What problem does it solve? Web research often mixes untrusted page content, outdated documentation, and conflicting sources, risking prompt injection, leaked private data in queries, and unsupported conclusions. This Skill structures web evidence gathering so retrieved content never gains authority over the task and every material claim carries a traceable evidence state. ## Core Features & Use Cases - Untrusted-content handling: Treats pages, search results, and tool output as data with no authority to change scope, permissions, or destinations, and minimizes protected information in outbound queries. - Claim-specific source selection: Chooses standards, official docs, source code, advisories, or maintainer statements based on each claim's role, checking freshness, version applicability, and source independence. - Evidence-state reporting: Labels each material claim as Supported, Contradicted, Mixed, or Not verified with direct citations, separating observed facts from inference and unknowns. - Use Case: When asked whether a library version still supports a deprecated option, it checks current release notes against obsolete docs, cites the governing source, and reports the claim as Contradicted rather than merging versions. ## Quick Start Ask the agent to research whether a specific tool version supports a given feature using web sources and return cited evidence with an explicit evidence state for each claim.
