Agent Skills Expand Beyond Developer Tools: PM, Science, and Context Engineering Lead the Way

The Agent Skills ecosystem started as a developer tools phenomenon. Six months ago, every top skill repo focused on code generation, testing, and deployment. That is changing. Three new categories have entered the top 20, and they are growing faster than the dev-tools incumbents.

The Shift

Three repositories signal where the ecosystem is heading:

phuryn/pm-skills (9,700 stars) brings structured product management to AI agents. 65 skills cover discovery, strategy, execution, research, and go-to-market. Built on frameworks from Teresa Torres, Marty Cagan, and Alberto Savoia. The key insight: PMs spend more time on process than on product, and AI agents can run that process faster and more consistently than humans.

k-dense-ai/claude-scientific-skills (17,000 stars) covers 140 scientific tools across bioinformatics, drug discovery, clinical research, and data analysis. This is not a developer tool. It is a research platform. Scientists can describe a drug discovery pipeline in plain English and have an AI agent query ChEMBL, run molecular analysis with RDKit, dock candidates with DiffDock, and generate a structured report.

muratcankoylan/agent-skills-for-context-engineering (14,000 stars) addresses a meta-problem: managing what goes into an AI agent's context window. As context windows grow but model attention does not improve proportionally, context engineering becomes a discipline in itself. This collection teaches agents how to design, optimize, and evaluate their own context.

Why It Matters

The dev-tools category still dominates by volume. But the new categories solve different problems:

The pattern is consistent: skills succeed when they enforce discipline that humans know they should follow but often skip. Developers skip tests. PMs skip assumption validation. Scientists skip documentation. Context engineers skip compression. Skills make the right path the default.

The Engineering Skills Benchmark

Among dev-tools, addyosmani/agent-skills (9,500 stars) stands out for its structured approach. Rather than individual utilities, it covers the entire software development lifecycle in six phases: Define, Plan, Build, Verify, Review, Ship. Each skill enforces senior-engineer-level discipline, drawing from Google's engineering culture.

And sickn33/antigravity-awesome-skills (31,000 stars) takes an aggregator approach: 1,400+ skills installable via a single npm command, with role-based bundles and support for 10+ AI coding assistants.

What to Watch

Three trends to track:

1. Domain-specific skills will keep expanding. Legal, finance, education, and healthcare are underserved. The first high-quality skill collection in each domain will capture significant attention.

2. Skill quality will matter more than quantity. The ecosystem is still in the aggregation phase. As users accumulate skills, context window overload becomes a real problem. Skills that are precise, well-scoped, and progressive-disclosure-friendly will win over sprawling collections.

3. Cross-tool compatibility will become a differentiator. Skills that work across Claude Code, Cursor, Codex, and Gemini have a structural advantage over tool-specific collections.

What We Published This Week

• Production-Grade Agent Skills for the Full Development Lifecycle -- Addy Osmani's structured engineering skills
• Agent Skills for Scientific Research -- 140 scientific tools and databases
• Master Context Engineering -- Build better AI agent systems
• Browse and Install 1,400+ Agent Skills -- The antigravity skill library
• AI-Powered Product Management -- From discovery to launch

Get weekly ecosystem updates. Bookmark the news page for regular coverage.