Show HN: TrustMesh – Open-source reputation layer for AI agents (github.com)

TrustMesh is an open-source reputation layer for AI agents that plugs into the Agent2Agent (A2A) protocol to help agents evaluate peers before delegating work. It provides portable trust scores across any A2A-compatible platform, simple three-line integration, and an API/SDK-backed service for registering agents, fetching trust-scores, logging interactions, leaderboards and immutable audit trails. The project targets a practical gap in agent ecosystems—knowing whether a hired agent will succeed, leak data, or disappear—and is positioned as an extendable, community-driven standard (v0.1, MIT-licensed) with PostgreSQL support planned for production. Technically, TrustMesh uses a Beta-Binomial Bayesian model with a neutral prior (0.5) to handle cold starts, time-weighted interaction updates so recent behavior matters more, and growing confidence as interaction counts rise. Trust is computed as trust_score = α / (α + β) where α = prior_successes + weighted_successes and β = prior_failures + weighted_failures; examples show a neutral new agent at 0.5, jumping to ~0.83 after five successes and stabilizing near 0.89 with many interactions. v0.1 runs a REST service (SQLite default), enforces API keys and rate limits, and logs disputes; the repo invites contributors for SDKs (TypeScript/Rust), dashboard UX, dispute resolution and multi-dimensional trust.