Deep Research
Deep research that doesn't stop at the third hit: search broad, hunt primary sources, cross-check, show contradictions, cite cleanly. Neutral — it reports sources and how strong they are, never a true/false verdict.
SKILL.md in the open agentskills.io standard — works directly in Claude, ChatGPT/Codex, Cursor, Copilot & more.
Deep Research is an AI research skill for deep, multi-source research and fact-checking — it hunts primary sources and cross-checks them. Breaks your question into checkable claims, verifies each against independent sources, surfaces contradictions, and delivers a traceable, cited report — neutral, never a true/false verdict.
What this skill does
- Breaks complex questions into checkable, atomic claims and prioritizes them.
- Searches across several rounds instead of just top hits — synonyms, other languages, specialist sources, forums, video transcripts and archives.
- Actively hunts primary and authoritative sources: laws, studies, registries, changelogs, official statistics.
- Cross-checks every key claim adversarially against independent sources and actively seeks counter-evidence and criticism.
- Detects fake independence via citation lineage: copied press releases or the same cited study count as one source.
- Rates sources by authority, recency, primary/secondary nature, bias and conflict of interest.
- Surfaces contradictions and uncertainties instead of smoothing them, and documents negative findings.
- Delivers an evidence-first report: a compact source table first, then evidence, confidence and a source list.
- Offers four depth levels (Quick, Standard, Deep, Exhaustive) and fitting output modes (Research Brief, Evidence Check, Decision Memo, Source Map, Technical, Legal/Policy, Historical).
- Can evaluate, structure and summarize sources you provide even without web access.
Description
Deep Research is a skill for deep, multi-source research — it breaks complex questions into checkable claims, searches broadly across several rounds, evaluates sources, checks contradictions and produces traceable reports with evidence and uncertainties. It works like a careful newsroom: it doesn't stop at the first hits, actively hunts primary sources and counter-evidence, and uses a scoring system to reach saturation (several independent sources, at least one primary, multiple domains and perspectives), with clear hard-stops preventing endless loops. Its core principle is neutrality: Deep Research never decides what is true or false — it reports sources and rates the evidence (number, independence, quality, recency, agreement and conflict) and leaves the judgement to you. Every report opens with a compact evidence table (sources, fact variants, agreement, links), followed by key findings with citations, contradictions, negative findings (what was searched but not found), a confidence / research-ledger overview, and an Evidence Summary including "what would change the picture". A standout: Deep Research checks real source independence via citation lineage — ten hits that copy the same press release or study count as one source. It offers four depth levels (Quick, Standard, Deep, Exhaustive), seven output modes (incl. Research Brief, Evidence Check, Decision Memo, Source Map), a high-stakes protocol for sensitive topics and a freshness protocol for time-critical questions. It includes 31 reference files (methodology plus curated source libraries for news, science, law and government, medicine, tech, finance, statistics, international/multilingual, fact-checking and archives), ready-to-run query recipes, six optional Node helpers and eight example reports. It activates only on explicit request (opt-in with a mode gate; in file-aware tools via an editable deep-research.config.json). For real depth it uses the host AI's web and search tools; without them it evaluates and condenses sources you provide. One-time purchase, runs locally with your own AI.
Examples
What it does not do
- Never decides what is true or false — it reports sources and rates the evidence; the judgement stays with you.
- Invents no sources, numbers, quotes or ratings; uncertainties and gaps are stated openly.
- Is not a substitute for professional legal, medical or financial advice — for sensitive topics it shows the authoritative sources and points to professionals.
- Bypasses no paywalls, logins or platform restrictions, and lists only publicly accessible sources.
- Brings no runtime or scraping of its own — it steers the host AI's web tools.
- Doesn't run on every question: opt-in with a mode gate; "always" is a copy-paste line or an editable config file.
- Doesn't turn weak sources into strong evidence.
Compatibility & tech
- Tested (internal)
- 6 scenarios
- Recommended runtime
- Any capable chat model with web/search tools for real deep research; especially smooth in tools with file/project context (Claude Code, Cursor, etc.) because the skill can use its references, source libraries, helper scripts and editable config directly in the project.
- Modes
- Quick · Standard · Deep · Exhaustive · Research Brief · Evidence Check · Decision Memo · Source Map · Technical Research · Legal/Policy Scan · Historical Reconstruction
- Inputs
- A research question in natural language ("Research thoroughly whether …") · A claim or figure to check (Evidence Check) · Provided sources/articles to evaluate (offline mode) · An A-vs-B decision question (Decision Memo)
- Output format
- An evidence-first report — a compact source/evidence table first, then key findings with citations, contradictions, negative findings, confidence / research ledger and an Evidence Summary. Shaped as a Research Brief, Evidence Check, Decision Memo, Source Map or Technical/Legal/Historical as the task needs.
- Subcategory
- Research / Source work
- License
- Proprietär
Security profile
Runs entirely on your machine with your own AI — no external runtime, no running costs.
Contains executable code or runs actions/tools — take a quick look before using.
Accesses external sources / the network in normal use (e.g. live pages, search/data APIs).
