GitHub 2026-02-16 Weekly Trending Projects

Shannon is a white‑box autonomous AI pentester that, from source‑aware analysis, constructs and executes real exploit PoCs and produces reproducible reports—enabling security and development teams to move pentesting from a yearly event to a continuous process.

GitHub Agentic Workflows enables natural-language Markdown definitions of AI agentic workflows run in GitHub Actions with layered guardrails for controlled repository automation; evaluate license, community maturity, and auditability before production deployment.

LangExtract is a Python library for production and research that uses LLMs and example-driven constraints to turn unstructured text into source-grounded, auditable structured data, with interactive HTML visualization for human review and large-scale document workflows.

Claude‑skills supplies 66 specialized skills and 9 workflows for Claude‑based conversational programming, emphasizing context engineering and progressive disclosure—ideal for full‑stack teams using Claude as an expert pair programmer; however, there are no formal releases and contributor activity is low, so evaluate maintenance and compatibility risks before adoption.

Monty: Rust minimal secure Python VM for safe LLM code execution.

Dexter is an autonomous multi-agent system designed for professional financial research that plans and executes data retrieval, computation, and self-validation—suitable for quant research and enterprise finance automation.

GitButler uses virtual branches to manage multiple uncommitted changes in a single working directory, simplifying splitting, pushing and review workflows with AI-assisted commits and a cross-platform desktop client — targeted at developers and teams needing parallel context management and higher multitasking efficiency.

Pi is a developer-focused monorepo toolkit offering a unified multi-provider LLM API, agent runtime, coding agent CLI, terminal and web UI components, and tools to manage vLLM GPU pods—suited for teams with TypeScript/Node skills building and deploying custom AI agents.

A Chinese-focused multi-agent LLM financial decision framework integrating multi-vendor models and A/HK/US data—well suited for research and enterprise prototyping but requires careful review of license, maintenance, and production/compliance risks.

Chrome DevTools MCP exposes Chrome DevTools to AI coding agents via an MCP server, enabling automated browser control, deep debugging and performance analysis; however, the repository lacks contributors, commits, and releases, so evaluate maintenance and compliance risks before production use.

Aggregates numerous LLM application examples and links with emphasis on Agents and RAG practices; well suited for learning, comparison and rapid prototyping, but users should watch entry quality and maintenance stability.

LiteBox is a security-focused library OS that uses North/South abstractions to provide portable sandboxed execution in kernel and user modes; it is suited for cross-platform isolation, platform integration, and security research in experimental deployments.

For developers and teams using Claude Code: automatically capture, semantically compress, and persist session outputs, then inject relevant context into future sessions to improve continuity and retrieval efficiency.

A Claude Code-focused compound-engineering plugin and CLI that defines cyclic workflows, exposes a plugin marketplace, and offers experimental OpenCode/Codex conversions; suitable for engineering teams with self-managed tooling to trial in controlled environments.

PAI is a personal AI infrastructure template emphasizing modular skills, history capture, and observability with a backend-agnostic design—enabling technically proficient users to build their own personal AI systems.