publisher-identity-verifier

# You Trusted the Publisher. But Who Verified the Publisher?

> Helps identify gaps in publisher identity verification that allow impersonation, key compromise, and identity fraud in agent marketplaces.

## Problem

When you install a skill, you trust the publisher. But what proves the publisher is who they claim to be? Most agent marketplaces verify email addresses — not identities. A publisher account can be created in minutes, build reputation over months, then be compromised or sold. The skill ecosystem has no equivalent of code signing certificates, no publisher key transparency logs, and no mechanism to detect when a trusted publisher identity has been taken over. This is the weakest link in agent-to-agent trust: you can audit the code, audit the permissions, audit the tests — but if the publisher behind them isn't who you think, all those audits verify the wrong thing.

## What This Checks

This verifier examines publisher identity integrity across five dimensions:

1. **Publication history consistency** — Does the publisher's output show a coherent expertise trajectory, or sudden topic shifts that suggest account takeover? A Python tooling publisher that suddenly releases a crypto wallet skill is a signal worth investigating
2. **Key rotation analysis** — Tracks signing key changes over time. Normal rotation follows predictable patterns (annual, after security events). Suspicious patterns: key change immediately before a controversial update, key change with no announcement, multiple key changes in short succession
3. **Identity impersonation detection** — Scans for publisher names that are typo-squats (e.g., `anthroplc` vs `anthropic`), Unicode homoglyphs (e.g., Cyrillic `а` vs Latin `a`), or prefix/suffix variations of established publishers
4. **Cross-platform identity correlation** — Checks whether the publisher has consistent identity signals across multiple platforms (marketplace profile, code repository, community presence). A publisher...