elephant-goldfish-model jumps from 12th to first as new agent apps fill the top 10

A feature-planning skill takes first place in week 44

elephant-goldfish-model moved from 12th to first, adding 8,650 GitHub stars for a total of 9,808. It is a skill that helps teams plan software features with AI agents and stress-test the design before coding starts. Founders and vibe coders who brief an agent on a feature can use it to check the plan first. The board turned over hard this week. Every project in last week's top 10 fell out of it, including Clean Your TV, which went from first to 19th.

Two new agent apps sit right behind it. agent-desktop entered at second with 5,800 stars gained (7,622 total). It is a command-line tool that lets AI agents read and operate desktop apps, aimed at people who want reliable computer control. Orkas, also new, landed at fourth with 5,527 stars. It turns your goals into finished work and keeps ongoing projects moving, built for business owners and solo founders on macOS and Windows.

android-ad-cleaner climbed from 20th to third on 4,666 stars. It is a desktop app that turns off ads, hot-search feeds, and marketing notifications on Chinese Android phones without root. The audience is narrow. If you own one of those phones and want the clutter gone, it is the most direct tool on this week's board.

New agent apps and fast climbers from week 44

OpenBot made the largest jump of the week, from 88th to sixth, on 3,590 stars. It is a desktop app that runs persistent AI teammates on your own computer, each with its own workspace and task queue. ThinkRail rose from 46th to fifth with 5,015 stars. It is an agentic IDE that runs the pi coding agent in isolated workspaces for vibe coders. Hallmark went from 35th to 14th with 10,950 stars, a skill for building web pages that do not look AI-generated. claude-seo rose from 74th to 33rd.

There are 61 projects new to the top 100, and several target daily work outside code. OpenDuet (ninth) is a self-hosted agent that sorts email and calendar, writes a morning brief, and asks before it acts. OpenMausBot (10th) runs a team of AI bots in a chat app, each with its own model and cloud computer, pitched as an open-source alternative to Grok Bot. Digital Human Koubo (11th) turns one photo and one sentence into a talking-head video for creators and marketers. Floorplan (12th) draws a plain-English map of an app's code for people who built it with AI help.

Survival LLM entered at 13th and had the week's top Hacker News post, with 543 points and 203 comments. It answers emergency and survival questions offline on your own computer and cites its sources, for households and anyone without reliable internet. DeepTutor, new at 15th with 11,167 stars, is a self-hosted tutor that turns your own documents and notes into personalized lessons. Elsewhere on Hacker News, PaperPilot drew 236 points and Hob drew 177.

Skills and plugins outnumber agent apps in the top 50

Skills and plugins held 24 spots in the top 50, against 20 for agent apps. MCP servers and connectors had 2, and agent builders or no-code platforms had 1. The top project, elephant-goldfish-model, is a skill. So is tstack, new at seventh with 2,497 stars, which takes a pull request from change to merge for people using Claude Code or Codex.

By task, coding still leads but slipped. Projects that write, fix, and review code went from 18 to 16 in the top 50. Tools that connect apps and automate repetitive work rose from 10 to 12, and project, task, and note management rose from 7 to 9. Document summarizing went from 3 to 6, and turning long videos into short clips went from 4 to 6. Desktop control held steady at 10.

Builder tools for people making their own agents

Open Ontologies tops the builder board as a new entry with 1,361 stars. It plans changes to a production ontology and checks their consequences, for developers and data teams. Hindsight climbed from 11th to second on 4,065 stars. It is a memory system that helps agents learn over time instead of only recalling chat history. RocketRide, new at third with 1,920 stars, builds, debugs, and runs AI workflows inside a code editor.

mcp-server-gateway routes agent tool calls through a secured gateway for enterprise teams running locally with Docker (368 stars). MasterAgent is a C++ framework that turns requests into validated tool actions locally, for developers putting agents on devices (455 stars). Both are new to the board.