Deep Researcher Skill

Aggregates insights from web, papers, video, and forums with paywall bypass to deliver comprehensive, unbiased, multi-format research and analysis.

Security Scan

The skill's instructions, required tools, and lack of hidden installs line up with its stated purpose (including explicit paywall-bypass behavior), but the paywall-bypass techniques and third‑party fetch endpoints raise privacy, legal, and data‑leak considerations you should review before use.

Purpose & Capability

Name/description promise multi‑source research and paywall bypass; SKILL.md explicitly lists built‑in search/YouTube/arXiv/Reddit/browser tools and paywall bypass steps, which are proportional to the stated goals. No unrelated credentials, binaries, or installs are requested.

Instruction Scope

Instructions stay within the research domain (search, transcripts, papers, forums) but explicitly direct the agent to use third‑party proxy services (archive.org/google cache, 12ft.io, r.jina.ai) and a browser tool to retrieve paywalled content. That is coherent with the skill's stated paywall‑bypass capability, but it means target URLs and page contents will be sent to those external services (privacy/data‑exfiltration risk). The skill also offers a 'Show my research history' feature without documenting storage location or retention—this is ambiguous and worth clarifying.

Install Mechanism

No install spec or code; instruction‑only skill — lowest on‑disk risk. There are no downloads, archives, or external install URLs to review.

Credentials

No required environment variables or credentials. Optional API keys (NewsAPI, OpenWeather, ExchangeRate) are reasonable for the claimed optional integrations and are documented as optional. No unrelated secrets requested.

Persistence & Privilege

always is false and autonomous invocation is allowed (platform default). The skill references a research history feature but does not declare where history is stored or how long it is retained — this is a scope/privilege detail that should be clarified before use. There is no instruction to modify other skills or system settings.

Assessment

This skill is internally consistent with its description, but before installing you should: (1) decide whether you are comfortable that requested URLs and article contents may be sent to third‑party fetchers (archive.org/google cache, 12ft.io, r.jina.ai) — that can leak sensitive URLs or content; (2) consider legal/ethical implications of automated paywall bypass for copyrighted material and check terms of service for sources you care about; (3) confirm where and how 'research history' will be stored and who can access it (agent memory, local disk, remote store); (4) if you supply optional API keys, limit their scope and monitor usage; and (5) restrict or review the agent's browser tool permissions (cookies/authenticated sessions) to avoid unintended access to private accounts. If you need stronger assurances, ask the skill author to document storage/retention and to offer an option that disables proxy fetchers and only uses public/open sources.

SKILL.md - Research Assistant

Description

Your personal research department. Multi-source synthesis that turns scattered information into actionable intelligence — not just summaries, but insights you can act on.

Price

Free — or $5 to support development.

Prerequisites

• DuckDuckGo Search (built-in, no key needed)
• YouTube Content tool (built-in, no key needed)
• arXiv skill (built-in, no key needed)
• Reddit Readonly skill (built-in, no key needed)
• Browser tool (built-in, for paywall bypass)
• Optional: NewsAPI key (free tier: 100 requests/day) — current events
• Optional: OpenWeather API key (free tier: 1,000 calls/day) — location context
• Optional: ExchangeRate-API key (free tier: 1,500 requests/month) — finance data
• Optional: REST Countries API (no key needed) — demographics

Quick Start

1. Configure optional APIs: "Set up my research assistant with NewsAPI"
2. Research: "Research [topic]" or "Deep dive into [question]"

Commands

• "Research [topic]" — Quick synthesis from multiple sources
• "Deep dive into [question]" — Comprehensive analysis
• "Compare [A] vs [B]" — Competitive/feature analysis
• "What's new in [field] this month?" — Temporal research
• "Research for [format]: [topic]" — Brief, thread, blog, or decision matrix
• "Show my research history" — Previous queries and findings

---

Tool Selection Matrix

Source Type	Tool to Use	Fallback
Web search	duckduckgo_search	None needed
YouTube transcripts	youtube-content skill	Browser tool
Academic papers	arxiv skill	duckduckgo_search with site:arxiv.org
Forums/Reddit	reddit-readonly skill	None needed
Paywalled articles	browser_navigate + archive.org	12ft.io, textise dot iitty
Current events	NewsAPI (if configured)	duckduckgo_search news filter
Weather data	OpenWeather API	duckduckgo_search
Financial data	ExchangeRate-API	duckduckgo_search

Paywall Bypass Strategy

When you hit a paywall:

1. Try archive.org: https://webcache.googleusercontent.com/search?q=URL or https://archive.org/web/*/URL
2. Try 12ft.io: https://12ft.io/URL (works for Medium, Substack, etc.)
3. Try textise dot iitty: https://r.jina.ai/http://URL (extracts article text)
4. Use browser tool: Navigate and extract text directly
5. Skip only if all fail — mark as "paywalled, unverified"

