Two of today’s entries are built on the same engine, TypeSafe’s Jev. browser-use/jev-ultrafast renders a page as a numbered table of controls and lets Jev choose the operation and target in a single request, with a small LLM writing text only when a field needs it; the README’s measured demo finds flights from Zürich to London in about seven seconds. tamaratran/fast-jev-compaction aims the same idea at Claude Code’s context window: every tool call and result gets a keep-or-drop score, stale ones are deleted or truncated, and the text that stays is verbatim, so exact paths and error messages survive a long session.

AI-assisted review keeps stacking up. cloudflare/security-audit-skill maps a codebase’s trust boundaries and hands every candidate finding to a fresh verifier that tries to disprove it, while alibaba/open-code-review runs deterministic checks first and pins each AI comment to the exact line it refers to. Tencent/BrowserSkill lets an agent borrow tabs in the browser you are already logged into, hypit-ai/hypit rebuilds a reference video as a word-timed production workflow, and deepseek-ai/deepseek-harness is a plugin-based agent runtime for your own experiments.

The non-agent picks are the quiet reads. eternity4719/HowToLiveBetter records each life suggestion with what you spend, what you get back, and how strong the cited evidence is. bilawalsidhu/gods-eye-view tracks live aircraft and ships on an interactive 3D globe, and tt-a1i/archify turns an agent’s description of a system into architecture diagrams you can compare between versions.