Article URL: https://btao.org/posts/2026-08-17-i-dont-enjoy-the-internet/ Comments URL: https://news.ycombinator.com/item?id=49341551 Points: 52 # Comments: 37
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
A developer tutorial published on DEV Community walks through building plugins for DeepSeek Harness (dsh), an open-source AI tooling framework driven by a vendored version of Cordis. The guide explains that dsh has no privileged core — model adapters, tool registries, session logs, and the agent main loop are all plugins, making every component replaceable via configuration. Developers are shown how to export the four standard plugin fields (name, inject, Config, and apply) and how to register tools and event hooks in a reversible, hot-reload-safe manner. The tutorial covers setting up a local clone of the GitHub repository, running type checks and unit tests without an API key, and eventually packaging a plugin for formal release. It also details the layered loading order dsh uses to compose its runtime plugin tree from profiles, patch files, and command-line overrides.
On August 17th, an engineer's new alert rules flagged two separate IO pressure incidents on a four-core ARM server, neither of which triggered any error from Kubernetes, ArgoCD, or Helm. Investigation revealed the heaviest disk reader on the node was a restarted Jaeger container — the third victim of OOM kills in 16 hours — thrashing the page cache rather than actively requesting reads. The node's memory limits were set at 110 percent of capacity, while the scheduler only reads memory requests, which stood at just 41 percent, giving it no reason to stop placing pods. Each layer of the stack answered its own narrow question correctly, but the composite picture they painted was entirely misleading. The incidents illustrate how overcommitted resources and siloed health signals can mask systemic instability until low-level kernel metrics are examined directly.
A developer partnered with IBM's AI assistant Bob to build a working Loop Engineering application from scratch, moving beyond theory to explore production-grade implementation. Loop Engineering replaces single-pass LLM prompts with iterative, self-correcting cognitive cycles structured around five stages: Perceive, Plan, Act, Observe, and Reflect. The sample project, an Autonomous Research Summarizer, uses a two-tier nested loop architecture where an outer orchestrator manages high-level research phases and an inner engine handles atomic cognitive steps. State is tracked across three distinct memory types — volatile working memory, immutable episodic records, and persistent semantic memory — ensuring deterministic termination and quality control. The walkthrough covers code, design patterns, and schemas, offering a concrete reference for developers looking to implement autonomous agent frameworks.
A developer has created SonicLab's Equalizer, a free, fully client-side 10-band audio equalizer that runs entirely in the browser without requiring file uploads or user accounts. The tool leverages the Web Audio API's BiquadFilterNode with a peaking filter type to boost or cut across ten frequency bands ranging from 32Hz to 16kHz. Processing audio locally ensures user privacy, eliminates server costs, enables real-time playback preview, and allows offline use after the initial page load. The implementation chains ten filter nodes in series and uses a standard AudioContext for live preview alongside an OfflineAudioContext for faster-than-real-time audio export.
Discussion (0)
Log in to join the discussion and vote.
Log in