TrustNet

TrustNet helps you decide if an AI agent should be allowed to do a specific action, and explains why. it is an initial implementation of a web-of-trust network and trust score propagation, where if agent A trusts agent B and agent B trusts C, it is expected that A has some trust to C.

Think of TrustNet like task-based references for AI agents:

1. You decide whose judgment you trust.
2. Those trusted people or agents rate agents for specific tasks.
3. TrustNet checks those ratings and returns a decision with a reason.

Important: trust is scoped by task.
An agent trusted for payments is not automatically trusted for code-exec.

What TrustNet Returns

At core, TrustNet returns a score and an explanation. Your policy uses that score to decide:

• allow
• deny

If you use the OpenClaw plugin, you can also enable a human-in-the-loop ask step before execution.

Why This Matters

• No single global reputation score.
• Decisions are explainable ("which trust links were used").
• Technical teams can verify results against committed roots and proofs.

Simple Example

Question: "Can this agent send a payment?"

1. FinOps trusts CFO for payment-related judgments.
2. CFO positively rates the agent for payment tasks.
3. Policy says payment actions need score >= +1.
4. TrustNet returns allow with a "why" trail.

For Engineers

• Spec (v1.1): docs/TRUSTNET_v1.1.md
• Test vectors: docs/Test_Vectors_v1.1.json
• Server smoke test guide: docs/Server_Smoke_Test.md
• Chain smoke test guide: docs/Chain_Smoke_Test.md
• Base Sepolia rehearsal: docs/Base_Sepolia_Dress_Rehearsal.md

Components

• Indexer: crates/indexer
• API: crates/api
• CLI: crates/cli
• OpenClaw plugin: plugin-openclaw/README.md

Core Commands

# format
cargo +nightly fmt --all

# lint
RUSTFLAGS="-D warnings" cargo +nightly clippy --workspace --all-features --locked

# tests
cargo nextest run --workspace