danium

Fanout Audit — Security skill for Claude Code

Security community

Audit content for AI search retrievability by simulating query fan-out and checking passage extractability.

How to install Fanout Audit

This entry records only its repository, not the path inside it, so there is no exact command to give. Open danium/fanout-audit and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Fanout Audit does

Audit content for AI search retrievability by simulating query fan-out and checking passage extractability. Runs offline, invents nothing.

Alternatives in Security

Query Token Audit — Audit token security to detect scams, honeypots, and malicious contracts across BSC, Base, Solana, and Ethereu 483 ★
Gsc Audit — Run a full Distribb SEO audit from Search Console + on-page + competitor + cannibalization + topic clusters 195 ★
List All Global Skills Sorted By Provider Location — asm list --scope global --sort location asm search "code review" --json asm inspect my-skill asm audit --yes a 185 ★

README

fanout-audit

fanout-audit is a markdown-only agent skill that checks whether a page answers likely AI-search sub-queries in passages that survive being lifted out.

It audits content for retrievability by an LLM grounding pass. It runs offline by default, has no built-in fetcher, and does not rewrite the page. A host fetcher may be used only when the user explicitly permits it and returns the page's authored text rather than a summary.

The caveat, up front

fanout-audit simulates a plausible query decomposition rather than reproducing Google's, because the decomposition itself — which sub-queries get issued for a given question, and how they are chosen — is disclosed nowhere. What Google has published has grown from one sentence to a few: Search Central now documents that both AI Overviews and AI Mode "may use a 'query fan-out' technique — issuing multiple related searches across subtopics and data sources," and Google states that Deep Search "can issue hundreds of searches." None of that reveals the decomposition.

Every simulated report says so. A tool that quietly implies it knows Google's fan-out is making a claim this skill cannot support.

Coverage gaps are a **content completeness signal**. They are not a ranking prediction.

What it does

**Module A — fan-out coverage.** Uses a user-named head query verbatim, observed queries when available, or an inferred head query otherwise. In inferred mode it simulates six to twelve well-fitted sub-queries from nine gated intent classes, using fewer when fewer than six distinct candidates exist and never padding. It checks whether the content contains a bounded passage answering each. Every sub-query lands in one of three states: `COVERED`, `CONTEXT-DEPENDENT`, or `ABSENT`.

If you have the real queries an engine issued — from an API's grounding metadata or an analytics surface that exposes them — paste them: fanout-audit audits those instead of simulating, and the report header says so. Observed queries beat