Urgent.News

What's breaking now, across thousands of outlets.

Tech

Separate decisions from context, and let agents propose but not ratify

When I build with AI coding agents, I keep running into the problems classical projects have. A lot of talk and notes, and no clear focus on what the decisions actually are. Who said it? Was it approved? Was it just a brainstorm? A decision made on Monday meets an agent working from last week's version on Thursday. Notes in a rules file say what was written. A pile of remembered context says what…

When building software with AI coding agents, I have found that many projects suffer from a lack of clear decision-making processes. Conversations and notes are abundant, but it's difficult to determine which decisions are being made and who has the authority to make them. A decision made on Monday may be working with an outdated version from a week ago.

This can lead to confusion and errors, as work is built upon the wrong decisions. Smeldr is a solution that separates context from reasoning and decision-making. It holds all relevant information, such as chats, notes, code, and documents, as references. Smeldr tracks the decisions that are currently in force and their lineage, governance, and traceability.

This allows users to reason from them and make new decisions with the necessary approvals. The process begins with a decision being proposed and then ratified by a person with the appropriate authority. Once ratified, the decision cannot be edited. Changes are made by creating a new decision that replaces the old one, with the old decision staying on record.

Permits for ratification are set at different levels, such as the entire organization, a domain, or an area within a repository. When a decision is ratified, it becomes the basis for further work, and any changes or updates are built upon it. This approach ensures that work is always made based on the current and correct decisions, reducing confusion and errors.

The server enforces these rules and will not allow an agent to ratify a decision, only a human approver. This ensures that the right person is responsible for making the final decision. The system is self-hosted, runs on the user's own instance, and can be accessed from different operating systems, such as Linux and Windows. It integrates with tools like GitHub, where pull requests can be approved based on the current decisions in force.

This system makes it easy to track and manage decisions, ensuring that work is always built upon the correct and most up-to-date information.

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 Tech

I wanted my PC in Home Assistant without an MQTT broker, so I wrote my own integration

I have a Snapdragon laptop, a desktop PC and a Home Assistant box on a Raspberry Pi. I wanted the computers in Home Assistant like any other device: sensors, power buttons, notifications, media.

  • Developer created Deskmate integration for Home Assistant without MQTT broker.
  • Deskmate uses encrypted WebSocket connection with X25519 key exchange per session.
  • Features include file sharing, screen time management, sensor monitoring, and power management.

Technicolor CGM4140COM (Xfinity XB6) Login: 10.0.0.1 and the App-First Fix

If you are an Xfinity customer, the black tower sitting next to your TV is probably a Technicolor CGM4140COM — the gateway Xfinity ships as the XB6.

  • Access Technicolor CGM4140COM gateway admin page at 10.0.0.1
  • Use default username "admin" and password "password" to log in
  • App-first fix involves using Xfinity app for login issues

How to Tell Whether the Model Behind Your API Got Worse

A serving bug can break a few percent of a model's answers while every eval score you track stays flat. Counting the answers that fail a simple check catches it with thirty times less traffic.

  • Count failed responses to detect model degradation
  • A single bad server can break ~1/3 of model's answers
  • Compare failure rate to normal rate to spot issues

More from Monday 16 November →