What problem does it solve? Answering complex questions from memory alone risks hallucinated facts and missing sources. This Skill runs a structured, evidence-grounded research workflow that decomposes a question, retrieves real web sources, verifies every claim against its citation, and produces a cited report. ## Core Features & Use Cases - Iterative Retrieval Loop: Decomposes the question into subtopics and runs up to 4 rounds of web_search and web_fetch, maintaining an evidence ledger keyed by subtopic. - Parallel Synthesis & Verification: Dispatches parallel sme workers to synthesize findings, then two reviewer agents verify each claim as SUPPORTED, UNSUPPORTED, OVERSTATED, or CONTRADICTED, with a critic challenging high-stakes claims. - Cited Report Output: Produces a report or brief where every load-bearing claim carries a source citation, disagreements between sources are surfaced, and ungrounded subtopics are marked UNVERIFIED. - Use Case: Ask "What are the current best practices for PPO training stability in 2025?" and receive a structured report with verified claims, confidence levels, and a full source list. ## Quick Start Ask the agent to run deep research on a question, for example: run deep research on the current state of indoor drone navigation benchmarks with standard depth and a full report.
