What problem does it solve? Manual web research is slow, inconsistent, and prone to skipped steps, unverified claims, and lost progress when sessions crash. This Skill runs an enforced, code-driven research pipeline that plans, searches, reads, verifies, and synthesizes findings into a cited report. ## Core Features & Use Cases - Code-enforced research stages: Plan, diverge, challenge, search, triage, read, reassess, verify, and synthesize stages run as a Python pipeline with gates the LLM cannot bypass, plus per-source checkpointing for crash recovery. - Multi-provider search with quotas: Routes queries across Claude's built-in WebSearch/WebFetch, Tavily, Exa, Brave, Serper, and You.com with free-tier quota tracking and circuit breakers. - Quality controls: Source diversity enforcement, five-tier reputation assessment, claim verification with confirmed/refuted verdicts, falsifiability statements, and an optional adversarial review that scores report robustness. - Use Case: Ask for a deep dive on the AI job market in Toronto; the pipeline asks clarifying questions, gathers ~30 sources across diverse domains, verifies load-bearing claims, and delivers a detailed markdown report with citations and a YAML index appendix. ## Quick Start Ask the agent to research a topic, for example: research Anthropic's company background and recent funding for a job application.
