GitHub 2026-02-15 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.

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

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.

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.

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.

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.

Provides a discoverable, installable catalog of agent skills for Codex to help teams package and reuse task capabilities; the repository currently lacks license information and active maintenance, so it is best evaluated and customized within controlled environments.

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.

MiniCPM-o is an on‑device multimodal MLLM enabling low‑latency full‑duplex vision, speech and text streaming—suited for real‑time edge applications and research teams.

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.

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.

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.

QMD is a privacy-oriented on-device document retrieval tool that combines BM25 full-text, vector semantic search and LLM reranking — suited for individuals or teams needing offline, controllable search with structured outputs for agent workflows.