social-trust-manipulation-detector

# Your Trust Score Is Real. The Signals Behind It Are Manufactured.

> Helps identify when a skill's trust reputation is built on coordinated
> social manipulation rather than genuine community validation.

## Problem

Trust in agent marketplaces flows through social signals: upvotes, downloads,
comments, and follow counts. These signals are valuable precisely because they
aggregate distributed judgment — when thousands of independent users find a
skill useful and safe, their collective assessment carries real information.

The assumption of independence is the attack surface. A coordinated network
of accounts can manufacture the appearance of distributed consensus. A skill
with 500 upvotes from a bot network looks identical to a skill with 500
upvotes from 500 independent developers. The marketplace's reputation system
cannot distinguish manufactured trust from earned trust — and neither can
most agents that rely on reputation as a trust signal.

Social trust manipulation is the third pillar of the trust attack surface,
alongside technical attacks (code injection) and structural attacks (supply
chain compromise). It is the most scalable: a well-constructed sockpuppet
network can manufacture trust faster than any code-level auditing can catch
it, and the manufactured trust persists long after the network is dismantled.

Legitimate skills earn trust gradually, from a diverse user base, with
engagement patterns that correlate with actual skill utility. Manipulated
skills earn trust in coordinated bursts, from accounts with suspicious
creation patterns, with engagement that does not correlate with usage or
outcomes.

## What This Detects

This detector examines social trust integrity across five dimensions:

1. **Engagement velocity anomalies** — Does the skill's vote/download
   trajectory show natural growth curves, or coordinated burst patterns?
   Organic trust accumulates gradually; manufactured trust arrives in
   synchronize...