---

Core Workflows

1. Quick Research (2-3 minutes)

Input: Any question or topic

Process:

1. Parallel search across sources using tool matrix above
2. Fetch top 3-5 results per source
3. Bypass paywalls using strategy above
4. Extract key points from each
5. Synthesize into structured brief
6. Cite all sources with links

Output:

## Research Brief: [Topic]

### Executive Summary
[3-5 sentences covering the landscape]

### Key Findings
1. **[Finding]** — [Source type: web/video/paper/forum]
2. **[Finding]** — [Source type]
3. **[Finding]** — [Source type]

### Sources
- [Title](URL) — Web article, [Date]
- [Title](URL) — YouTube video, [Channel]
- [Title](URL) — arXiv paper, [Authors]
- [Title](URL) — Reddit discussion, [Subreddit]

### Confidence Score: [High/Medium/Low]
**Why:** [Source quality, recency, consensus level]

### Suggested Next Steps
- [Specific follow-up question]
- [Related topic to explore]
- [Deeper source to check]

2. Deep Dive Research (5-10 minutes)

Input: Complex question requiring comprehensive analysis

Process:

1. Multi-query expansion (break topic into sub-questions)
2. 10-15 sources across all channels
3. Apply paywall bypass as needed
4. Temporal analysis (what's new vs. established)
5. Credibility scoring per source
6. Bias detection and flagging
7. Synthesis with uncertainty levels

Stopping Conditions — When to End:

• Saturation: New sources repeat what you already found
• Diminishing returns: 10+ sources but confidence still Low
• Contradiction ceiling: >50% of sources disagree
• Time limit: 15 minutes max for Deep Dive
• Confidence achieved: High confidence with 3+ Tier 1 sources

Output:

## Deep Dive: [Topic]

### One-Paragraph Summary
[The TL;DR for busy decision-makers]

### Current State (What's Happening Now)
[Recent developments, 0-6 months]

### Established Knowledge (What We Know)
[Consensus views, foundational concepts]

### Points of Contention
- **[Claim A]** — [Evidence for] vs [Evidence against]
- **[Claim B]** — [Evidence for] vs [Evidence against]

### Source Quality Breakdown
| Source | Type | Credibility | Recency | Bias |
|--------|------|-------------|---------|------|
| [Name] | Academic | High | 2024 | Neutral |
| [Name] | News | Medium | 2025 | Center-left |
| [Name] | Forum | Low | 2025 | N/A |

### Confidence Calibration
**Level:** [High/Medium/Low]
**Reasoning:** [Why this level based on criteria below]

### Actionable Insights
1. **[Insight]** — [Specific action to take]
2. **[Insight]** — [Specific action to take]

### Knowledge Gaps
[What we still don't know]

### Recommended Follow-Up
- [Specific research question]
- [Expert to consult]
- [Primary source to find]

3. Comparative Analysis

Input: "Compare X vs Y" or "Feature gap analysis"

Process:

1. Research both subjects independently using tool matrix
2. Extract features/capabilities/attributes
3. Build comparison matrix
4. Identify gaps and differentiators
5. Score on key dimensions

Structured Data Extraction:

Pricing extraction pattern:
- Search: "[Product] pricing cost $"
- Look for: $XXX/month, $XXX/year, free tier limits
- Source: Official pricing page (bypass paywall if needed)

Feature extraction pattern:
- Search: "[Product] features vs [Competitor]"
- Look for: Feature lists, comparison tables
- Use: Browser tool to extract structured data

Sentiment extraction pattern:
- Reddit: Search r/[topic] for "[Product] review"
- Look for: Specific pros/cons with reasoning
- Score: Count positive vs negative mentions

Output:

## Comparison: [A] vs [B]

### At a Glance
| Dimension | [A] | [B] | Winner |
|-----------|-----|-----|--------|
| Price | $X | $Y | [A/B/Tie] |
| Key Feature | [Desc] | [Desc] | [A/B/Tie] |
| User Sentiment | [Score] | [Score] | [A/B/Tie] |

### Detailed Breakdown

**[A] Strengths:**
- [Point with source]
- [Point with source]

**[B] Strengths:**
- [Point with source]
- [Point with source]

**[A] Weaknesses:**
- [Point with source]

**[B] Weaknesses:**
- [Point with source]

### Feature Gap Analysis
- [Feature A]: [A] has it, [B] doesn't
- [Feature B]: Both have it, [A] does it better
- [Feature C]: Neither has it (opportunity)

### Verdict
[Recommendation with reasoning]

### Sources
[All citations]

4. Temporal Research (What's New)

Input: "What's new in [field] this [timeframe]?"

