Weekly Trending

An agent skill that turns a codebase or system description into a validated, interactive architecture, workflow, sequence, data-flow, or lifecycle map packaged as self-contained HTML.

A cross-harness coding-agent plugin that encodes a reuse-first, YAGNI decision ladder before new code is written.

A self-organizing plain-Markdown second brain for Obsidian and Claude Code that reads, links, and files sources into a connected knowledge graph.

A self-hostable private search engine for browser pages and local files that offers controlled MCP access to AI assistants.

abi/screenshot-to-code is an open-source tool that converts screenshots, mockups, Figma designs, and screen recordings into functional code. It supports stacks including HTML/Tailwind, React, and Vue, and can run locally or through a hosted app.

CogniRoute studies how omni-modal models can route text, audio, and video evidence for social reasoning, where tone, gesture, timing, and contradiction can carry different parts of the meaning.

The paper tests whether injecting static-analysis facts as plain-text structural anchors makes code-agent navigation more stable and reproducible.

The paper frames benign fine-tuning as a potentially unstable displacement: a model may be pulled back toward earlier behavioral manifolds after a new alignment layer is applied.

Apodex 1.1 describes an agentic-work framework with explicit task state, provenance, delegated work, recovery, and a record of what actually happened.

PolicyGuard uses a dialogue-grounded sub-agent verifier to assess policy adherence during LLM-agent operation.

This paper models delayed verification in multi-agent LLM systems as a graph problem: false claims can spread while correction is in transit. It derives stability thresholds and corrector-placement rules, then tests predicted oscillations across five open models; grounded factual answering stabilised the tested systems.