Signal Box

SignalBox aggregates social feedback from platforms like Twitter/X, classifies each item with AI (bug, complaint, praise, feature request, question), scores overall sentiment, and publishes the result on-chain via Chainlink CRE.

Any protocol can call getSentiment("chainlink") and gate actions on community health.

The CRE workflow is a 5-step pipeline:
(1) HTTP fetch community feedback from the SignalBox API,
(2) Claude AI classifies each mention with category, priority, and bot probability,
(3) second Claude call aggregates everything into a single score with risk assessment,
(4) conditional risk alerting if sentiment crashes,
(5) writes the verified report to SentimentOracle on Sepolia.

On top of that, SentimentGate is a consumer contract that gates DAO governance proposals based on sentiment score. SentimentSentinel uses Chainlink Automation to monitor oracle data freshness.

So you have two Chainlink services working together: CRE pushes data in, Automation watches data health.

CRE workflow in TypeScript (CRE SDK v1.0.9) using HTTPClient for 3 calls (API fetch + 2 Claude AI calls), EVMClient for on-chain writes, Runtime.report() for ABI encoding, Runtime.getSecret() for API keys, and consensusIdenticalAggregation for DON consensus.

Contracts are Solidity 0.8.24 built with Foundry, 31 tests across 3 contracts. The API is Python/FastAPI. Dashboard is vanilla HTML/CSS/JS.

AI runs on Claude Haiku 4.5 called through CRE's HTTPClient. I split the AI into two steps (classify each item first, then aggregate the classified data into a score) because the second model gets structured input instead of raw text, which gives more consistent results.