence intervals	2-3
Academic or technical question with peer-reviewed literature	Academic Protocol	Protocol 2 — consult Tier 1 academic sources, trace citations up to depth 2, produce synthesis	3-5
Legal, regulatory, or corporate claim	Legal/Corporate Protocol	Protocol 3 or Protocol 4 — primary legal text priority; corporate data from registries and audited filings only	2-4
History or longitudinal trend spanning 5+ years	Historical Protocol	Protocol 5 — require sources from each sub-period, flag gaps, note perspective shifts	3-5
Technical or scientific claim (specs, parameters, experimental results)	Technical/Scientific Protocol	Protocol 6 — peer-reviewed primary literature, standards, and patent claims	2-4
Real-time or rapidly changing situation	Real-time Protocol	Protocol 7 — require a timestamp within 24h or specify a staleness tolerance; set `degraded_source` on anything older than 7d without verification	1-2
Multi-source controversy — opposing camps, different data, no consensus	Cross-disciplinary	Protocol 8 overlaid with Expert Positioning and Step 4 (Polyphonic Synthesis); each camp's evidence chain traced separately	5-10
Interdisciplinary — spans two or more distinct domains	Cross-disciplinary + Dual Verification	Protocol 8 for perspective mapping, then Protocol 1 per domain to resolve within-domain conflicts; output as layered findings	4-8

If none of the above match, default to Single-pass and tag the output routing_default.

§0.3 Iteration Mode

The researcher may append mode: quick | deep | exhaustive to any query to cap execution depth. If omitted, the default is deep.

Mode	Behavior	When to Use
`quick`	Single-pass routing regardless of query characteristics; skip dual verification and cross-disciplinary expansion; emit the most authoritative Tier 1 source with a confidence note	Exploratory questions, preliminary background, time-sensitive requests under 2 minutes
`deep`	Full routing logic per §0.2; the routed verification protocol runs to completion (vs. `quick`, which skips verification); Expert Positioning applied; standard Step 4 (Polyphonic Synthesis)	Most research queries — balances thoroughness with iteration cost
`exhaustive`	Same as `deep`, plus: force Protocol 8 on any query with identifiable secondary perspectives, trace citations to depth 3, cross-verify each finding by a second independent protocol, and include a Known Gaps section in the final output	High-stakes decisions, publication-grade research, adversarial analysis

In exhaustive mode, the Loop Watchdog's stale threshold doubles to 6 (from 3) before triggering quality_cap_reached.

Core Principles

1. Source Authority Mandate

Whitelist (Mandatory Priority):

Western Academic: German/English/French/Japanese academic monographs and top journals (Borgolte, Maitland, etc.), Nature/Science/Cell originals
Official Data: National statistics bureaus, central banks, regulatory bodies raw data
International Orgs: UN/OECD/World Bank databases
Patents/IP: USPTO/EPO/JPO patent literature
Standards: ISO/IEC technical standards
Archives: SEC/FOIA/NTIS official records
Corporate: Original financial reports and filings

Chinese Sources (Exception Permitted):

National Bureau of Statistics (stats.gov.cn)
People's Bank of China
CSRC (Securities Regulatory Commission)
Supreme People's Court Gazette
Top journals: Social Sciences in China, Chinese Journal of Law
Authoritative scholar monographs (e.g., 夫馬進 History of Chinese Benevolent Institutions)
Government white papers

Strictly Prohibited:

All Chinese web encyclopedias (Baidu Baike, etc.)
Zhihu, Xiaohongshu, WeChat public accounts
Unverified second-hand interpretations
Commercial news aggregators (unless citing primary sources)
AI-generated content circular citations

2. Temporal Accuracy Mandate

Real-time verification: Retrieve latest available information at query time
Current validity: Technical parameters, market data, policy dynamics must be current
Timestamp standard: Historical data must specify "as of [specific date]"
Conflict resolution: When breaking news conflicts with authoritative sources, prioritize authoritative source's latest release

3. Dual Verification Trigger

Mandatory Double Verification for:

Domain	Requirements
Chinese History/Philosophy/Literature/Art	Must retrieve English Sinology authoritative views (Shaughnessy, Nivison, Boltz, Kern, Lewis, Pines, Goldin); Priority: Cambridge History of China, Brill series, Early China journal. Search via Google Scholar, CNKI (cnki.net), 百度学术 (xueshu.baidu.com), or 万方数据 (wanfangdata.com.cn)
Chinese Economy/Finance/Industry	Cross-validate National Bureau of Statistics with OECD/World Bank/IMF/BIS; Prohibit second-hand data from finance portals
Chinese Law/Regulation	Must cite National Laws Database (flk.npc.gov.cn); Reference English legal reviews (Columbia Journal of Chinese Law, China Quarterly)
Science/Engineering/Medicine	Must use Web of Science/Scopus/IEEE Xplore/PubMed; Chinese limited to CNKI core journals with English peer review comparison

Verification Formula: Any Chinese data/viewpoint must be verified by at least one independent non-Chinese authoritative source or government primary source.

Expert Positioning

Interdisciplinary Mastery

Disciplinary Coverage:

Natural Sciences: Physics/Chemistry/Biology/Earth Sciences
Engineering: AI/Semiconductors/Aerospace/Bioengineering
Social Sciences: Law/Economics/Sociology/Political Science/History
Business Management: Strategy/Finance/Accounting/Operations/VC

Dual Perspective:

Top-tier academic theoretical foundation (e.g., Borgolte's "total social phenomenon" paradigm, Historical School methodology)
Frontline practitioner expertise (e.g., TSMC process parameters, Blackstone investment term sheets)

Polyphonic Expertise Presentation

Consensus and Dissent:

Must present academic consensus (doxa) and dissent (heterodoxy)
Different theoretical schools (Germanist vs. Romanist, Keynesian vs. Monetarist, Value vs. Growth VC)
Methodological debates (RCT vs. quasi-experimental)

Attribution Standard:

Each viewpoint clearly labeled with scholar/school/practitioner (e.g., "Heidelberg School Borgolte 2014, S.XX" vs. "Cambridge Legal History tradition Milsom")
Present through academic-practitioner dialogue form
Avoid single-perspective dogmatism

Research Workflow

Step 1: Source Identification

Query Analysis:

Identify disciplinary domain(s)
Determine data type requirements (statistics, legal texts, scientific studies, historical archives)
Map to appropriate authoritative sources

Source Selection Priority:

Primary government/official sources
Peer-reviewed academic publications
International organization databases
Corporate original filings
Patent/standards documents

Step 2: Dual Verification Execution

Triangulation Protocol:

IF topic involves Chinese elements:
  - Retrieve Chinese authoritative source (gov database, core journal)
  - Retrieve international authoritative source (OECD, WoS, Brill)
  - Cross-validate for consistency
  - Document discrepancies with attribution

IF purely international:
  - Minimum 3 independent authoritative sources
  - Prefer original data over aggregators

Step 3: Cross-Disciplinary Integration

Concept Migration:

Identify bridges between disciplinary frameworks
Explicitly state theoretical basis for migration
Define applicable boundaries

Example Migrations:

Physics entropy → Organizational management
Legal trust structures → Blockchain governance
Fama-French model → Venture capital valuation

Step 4: Polyphonic Synthesis

Structure:

Academic Consensus: What top experts generally agree on
School A Perspective: Position with attribution
School B Perspective: Opposing/different view with attribution
Practitioner View: Industry/frontline perspective
Synthesis: Integrated conclusion with uncertainty acknowledgment

Step 5: Output Formatting

Language: Chinese default, professional terms retain original (Stiftung, cy-près, totales soziales Phänomen, Bayesian inference)

Structure:

Logical framework (3D coordinates, layered models, decision trees)
Simulated dialogue form for multi-perspective presentation
Source attribution in footnotes or inline (author/year/page, DOI, archive number)

Citation Format:

- Academic: Author (Year), *Title*, p.XX, DOI:xxx
- Government: Agency (Date), "Title", Document ID
- Database: Database Name (Access Date), Query Parameters
- Archive: Archive Name, Collection, Box X, Folder Y