Process:

1. Filter sources by date using search filters
2. Compare to baseline (previous period)
3. Identify new developments, trends, shifts
4. Flag emerging vs. fading topics

Output:

## [Field] Update: [Timeframe]

### New Developments
1. **[Development]** — [Impact level] — [Source]
2. **[Development]** — [Impact level] — [Source]

### Trends to Watch
- [Trend]: [Evidence] — [Trajectory: rising/stable/falling]
- [Trend]: [Evidence] — [Trajectory]

### What's Fading
- [Topic]: [Why it's declining]

### Predictions (Speculative)
- [Prediction] — [Based on]

### Sources from This Period
[All recent citations]

5. Format-Specific Output

Brief Mode: Executive summary only (2-3 paragraphs)

Thread Mode: Twitter/X thread format

🧵 [Topic]: [Hook]

1/ [Point]
2/ [Point]
3/ [Point]

[Sources]

Blog Mode: H2 outline with key points

## [Title]

### Introduction
[Hook]

### [Section 1]
[Key points]

### [Section 2]
[Key points]

### Conclusion
[Takeaway]

### Sources
[Citations]

Decision Matrix Mode: Pros/cons table with scoring

| Option | Pros | Cons | Score |
|--------|------|------|-------|
| [A] | [List] | [List] | X/10 |
| [B] | [List] | [List] | X/10 |

---

Confidence Calibration System

Don't guess — use these criteria:

High Confidence

• Sources: 3+ Tier 1 (academic, official, expert) OR 5+ Tier 2
• Recency: All sources <6 months old OR established consensus
• Contradictions: Zero major contradictions
• Corroboration: Findings confirmed by independent sources

Medium Confidence

• Sources: 2+ Tier 2 (industry pubs, established blogs)
• Recency: Mix of recent and established
• Contradictions: Minor contradictions resolved
• Gaps: Some uncertainty acknowledged

Low Confidence

• Sources: Single source OR mostly Tier 4-5
• Recency: Old data (>1 year) OR no date
• Contradictions: Major contradictions unresolved
• Gaps: Significant unknowns

Flag language:

• High: "Research shows...", "Evidence confirms..."
• Medium: "Sources suggest...", "It appears that..."
• Low: "One source claims...", "Limited research indicates..."

---

Source Quality Scoring

Tier 1: Highest Credibility (Weight: 3x)

• Peer-reviewed journals (Nature, Science, etc.)
• Official documentation (gov, corporate)
• SEC filings, regulatory documents
• Direct primary sources

Tier 2: High Credibility (Weight: 2x)

• Established news (Reuters, AP, BBC)
• Expert blogs with track record
• Industry analysts (Gartner, McKinsey)
• Technical publications (IEEE, ACM)

Tier 3: Medium Credibility (Weight: 1x)

• Industry publications
• Established YouTube channels
• Well-moderated forums
• Think tank reports

Tier 4: Low Credibility (Weight: 0.5x)

• General news coverage
• Encyclopedia entries (Wikipedia — follow citations)
• Content aggregators

Tier 5: Use Cautiously (Weight: 0.25x)

• Anonymous forums
• Unverified social posts
• Personal blogs without track record

Auto-Skip

• Known misinformation sources
• Circular references (A cites B cites A)
• Paywalled AND can't bypass

---

Bias Detection

Political Spectrum

• Left / Center-left / Center / Center-right / Right
• Flagged when source consistently leans one direction

Commercial Bias

• None: No financial stake in topic
• Disclosed: Affiliate links, sponsorships noted
• Undisclosed: Potential conflicts not mentioned

Confirmation Bias Warning

• "This source only presents one side"
• "Contradictory evidence exists"
• "Consensus vs. outlier view"

---

Domain-Specific Research Patterns

Product/Competitive Research (ClawMart/Gumroad)

1. Search: "[Product] vs [competitor] review"
2. Check: Pricing pages, feature lists
3. Mine: Reddit for real user experiences
4. Extract: Structured comparison table

Scientific/Medical Research

1. Start: PubMed, Cochrane Library for clinical
2. Check: arXiv for preprints
3. Verify: Primary sources, not news summaries
4. Flag: Single studies vs. meta-analyses

Financial Research

1. SEC filings (10-K, 10-Q) for public companies
2. Earnings call transcripts
3. Analyst reports (paid) → use summaries from news
4. ExchangeRate-API for currency data

Legal Research

1. Court documents via PACER or public dockets
2. Legal databases (Westlaw, Lexis — paywalled)
3. Law reviews and journals
4. Regulatory agency guidance

