Trending archive for Oct 3, 2026

ninjahawk/livenerf

Watching Claude Opus for silent post-launch degradation — livenerf runs a frozen, append-only daily benchmark tracking whether Claude Opus 5.5 quietly gets worse after shipping, addressing recurring community claims of post-release 'nerfing'. It's a pre-registered, reproducible test for researchers and skeptics who want hard data instead of anecdotes about model drift.

Benchmark for tracking model capability after release.

streetcomplete/StreetComplete

Quest-based OSM editing now spans Android and iOS via KMP — StreetComplete turns OpenStreetMap contribution into simple on-site quests like naming a road or describing its surface, no tagging knowledge required. Built as a Kotlin Multiplatform app, it now targets both Android and iOS, works offline, and uploads answers directly as OSM changesets.

Easy to use OpenStreetMap editor for Android and iOS

reladraw/reladraw

A text language for diagrams where relative placement, not layout algorithms, rules — reladraw is a text-to-SVG diagram language that sits between auto-layout tools like Mermaid/Graphviz/D2 and manual editors like draw.io/Excalidraw. Instead of absolute coordinates or algorithmic layout, you describe relative positions ('below', 'right of'), giving custom arrangements without hand-dragging nodes or editing XML.

A text diagram language where you say where things go. Playground: reladraw.github.io/reladraw

ntfargo/Relapse-Exploit

Exploit chain for PS5 7.00 - 13.60

maanHimself/OpenDLSS-NR

A bit-exact, open Vulkan rebuild of NVIDIA's DLSS 5 neural renderer — OpenDLSS-NR reimplements DLSS 5's 71-block Swin/ViT generative rendering network in Vulkan, matching NVIDIA's build byte-for-byte at every block boundary, not just the final frame. A second WebGPU port runs the same network in-browser without tensor cores. Useful for graphics researchers and engine devs studying or porting the architecture; you must supply your own weights.

A Vulkan reimplementation of NVIDIA's DLSS 5 Neural Rendering network, bit-exact against the original.

nanaism/yomiyasu

An Agent Skill that cleans up AI-sounding Japanese prose — A Claude Code/Codex/Cursor skill that fixes the structural tics of AI-generated Japanese—vague metaphors, unclear subject-predicate links, inanimate subjects—rather than just banning buzzwords. Aimed at writers of technical docs, specs, PR descriptions, and internal reports who need naturally readable Japanese.

AI生成の日本語を自然な日本語へ推敲するAgent Skill / Agent Skill for Refining AI-Generated Japanese into Natural Japanese

coji/natural-japanese

An Agent Skill that strips AI stiffness out of Japanese work writing — This Claude/Codex Agent Skill rewrites AI-generated Japanese reports, minutes, and PR descriptions into natural phrasing without adding facts or changing certainty levels. Tests across 25 documents show bold-text overuse dropping from 551 to 13 instances, aimed at writers polishing AI drafts.

仕事の日本語を、読みやすくわかりやすく書く・直すための Agent Skill です。

modelcontextprotocol/experimental-ext-triggers-events

Designing push notifications for the Model Context Protocol — This incubation repo hosts early design work from the MCP Triggers & Events Working Group, aiming to replace polling or held-open SSE connections with a standardized callback mechanism for server-initiated updates. It's relevant for MCP SDK and spec contributors tracking proposals on subscription lifecycle, delivery semantics, and cross-transport event ordering.

Incubation space for the MCP Triggers & Events Working Group

mizchi/explainer

Skills for AI-to-human explanations, tool-verified claims included — A Claude Code plugin bundling five skills for writing explanations tailored to one reader's knowledge gaps, with claims and figures checked by tools rather than trusted blindly. Useful for anyone building AI-generated docs, crash courses, or chaptered guides where understanding—not writing speed—is the bottleneck.

midorisawa/Shirakawa

AI-powered post filtering for X, judged by meaning not keywords — A browser extension that hides X (Twitter) posts matching user-defined natural-language conditions, using the Jev AI model to judge context and intent rather than simple keyword matches. It offers adjustable thresholds, usage/cost tracking, caching to limit API calls, and JSON export of rules. Useful for anyone wanting finer control over their timeline beyond basic mute filters.

JevでX投稿を判定し、条件に合う投稿を非表示にするブラウザ拡張機能