The wshobson/agents repository is a multi-harness agentic plugin marketplace designed to serve as a unified source of truth for AI agent workflows across six different development environments: Claude Code, Codex CLI, Cursor, OpenCode, GitHub Copilot, and Gemini CLI. Written in Python, the marketplace contains 88 production-ready plugins, 194 domain-expert agents, 158 modular skills, 106 slash commands, and 16 multi-agent orchestrators that enable developers to build sophisticated agentic workflows without switching between tools.
The core innovation of this repository is its single-source-of-truth architecture. Rather than maintaining separate codebases for each harness, the marketplace uses a unified Markdown source in the plugins directory that generates harness-native artifacts tailored to each platform's specific requirements and capabilities. Claude Code receives native marketplace.json and plugin structures, Codex CLI and Cursor install from committed registries, while OpenCode and Gemini CLI install via clone and generation with gitignored transformed trees. GitHub Copilot receives Markdown agent profiles and native skill definitions. This approach ensures consistency across platforms while respecting each harness's idiomatic patterns and constraints, such as Codex CLI's 8 KB skill cap.
The agents within the marketplace are organized by domain expertise, covering architecture, programming languages, infrastructure, security, data engineering, machine learning, documentation, business logic, and SEO. A tiered model strategy assigns different Claude models to different task categories: Fable 5 for longest-horizon autonomous work, Opus for architecture and security-critical decisions, inherited user-chosen models for specialized domains, Sonnet for documentation and testing, and Haiku for fast operational tasks. This stratification optimizes both capability and cost across the marketplace's diverse use cases.
Quality assurance is built into the repository through the plugin-eval framework, a three-layer evaluation system that performs static structural analysis, LLM-based semantic evaluation across four dimensions, and Monte Carlo statistical reliability testing via 50 to 100 simulated runs. This ensures that plugins and skills meet production standards before deployment.
According to GitGenius tracking data, the repository has demonstrated steady growth with stargazers increasing from 37,518 to 37,519 as of July 4, 2026. The project maintains active issue and pull request management with a median response latency of 0.2 hours across 97 tracked items, though the mean response time is 42.2 hours, indicating that while critical issues receive immediate attention, comprehensive responses may take longer. Bug reports represent the most active issue category with 29 tracked items, followed by enhancement requests with 9 items. The primary maintainer wshobson has logged 185 events, with secondary contributors shintaii and t34wrj contributing 8 and 5 events respectively.
The repository overlaps with related projects including anthropics/claude-code, alirezarezvani/claude-skills, and crewaiinc/crewai through shared contributors, positioning it within a broader ecosystem of agentic AI development. External memory integration is provided through Pensyve, included as a git-subdir entry with direct upstream integrations maintained for all supported harnesses. The marketplace is licensed under MIT and includes comprehensive documentation covering plugin catalogs, agent categories, skill definitions, usage patterns, architecture principles, harness-specific capabilities, authoring guidelines, and quality evaluation methodologies.