What you get
- deep-research-pro-1.1.0/50 files
- .claude-plugin/marketplace.json
plugins/deep-research-pro/skills/deep-research-pro/50 files
- SKILL.md
- README.md
- manifest.json
examples/9 files
- example-conflicting-sources.md
- example-decision-memo.md
- example-fact-check-with-contradictions.md
- example-full-research-run.md
- example-negative-findings.md
- example-research-ledger.md
- research-report-template.md
- scorecard.sample.json
- source-evaluation-matrix.md
references/32 files
- 00-research-loop-and-scoring.md
- 01-question-decomposition.md
- 02-search-strategy-fan-out.md
- 03-beyond-page-one.md
- 04-source-evaluation.md
- 05-adversarial-verification.md
- 06-conflicting-sources.md
- 07-citation-and-report.md
- 08-confidence-and-uncertainty.md
- 09-bias-and-fallacies.md
- 10-platforms-youtube-video.md
- 11-platforms-forums-social.md
- 12-modes-and-activation.md
- 13-offline-no-websearch-mode.md
- 14-source-independence-and-citation-lineage.md
- 15-research-ledger-schema.md
- 16-high-stakes-and-freshness.md
- 17-query-recipes.md
- 18-negative-findings.md
- 19-research-output-modes.md
- 20-sources-news-media.md
- 21-sources-science-academic.md
- 22-sources-law-government.md
- 23-sources-medicine-health.md
- 24-sources-tech-software.md
- 25-sources-finance-business.md
- 26-sources-international-multilingual.md
- 27-sources-data-statistics.md
- 28-sources-fact-checking.md
- 29-sources-archives-primary.md
- 30-neutrality-and-evidence-first.md
- deep-research-pro-checklist.md
scripts/6 files
- citation-validator.mjs
- claim-coverage-report.mjs
- query-recipe-generator.mjs
- research-ledger-builder.mjs
- research-scorecard.mjs
- source-diversity-check.mjs
- .agents/skills/deep-research-pro/→ universal — same content (Codex, Cursor, Copilot, Gemini, Windsurf, Cline)
- LICENSE.txt
Installation
Also works as a chat prompt
No AI tool? Paste it into Claude, ChatGPT or Gemini and use the method right away.
You get the full method. Only 6 helper script(s) that automate parts of it run once installed.
Works best when your chat has web access.
Installing is the full version — it triggers automatically, runs its scripts and loads references as needed. As a chat prompt you drive the method by hand.
Unlock to copy the ready-to-paste prompt — then in “My Skills”.
Reviews
No reviews yet – be the first.
Note
Generic methodology skill — it reports sources and evidence strength, not a verdict, and is not legal/medical/financial advice. For real depth it needs the host AI's web tools; without them it works with provided sources. Source domains can change; the skill prioritizes primary sources and checks recency.
Changelog
- v1.1.022.06.2026Platform-native output: now actively uses your platform's native formatting — tables, checklists, code blocks and clickable/button options, with a numbered fallback on non-interactive platforms. Plus internal metadata that prepares the upcoming agent builder.
- v1.0.022.06.2026Initial release: methodical deep-research skill — neutral (reports sources and evidence strength, not a verdict), evidence table first, claim decomposition, multi-round fan-out, primary-source hunting, adversarial verification, citation-lineage check, scoring with hard-stops, negative findings, Research Ledger and Confidence Card, Evidence Summary. Four depth levels, seven output modes, high-stakes and freshness protocols, opt-in mode gate and editable config. 31 references incl. 10 curated source libraries and query recipes, 6 optional Node helpers, 8 example reports and internal evals.
Frequently asked questions
What does Deep Research do?
Deep Research is an AI research skill for deep, multi-source research and fact-checking — it hunts primary sources and cross-checks them. Breaks your question into checkable claims, verifies each against independent sources, surfaces contradictions, and delivers a traceable, cited report — neutral, never a true/false verdict.
What makes Deep Research different from a normal AI answer?
Deep Research doesn't work off a few hits — it runs a structured research loop: claim decomposition, broad multi-round search, primary-source hunting, adversarial cross-checking, a source-independence check and a scoring system with clear stop rules. The result is a traceable report with a source table, evidence, confidence and open gaps — not a quick, unchecked answer.
Does Deep Research decide whether a claim is true?
No — neutrality is the core principle. Deep Research never rules on true or false. It reports sources and rates the evidence (how many independent sources, their quality, recency, agreement or conflict) and leaves the judgement to you. Instead of "false" it says, e.g., "no primary source found; the figure traces to a single origin".
When should I use Deep Research?
Deep Research suits questions where solid sources, multiple perspectives or a traceable line of evidence matter — fact-checks, market research, technical research, decision support or contested topics. For simple knowledge questions it deliberately stays off.
Does Deep Research need web access?
For real deep research it uses the host AI's web and search tools. Without web access it runs in offline mode: it evaluates, structures and condenses the sources you provide, and flags that live breadth was unavailable.
How does Deep Research avoid counting the same claim multiple times?
Deep Research checks real independence via citation lineage: several pages copying the same press release or study count as one source. Only independent sources with their own data count as genuine confirmation.
Can Deep Research stay active permanently?
Deep Research starts only on explicit request. For "always" it provides a copy-paste instruction for project or system prompts and, in file-aware tools, can keep an editable deep-research.config.json. There is deliberately no automatic, cross-tool permanent activation — you stay in control.
Which AI tools does Deep Research work with?
Claude (Projects & system prompt) & Claude Code · ChatGPT/Codex, Cursor, Copilot, Gemini CLI, Windsurf, Cline (SKILL.md/paste-in) · Uses the host AI's web/search tools; optional offline mode for provided sources; optional Node helpers
How do I use Deep Research?
Deep Research is a SKILL.md in the open agentskills.io standard: install it with one command (npx) or download it and add it to your AI tool — Claude (Projects), ChatGPT (Custom GPT), Cursor, Copilot, Gemini CLI and more. No code needed.