ndpvt-web

Siddhi — Data skill for Claude Code

Data community

SIDDHI: The first desktop computer-use agent with a runtime self-evolving skill pipeline.

How to install Siddhi

This entry records only its repository, not the path inside it, so there is no exact command to give. Open ndpvt-web/siddhi and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Siddhi does

SIDDHI: The first desktop computer-use agent with a runtime self-evolving skill pipeline. Autonomously promotes experience into reusable strategies and executable tools -- without model retraining.

Alternatives in Data

OpenViking — Self-evolving Context Database for AI Agents 33.3k ★
Skill Content Pipeline — Extract patterns and anatomy from URLs — use to reverse-engineer content strategies from live pages 2.8k ★
Claude Deep Research Skill — Enterprise-grade deep research skill for Claude Code with 8-phase pipeline, source credibility scoring, and au 1k ★

README

SIDDHI

Self-Evolving Desktop Computer-Use Agent .

Siddhi: Self-Improving Desktop Intelligence

It sees your screen. It clicks, types, and navigates. And unlike every other agent --
*it remembers what worked, learns from what didn't, and gets better every time it runs.*

**Nivesh Dandyan** · [LinkedIn](https://www.linkedin.com/in/nivesh-dandyan/) · [@ndpvt-web](https://github.com/ndpvt-web)
JC STEM Lab of Cyber Security, The University of Hong Kong

Every computer-use agent today -- Anthropic CUA, OpenAI Operator, Google Mariner -- starts from a blank slate every session. They burn the same tokens re-discovering the same workflows, paying the same cost, making the same mistakes. **Siddhi breaks this cycle.** It is a fully autonomous macOS desktop agent that watches itself work, extracts patterns from its own behavior, graduates successful patterns into reusable strategies, and even forges entirely new tools from repeated workflows -- all at runtime, stored as inspectable JSON, with **zero model fine-tuning**. After just a few tasks, Siddhi completes familiar workflows in 2 iterations instead of 8, at 1/14th the cost, in 1/3rd the time. It doesn't just use your computer. It masters it.

This is not a research prototype running on synthetic benchmarks. Siddhi operates on a real macOS desktop -- launching apps, navigating Safari, composing emails, managing files -- with a hybrid vision system that comb