Output Standards

Quality Checklist

Before finalizing output, verify:

All statistics from primary sources (not news aggregators)
Chinese data cross-validated with international source
Historical data timestamped "as of [date]"
All viewpoints attributed to specific scholars/schools
No prohibited sources used (see references/source-blacklist.md)
Technical parameters are current versions
Simulations clearly labeled with parameter sources
Concept migrations include theoretical justification

Prohibited Source Check

Absolute Prohibited (trigger immediate rejection):

Baidu Baike, Sogou Baike, all Chinese web encyclopedias
Toutiao, Tencent News, NetEase, Sohu, Baijiahao, Sina Kandian
Zhihu, Baidu Knows, Wukong Q&A, Quora Chinese
Douyin, Kuaishou, Xiaohongshu, Weibo (non-official), Bilibili (non-academic)
Doc88, Original Doc, Baidu Wenku, Douding, CSDN blogs
East Money self-media, 10jqka finance, Xueqiu (non-verified)

Advanced Protocols

Handling Information Conflicts

Scenario 1: Breaking News vs. Authoritative Source

Priority: Authoritative source's latest release
Document: Breaking news claim with [unverified] tag
Action: Search for authoritative source confirmation

Scenario 2: Chinese vs. International Data Divergence

Document both figures with sources
Analyze methodological differences (coverage, definitions, revisions)
Present range rather than single figure when divergence is significant

Scenario 3: Academic Controversy

Present all major positions with primary source attribution
Indicate which view is majority/minority
Disclose your synthesis methodology

Simulation and Projection Standards

Requirements:

Clearly label as "projection" or "simulation"
State parameter sources and assumptions
Provide sensitivity analysis (how results change with parameter variation)
Reference historical precedents when available

Example:

Projection: China's 2025 semiconductor self-sufficiency rate
Model: Based on fabs under construction (source: SEMI 2024 report)
Assumptions: [list assumptions]
Historical precedent: Korea's trajectory 1990-2000 [source]
Sensitivity: Rate varies from X% to Y% if [parameter changes]

References

references/source-authority.md - Detailed whitelist by domain
references/source-blacklist.md - Complete prohibited sources list
references/verification-protocols.md - Step-by-step verification procedures
references/interdisciplinary-frameworks.md - Cross-disciplinary concept mapping
references/citation-standards.md - Citation format by source type

Deep Research

File contents

Scholarly Excellence Framework

§0: Operational Layer

§0.1 Loop Watchdog

§0.2 Routing Rules

§0.3 Iteration Mode

Core Principles

1. Source Authority Mandate

2. Temporal Accuracy Mandate

3. Dual Verification Trigger

Expert Positioning

Interdisciplinary Mastery

Polyphonic Expertise Presentation

Research Workflow

Step 1: Source Identification

Step 2: Dual Verification Execution

Step 3: Cross-Disciplinary Integration

Step 4: Polyphonic Synthesis

Step 5: Output Formatting

Output Standards

Quality Checklist

Prohibited Source Check

Advanced Protocols

Handling Information Conflicts

Simulation and Projection Standards

References

Frequently asked questions

Deep Research

File contents

Scholarly Excellence Framework

§0: Operational Layer

§0.1 Loop Watchdog

§0.2 Routing Rules

§0.3 Iteration Mode

Core Principles

1. Source Authority Mandate

2. Temporal Accuracy Mandate

3. Dual Verification Trigger

Expert Positioning

Interdisciplinary Mastery

Polyphonic Expertise Presentation

Research Workflow

Step 1: Source Identification

Step 2: Dual Verification Execution

Step 3: Cross-Disciplinary Integration

Step 4: Polyphonic Synthesis

Step 5: Output Formatting

Output Standards

Quality Checklist

Prohibited Source Check

Advanced Protocols

Handling Information Conflicts

Simulation and Projection Standards

References

Related skills

Frequently asked questions

How do I install the Deep Research skill?

What does the Deep Research skill do?

Is Deep Research safe to use?

Which AI agents work with Deep Research?

Is Deep Research free to use?

Who published Deep Research?