What problem does it solve? Web search results often return hundreds of thousands of characters of raw HTML, navigation bars, and boilerplate that flood the AI's context window and degrade reasoning quality. This Skill processes search results inside Python scripts so only curated, filtered output enters the context window, achieving a 100-200x reduction in noise. ## Core Features & Use Cases - Context-Isolated Search: Pipes tvly search and tvly extract output through Python filtering code so only relevant print() output reaches the model. - Multi-Turn Research Workflows: Saves raw results to /tmp/ files, then explores titles, drills into selected sources, and follows leads across turns without re-fetching. - Query-Adaptive Filtering: Supports pipe mode, heredoc mode, and jq fallback, with filtering logic tailored to each query type (financial figures, technical specs, news dates). - Use Case: Researching a topic like solid-state battery commercialization — search broadly, triage 8 results by title and score, then extract only paragraphs mentioning specific companies and production timelines from the top sources. ## Quick Start Ask the agent to search the web for a current topic and extract only the key details, for example: search for the latest news on EU AI Act deadlines and summarize the important obligations.
