This edition covers 14 October, the latest fully settled UTC day. morluto/rea leads for the second day running: hand your coding agent a website, an Electron app, a .NET assembly or a native binary, and it returns a readable explanation of how the software works, each claim tied to cited evidence from local analysis. debpalash/VoiceStudio rises to two, a fully local Electron studio where you clone or design a voice, dub videos with timed speech, dictate from a floating widget and turn long text into audiobooks, all through a local API and MCP endpoint that agents can call. storytold/photocraft holds three, a clean-room Photoshop written in pure Rust that opens and saves real PSD files and treats every edit as a command, so the same engine runs from the UI, a CLI, JSON or an MCP server.
The middle of the chart mixes explanation, free models and console hacking. wheresryan22/anatomy at four is a Claude Code skill that explains a technical idea by building it as an interactive isometric machine with live readouts, every number driven by the same model that moves the parts. Ebony-Vinyl/dsh-our-free-model at five is a plugin for the DeepSeek Harness shell that fills the model picker with free frontier models such as DeepSeek V4.1 Flash and Kimi K3, no login or API key required. storytold/artcraft at six is an IDE for AI image and video where you block the scene in 3D before generating, which keeps character identity and camera placement consistent between shots, and boykopovar/AnyPS5 at seven keeps relinking PS5 executables into native Linux and Windows programs with no emulation layer.
The tail turns the camera on the agents themselves. ifixai-ai/iFixAi at eight independently audits AI agents: after a guided setup picks the provider, judge and test suite, roughly 60 inspections across five pillars produce an A to F scorecard, runnable as a CLI wizard, in CI or as a plugin that lets the agent audit itself. mattpocock/skills at nine packages the small agent skills Matt Pocock uses for daily engineering, including one that asks detailed questions before implementation so vague requirements get clarified first. vxcontrol/pentagi closes the list as a self-hosted system for authorized penetration testing: describe a target and a goal in plain language, and a team of agents plans the work and runs more than 20 professional tools inside an isolated Docker sandbox. If you try only one today, point iFixAi at an agent you rely on and see where its scorecard says the weak spots are.