Open Source Python Agent Skills

Trail of Bits Skills Marketplace is a specialized Claude Code skills marketplace built by the security research firm Trail of Bits that focuses on enhancing AI-assisted workflows for vulnerability discovery, testing, and secure development. The repository groups a set of plug-in skills tailored toward static analysis, code auditing, secure defaults detection, and other practices that matter in software security. Users can easily add the marketplace to a Claude Code environment, browse available plugins, and install specific skills for tasks like automatic Semgrep rule creation, entry-point analysis in smart contracts, or insecure defaults detection. This project leverages the agent skills architecture to let AI assistants take on detailed, repeatable security procedures that are typically manual, such as parsing Burp Suite projects or conducting variant analysis across codebases.

Claude Skills is a public repository that showcases and serves as a collection of skills — modular, reusable packages of instructions, scripts, and resources that Claude and other compatible agents can dynamically discover and load to extend their capabilities on specialized tasks. Rather than relying on handcrafted prompts every time, Skills teach an AI agent procedural knowledge and task-specific workflows so it can apply that expertise reliably, whether the task involves document creation, data analysis, design generation, or technical automation. Each Skill lives in its own directory with a SKILL.md file containing metadata and instructions, and can include supplemental scripts or assets that the agent uses to perform complex operations when relevant.

OpenAI Agent Skills is an open-source repository that serves as a broad catalog of agent skills designed to extend the capabilities of OpenAI Codex and other AI coding agents. It organizes reusable, task-specific workflows, instructions, scripts, and resources into modular skill folders so that an AI agent can reliably perform complex tasks without repeated custom prompting, making agent behavior more predictable and composable. Each skill is defined with clear metadata and instructions organizing how an AI assistant should complete specific tasks ranging from project management to code generation and documentation assistance. The repository supports community contributions, allowing developers to add new skills or update existing ones to keep the catalog relevant and practical for evolving use cases.

The Softaworks Agent Toolkit is a comprehensive collection of agent skills, commands, and sub-agents designed to augment AI coding assistants like Claude Code, Codex, and Cursor with practical workflow capabilities. It packages broad categories of modular skills that help with development automation, documentation creation, planning, architecture, testing, and soft professional workflows. Beyond simple skills, it also includes agents and CLI slash commands that help developers automate common tasks such as pattern finding, diagram generation, requirement drafting, and daily standup preparation. The toolkit’s modular design follows the Agent Skills format, making it easy for users to install only what’s needed via CLI installers or plugin marketplaces. Because the set spans from low-level utilities like dependency updaters to higher-level planning and communication aids, it can streamline many aspects of a developer’s day-to-day work.

Agent Skills for Context Engineering is a curated collection of reusable “agent skills” focused on helping AI agents perform better on long-horizon, multi-step work by managing context deliberately. Rather than being a single application, it packages practical guidance into skill modules that agents can load to improve planning, retrieval, memory usage, and overall reliability in real workflows. The repository emphasizes context engineering as a discipline, covering why agents fail when context gets too large, too noisy, or poorly structured, and how to mitigate those failure modes with repeatable patterns. It is designed to be used across modern agent environments that support skill folders and structured instructions, so teams can standardize how agents operate instead of relying on ad-hoc prompting.

AWS Agent Skills is a repository that curates AWS-focused agent skills — capability modules that give AI assistants like Claude Code and Codex deep, practical knowledge across key Amazon Web Services domains. Instead of streaming giant documentation sets or relying on episodic web search, this project compresses AWS best practices, usage patterns, edge cases, and real-world engineering guides into pre-structured skill definitions that are token-efficient and tailored for reasoning. The skills cover critical AWS services such as IAM, Lambda, DynamoDB, S3, API Gateway, EKS, and many more, letting agents offer actionable advice on infrastructure as code, debugging, security configurations, and architectural workflows. Skills are kept up to date with weekly documentation checks, ensuring they reflect current AWS patterns and service changes.

This repository is a public security-focused scanning tool intended to analyze and assess AI agent skills for potential issues, quality concerns, and vulnerabilities. It acts as a scanner that inspects Agent Skills packages to flag structural problems, inconsistencies, or security flaws before they are deployed or integrated into agent workflows. Because agent skills can contain executable instructions and logic, scanning them for risky patterns is essential to prevent inadvertent exploitation when used by intelligent systems. The tool supports maintainers and community contributors in automating quality checks and enforcing conventions across skill sets in a standardized way. While still evolving with contributions and issue discussions, it shows the community’s interest in building safer AI ecosystems around reusable capabilities. The scanner also serves as a foundation for more sophisticated vetting frameworks that might be incorporated into CI/CD pipelines.

Hive is an open-source agent development framework that helps developers build autonomous, reliable, self-improving AI agents by letting them describe goals in ordinary natural language instead of hand-coding detailed workflows. Rather than manually defining execution graphs, Hive’s coding agent generates the agent graph, connection code, and test cases based on your high-level objectives, enabling outcome-driven agent creation that fits real business processes. Once deployed, agents can capture failure data, evolve automatically to meet their success criteria, and redeploy without constant manual intervention, delivering continual improvement over time. The framework also includes human-in-the-loop nodes, credential management, cost and budget controls, and real-time observability so teams can monitor execution and intervene as needed. Hive is designed for production environments and supports a wide range of large language models, local models, and business system connectivity.

agentskills is the specification and documentation repository for the Agent Skills open format, which defines a standardized way to package capabilities that AI agents can discover and use. A “skill” is treated as a foldered bundle containing instructions, optional scripts, and supporting resources, so agents can reliably apply a workflow or expertise area when it becomes relevant. The central goal is portability: you can write a skill once and reuse it across different agent runtimes and developer tools that implement the format. This repo serves as the canonical reference for how skills should be structured, what metadata they should include, and how an SDK can load and apply them consistently. It also includes supporting materials like guides and examples so builders can create skills that are predictable, testable, and shareable with teams.

Continuous Claude v3 is a persistent, multi-agent development environment built around the Claude Code CLI that aims to overcome the limitations of standard LLM context windows. Rather than relying on a single session’s context, Continuous Claude uses mechanisms like ledgers, YAML handoffs, and a memory system to preserve and recall state across multiple sessions, ensuring that learned insights and plans are not lost when context compaction occurs. The project orchestrates many specialized agents and skills—109 skills and 32 agents—so that complex coding tasks can be broken down, analyzed, and executed collaboratively by different components. It also includes a layered code analysis pipeline to reduce token usage and maintain relevant context efficiently. This continuous learning environment enables workflows such as bug fixing, refactoring, planning, and exploratory investigation while minimizing the need to re-explain context manually.

Mini-Agent is a minimal yet production-minded demo project that shows how to build a serious command-line AI agent around the MiniMax-M2 model. It is designed both as a reference implementation and as a usable agent, demonstrating a full execution loop that includes planning, tool calls, and iterative refinement. The project exposes an Anthropic-compatible API interface and fully supports interleaved thinking, letting the agent alternate between reasoning steps and tool invocations during long, complex tasks. It includes a basic toolset for file-system operations and shell commands, plus integrations with MCP tools such as web search and knowledge graph access. Mini-Agent also comes with “Claude Skills”-style predefined skills for tasks like document processing, design work, and testing, packaged as reusable behaviors that can be invoked by the agent as needed.