Deep Research (No External API)

Multi-step research skill using native web tools (WebSearch/WebFetch) to decompose complex questions, gather authoritative sources, synthesize claim-level evidence, cite findings, and report confidence/gaps. It runs without external API keys by default and includes an optional dense/frontier mode for maximum-depth research.

When to use

• Research a topic with multiple sources before making a decision.
• Gather and synthesize technical documentation.
• Produce a structured report with cited evidence.
• Answer complex questions that require web searches.
• Run dense, Perplexity-like research with parallel agents, frontier-model recommendation, adversarial critique, and citation audit.

What is included

• SKILL.md — End-to-end research workflow: operating modes, sub-question decomposition, parallel source collection, evidence ledger, skeptical synthesis, and structured report with citations
• Native mode — no external API keys required; uses built-in WebSearch/WebFetch
• Dense/frontier mode — recommends the strongest available model and runs a wider parallel agent topology when the host platform supports subagents
• Sub-question decomposition into 3-5 concrete research threads
• Evidence ledger with source quality notes and claim status
• Required confidence/gaps section
• Skeptical conflict handling for contradictory sources
• evals/evals.json — realistic test cases with assertions for trigger accuracy and workflow correctness
• evals/trigger-eval.json — trigger and non-trigger examples for description optimization

Typical invocation

• "Research the best approaches to implement OAuth2 in Node.js."
• "Deep-research the current state of vector databases for RAG."
• "Research pricing models for SaaS developer tools."
• "Synthesize the latest findings on AI agent evaluation frameworks."
• "Use dense deep-research with a frontier model to create a Perplexity-like evidence report."

Operating modes

Native research

• Default mode.
• Uses built-in WebSearch and WebFetch.
• Decomposes complex topics into 3-5 sub-questions.
• Runs parallel ResearchScout agents where supported.
• Produces citations, evidence ledger, and confidence/gaps.

Dense / frontier research

• Activated when the user asks for maximum depth, exhaustive research, frontier models, adversarial review, or "Perplexity turbinado" style output.
• Recommends the strongest available model or model class for the task.
• Adds specialized research roles such as PrimarySourceHunter, ContrarianScout, RecencyScout, CitationAuditor, and SynthesisJudge.
• Falls back to native mode if frontier tooling or subagents are unavailable.

What's New in v2.3

• Research modes — Standard / Verbose / Exhaustive dial with source quotas, word targets, and section requirements per mode
• Friendly agent names — Scouts renamed (🔍 Scout Alpha/Beta/Gamma, ⚔️ Devil's Advocate, 📡 Trend Watcher, 🏛️ Source Hunter, 🔬 Citation Auditor, ⚖️ Synthesis Judge) with live status notifications on launch and completion
• Methodology section — Mandatory audit trail: queries used/discarded, inaccessible sources, filters applied, known limitations
• Source deep-dives — Per-source analysis in Verbose/Exhaustive: methodology, sample size, geo scope, conflict of interest, independence
• Sub-question expansion — Four-layer structure per sub-question: supporting evidence → counter-evidence → methodology critique → conflicting interpretations → conclusion with individual confidence score
• New report sections — Executive Summary, Background/Context, Stakeholder map, Timeline/Chronology, Devil's Advocate, Appendices (query log, rejected sources, paywalled sources)
• Purpose-based quote rule — Replaced length restriction with intent-based guideline: quote when exact wording is material, not by character count

What's New in v2.2

• Dense / Frontier Mode -- Maximum-depth research path with model recommendation, wider parallel topology, critique, rebuttal, and citation audit
• Sub-question decomposition -- Complex topics are broken into 3-5 concrete research threads before searching
• Evidence ledger -- Claims are tracked with source metadata, source quality, and evidence status
• Skeptical synthesis -- Conflicts and weak evidence are surfaced instead of smoothed over
• Confidence & gaps -- Reports close with confidence level, source disagreements, and missing coverage

What's New in v2.0

• Progress Tracking — 4-phase gauge bar (Query Decomposition → Source Collection → Synthesis → Report Generation) displayed during execution
• Error Handling — Handles search failures, paywalled sources, conflicting information, and narrow topics with fallback strategies
• EVals — evals/evals.json with 3 realistic test cases; evals/trigger-eval.json with 20 queries (10 trigger / 10 no-trigger) for description optimization
• Standardized description — SKILL.md description updated to Anthropic skill-creator format

---

Metadata

| Field | Value | |-------|-------| | Version | 2.3.0 | | Author | Eric Andrade | | Created | 2026-02-20 | | Updated | 2026-05-10 | | Platforms | GitHub Copilot CLI, Claude Code, OpenAI Codex, OpenCode, Gemini CLI, Antigravity, Cursor IDE, AdaL CLI | | Category | research | | Tags | research, search, analysis, synthesis, citations | | Risk | safe |