GitHub 2026-02-14 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.

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.

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.

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.

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.

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.

Grok2API is a FastAPI-based Grok-compatible multimodal API gateway that supports streaming chat, image/video generation, token-pool concurrency and automatic load balancing—suitable for self-hosting; however, license ambiguity and limited community contribution require assessment of compliance and maintenance costs.

Prompt Optimizer helps teams optimize and test prompts across models, supporting image generation and local/container deployment to improve output consistency and speed up prompt engineering.

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.

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.

LikeC4 provides a C4-like modeling language and toolchain to generate and collaboratively maintain live architecture diagrams from code, suited for teams that need code-document synchronization and custom notation; however, contributor and release history should be verified to assess risk.

R&D-Agent is a multi-agent automation platform for ML engineering and quantitative research; it coordinates agents to mine data, generate models, and alternate optimize strategies to reduce loop time, improve engineering efficiency, and enhance reproducibility.