Empire LLM Plugin Keeps AI Code Review Advisory, Not Autonomous

·1 views

A new plugin called Empire LLM for Codex aims to solve a common problem with multi-model AI code review workflows by enforcing a strict boundary between advice and authority. The plugin routes requests to external AI models via OpenRouter, applying capability, privacy, cost, and availability policies before returning a response labelled with its provenance and cost. Crucially, the external model cannot edit files, execute code, approve its own suggestions, or silently merge results — all decisions remain with Codex. The tool formalises what many developers already do informally, replacing ad hoc copy-pasting between AI chat tabs with a structured, policy-governed review loop. Its core design principle is that external models contribute evidence, not authority, addressing real failure modes seen with over-permissioned AI agents.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer Builds CheckForge, an Uptime Monitoring SaaS Using Fastify and Cloudflare

A developer has independently built CheckForge, a production-ready uptime monitoring platform designed to track websites, APIs, and SSL certificates. The platform uses Fastify for the backend, Cloudflare Workers to execute health checks, and Supabase to store results. When a failure is detected, CheckForge automatically creates an incident and dispatches alerts via Email, Slack, Discord, or Webhooks. The project was built over several months as a self-challenge to go beyond typical portfolio work and tackle real backend engineering problems like background scheduling. CheckForge is currently live at checkforge.in, with multi-location monitoring among the features planned for future development.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Why nested rounded corners look off and how to fix them with one CSS rule

When two rounded rectangles share the same border-radius value, their curves appear uneven because the gap between them widens at the diagonal. The fix relies on a simple geometric rule: the inner element's radius must equal the outer radius minus the spacing between them. CSS custom properties make this easy to maintain by using calc() to derive the child radius automatically from the parent's padding and border width. Developers should also use max(0px, calc(...)) to prevent negative values, which CSS silently discards rather than flagging as errors. When horizontal and vertical padding differ, the slash syntax in border-radius allows separate values for each axis to keep both curves visually parallel.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Supervisor Pattern Fixes AI Support Bot Confusion in Multi-Task Deployments

A developer building a customer support bot with LangGraph and MCP encountered production failures where the bot mishandled queries by confusing orders, shipments, and returns. The root cause was a single overloaded agent trying to manage query understanding, data retrieval, and response generation simultaneously. To resolve this, the developer adopted the Supervisor pattern, which uses a manager agent to delegate incoming queries to specialized worker agents. Three dedicated worker agents were created — one each for orders, shipments, and returns — each handling only its designated task. A key caution noted is ensuring worker agents are properly scoped, as misrouted delegation can still cause failures if agents lack the capability to handle assigned tasks.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

P2C Load Balancing Outperforms Round-Robin for Java Virtual Thread Workloads

A technical analysis argues that traditional round-robin load balancing degrades performance in high-concurrency Java environments where virtual threads can handle tens of thousands of simultaneous requests per instance. Round-robin fails because it assumes uniform request processing time, while real workloads face asymmetric I/O bottlenecks and microsecond-level queue spikes that host-level metrics cannot capture. The Power-of-Two-Choices (P2C) algorithm, combined with Exponentially Weighted Moving Average (EWMA) latency tracking, is proposed as a superior alternative for routing decisions. P2C works by randomly sampling two upstream nodes, scoring each based on active virtual threads and recent latency, then routing the request to the lower-scoring node. This approach achieves near-optimal load distribution in constant time without the lock contention or herd-effect failures associated with naive least-connections strategies.

0 comments Read more at DEV Community