Decision engines, disposable desktops and a fresh crop of agent skills

Fastest-growing this period

Counts come from GitHub's official stargazers/history endpoint: the stars gained on each day of the window. Project briefs and the reading notes below are editorial additions.

1. 1

   browser-use/jev-ultrafast : A browser agent that picks each click and typed word in one model call

   Python

   Give it one natural-language goal and it renders the page as a numbered table of controls. TypeSafe's Jev picks the operation and target element in one request, and a small LLM writes text only when a field needs filling, so no site scripts or screenshots drive the loop. The README's measured demo finds flights from Zürich to London in about seven seconds.

   +3,514
2. 2

   cloudflare/security-audit-skill : Turn your coding agent into a security auditor that cross-checks its own findings

   JavaScript

   Installs as a skill that makes your coding agent audit a codebase for vulnerabilities. It maps the architecture and trust boundaries first, sends isolated hunter agents after the uncovered areas, then gives every candidate to a fresh verifier that tries to disprove it, so a finding only reaches the report with a complete source trace.

   +2,594
3. 3

   robbietilton/Compositor : A free open-source image editor that mirrors the Photoshop compositing workflow

   Swift

   A free, open-source image editor for macOS built around the Photoshop compositing workflow: layers with blend modes and masks, adjustment layers, content-aware fill and non-destructive transforms that keep full resolution. Import JPEG, PNG, HEIC and TIFF, export with a live preview, and modify any feature through the included Xcode project.

   +1,860
4. 4

   NandhaKishorM/laya : Answer typed questions about any text in one 33-millisecond forward pass

   Python

   A non-autoregressive decision engine that reads text, an email or a JSON document and returns a typed answer, a choice, a score or a yes/no flag, in one 33-millisecond forward pass. Three checkpoints cover English, over one hundred languages and typed workflows, and a built-in router picks one per request. Nothing is generated, so there is nothing to parse or hallucinate.

   +1,463
5. 5

   deepseek-ai/deepseek-harness : Explore DeepSeek's building blocks for an extensible agent

   TypeScript

   DeepSeek's agent runtime built around replaceable plugins, with a web interface for running it. Useful for developers exploring how to compose an agent's capabilities; it is a developer preview, so expect interfaces to change as you experiment.

   +1,136
6. 6

   trycua/cua : Open-source computers for AI agents: cloud desktops, local VMs, benchmarks

   HTML

   An open-source stack that hands AI agents real computers to work with. Spin up an isolated cloud desktop for one task and tear it down afterwards, run lightweight macOS or Linux virtual machines on Apple Silicon, inspect and operate apps across operating systems with a single driver, and score agents on simulated tasks with the bundled benchmark suite.

   +1,003
7. 7

   hypit-ai/hypit : Let your coding agent turn one reference video into a hundred variants

   TypeScript

   A skill that gives coding agents a language for making video: drop in a reference clip and the agent rebuilds it as a complete workflow covering footage, captions, B-roll and effects, timed to words instead of seconds. You can also start from templates or a plain description, and a workflow can compile captions and motion graphics without calling a generation model.

   +997
8. 8

   alibaba/open-code-review : Put AI review comments on the lines that need attention

   Go

   Reviews code changes with an AI model and places comments on the relevant lines. File selection and comment positioning are handled by ordinary program logic, useful when you want review suggestions that are easier to locate and check in a large diff.

   +958
9. 9

   tamaratran/fast-jev-compaction : Compaction that scores each tool call and drops only the stale ones

   TypeScript

   Replaces Claude Code's compaction summary with per-call decisions. The whole conversation goes to TypeSafe's Jev, every tool call and result gets a keep-or-drop score, stale ones are deleted or truncated, and everything kept stays verbatim and in order, so exact file paths and error messages survive. Ships as both a Claude Code plugin and an npm library.

   +919
10. 10

    affaan-m/ECC : Give your coding agent a plan, tests and a memory

    JavaScript

    A collection of coding-agent workflows for planning, tests, review and remembering useful context. If your assistant jumps straight into edits or forgets earlier decisions, this gives you reusable routines to adapt to your project.

    +918

Ranking window: 2026-09-19 UTC, full day.