Urgent.News

What's breaking now, across thousands of outlets.

AI

Beyond Prompt Engineering: Governing Autonomous Agents with the Domain Claim Protocol (DCP)

Software is moving from fixed call chains toward systems that choose tools, services, and other agents as they work. That flexibility is useful, but it makes a basic engineering question harder to answer: What is each component actually allowed to do? An API can tell an agent how to call a function. A JSON schema can tell it what fields to send. Neither necessarily tells it who owns a business…

The Domain Claim Protocol (DCP) is an open draft specification aimed at governing autonomous agents in software systems. It addresses the challenge of making explicit agreements between components before they are assembled, as current methods like API descriptions and JSON schemas do not fully capture who owns business decisions, which providers are involved, or how failures should propagate.

DCP provides a structured description of a component's role, including its domain, capabilities, required providers, explicit refusals, conflict rules, integration ports, and declared faults. This allows machines and humans to understand a component's boundaries before it executes, eliminating "agent spaghetti" that occurs when implicit boundaries and drifting context cause failures that are difficult to trace.

A DCP claim includes details such as the component's URI, name, kind, version, publisher, domain summary, concerns, reconciliation boundaries, refusals, offers, interface details, and metadata. The claim identifies the component's role, who owns business decisions, which providers it depends on, and how failures should be handled. It also separates the workflow into three planes: component claim negotiation, composition of components under relevant context and policy, and deterministic runtime results.

The DCP specification provides schemas, vocabulary, and examples, with an ongoing review process that welcomes corrections and concrete cases. It is licensed under Apache 2.0 and forms part of the broader Unfurl ecosystem, which includes the DCP specification, Java library, runtime binding tool, and a human harness for inspecting and questioning assemblies before they run.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

Stop Burning AI Tokens: Why 'kern' is the Local Code-Intelligence Engine We've Been Waiting For , Face off Graphify vs Code Review Graph vs Kern

If you are using AI coding assistants like Claude Code, Cursor, or Codex , you already know the sinking feeling of watching your context window evaporate.

  • kern is a local code-intelligence engine with pure-Go implementation
  • 500-2,500 token usage reduces AI coding assistant context window exhaustion by 45%-80%
  • kern employs zero-telemetry, local-first policy with sandboxing for enterprise security

Building a Serverless AI Tech-News Bot with Node.js, Gemini and GitHub Actions

I wanted to build a project that combined several technologies into one practical automation pipeline: Node.js Web scraping RSS Offline NLP Generative AI Telegram Discord GitHub Actions Vercel The…

  • Chaos Routine Bot generates AI-assisted tech news briefings daily
  • Node.js framework with Gemini 2.5 Flash, JSDOM, and Compromise library
  • Automated workflow via GitHub Actions at 02:30 UTC, Vercel CLI deployment

More from Monday 21 September →