What problem does it solve? LLM-generated answers and reports often contain hallucinated URLs or citations that do not support the claims they reference. This Skill eliminates that risk by owning the URL-to-number mapping in a persistent ledger, so the model only ever emits small integer ids it was handed at retrieval time and never types a URL from memory. ## Core Features & Use Cases - Citation Ledger: Registers sources at retrieval time with idempotent, URL-normalized ids, then mechanically renders the Sources block in markdown, plain, footnotes, bibtex, or evidence styles. - Fact-Checking Mode: Attaches verbatim quotes to each source (rejected unless they literally appear in the fetched page text), flags model-knowledge claims with [unverified], and fails drafts whose cited sources carry no evidence. - Draft Verification: Checks that every [n] in a draft exists in the ledger, the Sources block matches the ledger, and citation coverage meets a configurable threshold. - Use Case: While writing a market research report from web searches, register each fetched page with the ledger, cite claims inline as [1][2], then run verify before delivery to guarantee no fabricated or mismatched references. ## Quick Start Ask the agent to research a topic using web search, register every retrieved source in the citation ledger, and write a cited answer ending with a rendered Sources list.