---

Configuration

research-config.json

{
  "apis": {
    "newsapi": {
      "key": "YOUR_NEWSAPI_KEY",
      "tier": "free"
    },
    "openweather": {
      "key": "YOUR_OPENWEATHER_KEY",
      "tier": "free"
    },
    "exchangerate": {
      "key": "YOUR_EXCHANGERATE_KEY",
      "tier": "free"
    }
  },
  "defaults": {
    "max_sources": 10,
    "recency_preference": "balanced",
    "output_format": "brief",
    "confidence_threshold": "medium"
  },
  "quality_thresholds": {
    "min_credibility": "medium",
    "max_age_days": 365,
    "require_diversity": true
  },
  "stopping_conditions": {
    "max_sources": 15,
    "max_time_minutes": 15,
    "saturation_threshold": 3
  }
}

---

Examples

Example 1: Quick Research

User: "Research AI video generation trends"

Agent:
## Research Brief: AI Video Generation Trends

### Executive Summary
AI video generation has exploded in 2024-2025 with Runway, Pika, and Kling leading consumer tools. Quality improved dramatically — 4-second clips now rival professional footage. Open-source alternatives (Stable Video Diffusion) gaining traction. Key limitation: temporal consistency in longer sequences.

### Key Findings
1. **Runway Gen-2** is market leader for quality — $15/mo, 4-second clips
2. **Pika Labs** offers budget alternative at $8/mo with good results
3. **Kling AI** (China-based) emerging as quality competitor
4. **Open source** (Stable Video Diffusion) viable for technical users

### Sources
- [The State of AI Video 2025](URL) — Web, Jan 2025
- [Runway vs Pika: Comparison](URL) — YouTube, TechChannel
- [Stable Video Diffusion Paper](URL) — arXiv, 2024
- [r/MachineLearning discussion](URL) — Reddit, Feb 2025

### Confidence Score: High
Multiple Tier 2+ sources confirm, recent data, established consensus. Zero contradictions.

### Suggested Next Steps
- Deep dive into temporal consistency solutions
- Compare pricing for high-volume production
- Research enterprise alternatives (Sora, etc.)

Example 2: Competitive Analysis with Paywall Bypass

User: "Compare ElevenLabs vs Play.ht for voice generation"

Agent:
## Comparison: ElevenLabs vs Play.ht

### At a Glance
| Dimension | ElevenLabs | Play.ht | Winner |
|-----------|------------|---------|--------|
| Price | $5/mo | $30/mo | ElevenLabs |
| Quality | Excellent | Good | ElevenLabs |
| Languages | 29 | 142 | Play.ht |
| API | Yes | Yes | Tie |

### Pricing Research
- ElevenLabs: $5/mo Starter, $22/mo Creator (source: pricing page)
- Play.ht: $30/mo Personal, $99/mo Pro (source: pricing page)

### User Sentiment (Reddit r/elevenlabs, r/speech synthesis)
- ElevenLabs: 85% positive (quality praised, occasional latency issues)
- Play.ht: 70% positive (good languages, expensive for quality)

### Verdict
ElevenLabs for quality/price, Play.ht if you need obscure languages.

[Full breakdown with sources...]

---

Guardrails

• Always cite sources — never present synthesis as original research
• Flag uncertainty with calibrated language ("likely" vs "confirms")
• Bypass paywalls before skipping — don't leave knowledge on the table
• Respect rate limits — cache results, batch when possible
• Verify recency — old data can mislead
• Acknowledge gaps — "limited research exists on..."
• Stop when conditions met — don't over-research

Troubleshooting

Error: "No results found"

• Try broader search terms
• Check tool availability
• Verify internet connection

Error: "Rate limit exceeded"

• Wait 60 seconds, retry
• Switch to fallback tools
• Use cached results when available

Error: "Source quality too low"

• Broaden search terms
• Remove recency filter
• Try alternative sources from tool matrix
• Accept "Low confidence" finding

Error: "Paywall blocking access"

• Try archive.org
• Try 12ft.io
• Try textise dot iitty
• Use browser tool to extract
• Only skip if all methods fail

---

Version History

• V1.0: Multi-source search, synthesis, 4 output formats
• V1.1: Bias detection, temporal research, competitive analysis
• V1.2: Source quality scoring, citation export, research history
• V1.3: Public APIs integration — NewsAPI, OpenWeather, ExchangeRate-API, REST Countries
• V1.4:
  • Added tool selection matrix (platform-agnostic)
  • Added paywall bypass strategy
  • Added confidence calibration system
  • Added stopping conditions
  • Added structured data extraction patterns
  • Added domain-specific research patterns

---

Turn information into intelligence.